Processing method and device for predicting ASPECTS score based on CT images

Abstract

The embodiment of the present invention relates to a processing method and device for predicting ASPECTS scores based on CT images, the method comprising: constructing four segmentation models, a classification model and a scoring model; receiving two CT images to form an image sequence after training; performing image segmentation on the image sequence by the four segmentation models to obtain the first / second / third / fourth semantic map group and classifying the image sequence by the classification model to obtain a predicted side group; comprehensively evaluating the side information; extracting left and right brain partition images to obtain a partition image set; performing median filtering on the partition image set; eliminating sulci / old infarction areas on the partition image set based on the first / second semantic map group; extracting left and right brain partition image pairs; scoring the partition image pair sequence by the scoring model; and scoring the sequence as a whole / affected side based on the assessed side and partition score to obtain a summary report. The present invention can improve the prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a processing method and device for predicting ASPECTS scores based on CT images. Background Art

[0002] Ischemic stroke (CIS), also known as cerebral infarction (CI), is a type of vascular event in the central nervous system. There are two main symptoms of ischemic stroke: 1) High-density arterial sign in cerebral blood vessels: thrombus formation in arterial vessels slows down or even stagnates blood flow and causes ischemic damage to blood vessels. Thrombus in blood vessels forms high-density arterial sign on non-contrast CT (NCCT) images; 2) Cytotoxic edema of brain tissue: After ischemic damage occurs, serum proteins and water in the blood begin to penetrate into nearby brain tissue, causing cytotoxic edema of brain tissue. Cytotoxic edema forms local low-density image areas on plain CT images.

[0003] The Alberta Stroke Program Early CT Score (ASPECTS) is a systematic evaluation method for early ischemic changes (EIC) of ischemic stroke based on plain CT images. The ASPECTS scoring method divides 7 pairs of left and right brain regions (left / right brain C region, left / right brain L region, left / right brain IC region, left / right brain I region, left / right brain M1 region, left / right brain M2 region, left / right brain M3 region) on axial plain CT images at the level of the basal ganglia nuclei, and divides 3 pairs of left and right brain regions (left / right brain M4 region, left / right brain M5 region, left / right brain M6 region) on axial plain CT images above the basal ganglia nuclei; and scores (also known as ASPECTS scores) are performed based on these 10 pairs of left and right brain regions (also known as ASPECTS regions), and the score of each region is 0 or 1. A value of 0 indicates that the possibility of ischemic stroke in the current partition is relatively low, and a value of 1 indicates that the possibility of ischemic stroke in the current partition is relatively high; and the scores of 10 pairs of left and right brain partitions are used to locate the main affected side, and the so-called affected side refers to the left and right sides of the brain where ischemic stroke occurs; and based on the located affected side (such as the left side, right side, and bilateral sides), the scores of 10 pairs of left and right brain partitions are screened and totaled to obtain the corresponding total score; finally, the final ASPECTS overall score is obtained by subtracting the total score from the ASPECTS score (usually fixed at 10 points). The higher the score, the lower the possibility of ischemic stroke, and the lower the score, the higher the possibility of ischemic stroke.

[0004] At present, the processing mechanism for ASPECTS scoring of plain scan CT images based on the ASPECTS scoring rules mostly relies on manual scoring, that is, doctors use their own professional experience and ASPECTS scoring rules to perform ASPECTS scoring on the plain scan CT images of the subjects; this manual scoring method is affected by human factors, and it is difficult to provide high scoring efficiency and to ensure stable scoring quality.

[0005] To solve this problem, some researchers have proposed an intelligent scoring method based on image detection model + omics feature extraction tool + regression calculation model. This intelligent scoring method first uses the image detection model to perform ASPECTS partition detection on plain scan CT images to obtain 10 pairs of left and right brain partition images; then uses the omics feature extraction tool to identify the left / right brain partition image omics features of each pair of left and right brain partition images; then based on the regression calculation model, the omics features of each pair of left / right brain partition images are compared (usually using a differential method) to obtain the corresponding comparison features (or differential features), and regression calculation is performed based on the left / right brain partition image omics features and comparison features of each pair of left and right brain partitions to obtain the corresponding partition scores, and the main affected side is located based on the scores of the 10 pairs of left and right brain partitions, and the scores of the 10 pairs of left and right brain partitions are screened based on the located affected side to obtain the corresponding total score, and the final ASPECTS overall score is obtained by subtracting the total score of the 10 pairs of left and right brain partitions from the total ASPECTS score.

[0006] In practical applications, we found that although the above intelligent scoring method can improve scoring efficiency and output stable scoring quality, the prediction accuracy of this type of scoring method is not ideal. The main reasons are as follows: 1) The roughness of the partition detection boundary output by the image detection model causes the boundary accuracy of the partition image to be insufficient, which reduces the accuracy of image feature recognition, the accuracy of affected-side positioning, and the overall prediction accuracy; 2) There are many interference factors in the partition image. In addition to noise, common interference factors include omics feature interference caused by asymmetric sulcus structure and omics feature interference caused by old infarction areas of cerebral arteries. The existence of these interference factors will also reduce the accuracy of image feature recognition, the accuracy of affected-side positioning, and the overall prediction accuracy. Summary of the invention

[0007] The purpose of the present invention is to provide a processing method, device, electronic device and computer-readable storage medium for predicting ASPECTS scores based on CT images in response to the defects of the prior art. The present invention constructs four image segmentation models (a first segmentation model for segmenting sulcus regions, a second segmentation model for segmenting old infarction regions of cerebral arterial vessels, a third segmentation model for segmenting high-density arterial sign regions of cerebral vessels, and a fourth segmentation model for segmenting left and right brain ASPECTS partitions and ventricular regions), a first classification model for classifying brain lateral information of cytotoxic edema of brain tissue, and a first scoring model for ASPECTS scoring of ten groups of left and right brain partition CT images; and after all models are trained, a CT image sequence is composed of an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei of the brain and an axial plain scan CT image of the brain located above the level of the basal ganglia nuclei of the brain; and the first / second / third / fourth segmentation models are used to perform image segmentation on the CT image sequence to obtain corresponding The first / second / third / fourth semantic map group is used, and the first classification model is used to classify and identify the CT image sequence to obtain the corresponding predicted side group; and based on the third / fourth semantic map group, the predicted side group and the CT image sequence, the side information is comprehensively evaluated to obtain the corresponding assessed side; and the left and right brain ASPECTS partition images of the CT image sequence are extracted to obtain the corresponding partition image set; and each partition image in the partition image set is filtered based on the median filtering method, and after the filtering, the interference area (brain sulcus area and cerebral artery old infarction area) of each partition image is eliminated based on the first / second semantic map group; and the first scoring model is used to perform ASPECTS scoring on the partition image pair sequence to obtain the corresponding partition scoring pair sequence; finally, based on the assessed side and the partition scoring pair sequence, the overall score summary and the affected side summary are performed to obtain the corresponding summary report. It can be seen that the present invention uses the fourth segmentation model to improve the boundary accuracy of the partition image, uses the median filter to reduce the noise of the partition image, uses the first / second segmentation model to eliminate two types of interference factor images in the partition image (brain sulcus area images and cerebral arterial old infarction area images), uses the third segmentation model + the first classification model and a side information comprehensive evaluation mechanism to improve the positioning accuracy of the affected side, and uses the batch processing mechanism of the first scoring model to improve the partition scoring efficiency; through the present invention, not only can the overall prediction efficiency be further improved, but also the boundary accuracy of the partition image can be improved, the noise and interference of the partition image can be reduced, and the positioning accuracy of the affected side can be improved, so as to achieve the purpose of improving the prediction accuracy.

[0008] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present invention provides a processing method for predicting ASPECTS scores based on CT images, the method comprising:

[0009] Construct an image semantic segmentation model for performing sulcus region segmentation on brain CT images, which is recorded as the corresponding first segmentation model; construct an image semantic segmentation model for performing cerebral arterial vascular old infarction region segmentation on brain CT images, which is recorded as the corresponding second segmentation model; construct an image semantic segmentation model for performing cerebral vascular high-density arterial sign region segmentation on brain CT images, which is recorded as the corresponding third segmentation model; construct an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on brain CT image sequences, which is recorded as the corresponding fourth segmentation model; construct a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images; construct a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs; and collectively train the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on a preset plain scan CT image library;

[0010] After the collective model training is completed, an axial plain scan brain CT image located at the level of the basal ganglia nuclei and an axial plain scan brain CT image located at a level above the basal ganglia nuclei are received and recorded as corresponding first CT images and second CT images; and the first and second CT images form a corresponding first CT image sequence;

[0011] The first, second, third and fourth segmentation models respectively perform image segmentation processing on the first CT image sequence to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group, and the first classification model classifies and identifies the first CT image sequence to obtain the corresponding first predicted side group; the first predicted side group includes the first image predicted side and the second image predicted side; the first and second image predicted sides both include empty, left, right and bilateral;

[0012] Based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence, a comprehensive evaluation of the side information is performed to obtain a corresponding first evaluated side; the first evaluated side includes empty, left, right and bilateral;

[0013] The first CT image sequence is subjected to left and right brain ASPECTS partition image extraction processing to obtain a corresponding first partition image set; and each partition image in the first partition image set is filtered based on a median filtering method; and after filtering, each partition image in the first partition image set is subjected to brain sulcus area and cerebral arterial vessel old infarction area elimination processing based on the first and second semantic graph groups to obtain a corresponding second partition image set; and based on the second partition image set, ASPECTS partition left and right brain CT image pairs are extracted and sorted to obtain a corresponding first partition image pair sequence;

[0014] Performing ASPECTS scoring on the first partition image pair sequence by the first scoring model to obtain a corresponding first partition score pair sequence;

[0015] Based on the first assessment side and the first partition score, the sequence is summarized as a whole and on the affected side to obtain a corresponding first summary report; the first summary report includes a first total score and a first affected side type; the first total score is an integer between 0 and 10; the first affected side type includes left side, right side and bilateral side;

[0016] The first, second, third and fourth segmentation models are used to perform image segmentation processing on the first CT image sequence to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group, respectively, and the first classification model is used to classify and identify the first CT image sequence to obtain the corresponding first predicted side group, specifically including:

[0017] The first and second CT images of the first CT image sequence are respectively input into the first segmentation model to perform brain sulcus region segmentation processing to obtain the corresponding first image semantic map A 11 and the second image semantic map A 12 The first image semantic graph A is composed of the corresponding first semantic graph group; wherein the first image semantic graph A 11 Including a plurality of first semantic image pixels a 11 , each pixel point a in the first semantic map 11 Corresponding to a first semantic type; the second image semantic map A 12 Including a plurality of second semantic graph pixels a 12 , each pixel a of the second semantic map 12 corresponds to a first semantic type;

[0018] The first and second CT images of the first CT image sequence are respectively input into the second segmentation model to perform segmentation processing on the old infarction area of ​​the cerebral artery to obtain the corresponding first image semantic map A. 21 and the second image semantic map A 22The first image semantic graph A 21 Including a plurality of first semantic image pixels a 21 , each pixel point a in the first semantic map 21 Corresponding to a second semantic type; the second image semantic map A 22 Including a plurality of second semantic graph pixels a 22 , each pixel a of the second semantic map 22 corresponds to a second semantic type;

[0019] The first and second CT images of the first CT image sequence are respectively input into the third segmentation model to perform high-density arterial sign area segmentation processing on the cerebral blood vessels to obtain the corresponding first image semantic map A 31 and the second image semantic map A 32 The corresponding third semantic graph group is formed; wherein the first image semantic graph A 31 Including a plurality of first semantic image pixels a 31 , each pixel point a in the first semantic map 31 Corresponding to a third semantic type; the second image semantic map A 32 Including a plurality of second semantic graph pixels a 32 , each pixel a of the second semantic map 32 corresponds to a third semantic type;

[0020] The first CT image sequence is input into the fourth segmentation model to perform left and right brain ASPECTS partitioning and ventricular region segmentation processing to obtain a corresponding semantic map sequence; and the first and second semantic maps in the semantic map sequence obtained this time are used as the corresponding first image semantic map A in sequence. 41 and the second image semantic map A 42 Form a corresponding fourth semantic graph group; wherein the first image semantic graph A 41 Including a plurality of first semantic image pixels a 41 , each pixel point a in the first semantic map 41 Corresponding to a fourth first semantic type; the second image semantic map A 42 Including a plurality of second semantic graph pixels a 42 , each pixel a of the second semantic map 42 corresponds to a 42nd semantic type;

[0021] The first and second CT images of the first CT image sequence are respectively input into the first classification model, and the first classification model classifies and identifies the brain side information of brain tissue cytotoxic edema on the first and second CT images to obtain the corresponding first image predicted side and the second image predicted side to form the corresponding first predicted side group.

[0022] Preferably, the cerebral vascular high-density arterial sign and the brain tissue cytotoxic edema are each a typical sign of ischemic stroke;

[0023] The ASPECTS partition is composed of ten types of brain tissue partitions, namely C partition, L partition, IC partition, I partition, M1 partition, M2 partition, M3 partition, M4 partition, M5 partition and M6 partition; the C partition is the caudate nucleus area of ​​the brain; the L partition is the lenticular nucleus area of ​​the brain; the IC partition is the internal capsule area of ​​the brain; the I partition is the insular cortex area of ​​the brain; the M1 partition is the anterior cortical area of ​​the middle cerebral artery; the M2 partition is the lateral cortical area of ​​the insular lobe of the middle cerebral artery; the M3 partition is the posterior cortical area of ​​the middle cerebral artery; the M4 partition is the middle cerebral artery cortical area above the M1 partition; the M5 partition is the middle cerebral artery cortical area above the M2 partition; the M6 ​​partition is the middle cerebral artery cortical area above the M3 partition; the various types of brain tissue partitions of the ASPECTS partition are symmetrical between the left and right brains.

[0024] Preferably, the first segmentation model is implemented based on a U-Net image segmentation network; the first segmentation model is used to perform sulcus region segmentation processing on the current CT image input to the model to obtain a corresponding first semantic map; the current CT image is an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei of the brain or an axial plain scan CT image of the brain located at a level above the basal ganglia nuclei of the brain; the first semantic map includes a plurality of first pixel points; each of the first pixel points corresponds to a first semantic type; the first semantic type includes sulcus region semantics and background semantics;

[0025] The second segmentation model is implemented based on the U-Net image segmentation network; the second segmentation model is used to perform brain artery old infarction area segmentation processing on the current CT image input by the model to obtain the corresponding second semantic map; the current CT image is an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain or an axial brain plain scan CT image located at a level above the basal ganglia nuclei of the brain; the second semantic map includes a plurality of second pixel points; each of the second pixel points corresponds to a second semantic type; the second semantic type includes old infarction area semantics and background semantics;

[0026] The third segmentation model is implemented based on the U-Net image segmentation network; the third segmentation model is used to perform brain vascular high-density arterial sign area segmentation processing on the current CT image input by the model to obtain a corresponding third semantic map; the current CT image is an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain or an axial brain plain scan CT image located at a level above the basal ganglia nuclei of the brain; the third semantic map includes a plurality of third pixel points; each of the third pixel points corresponds to a third semantic type; the third semantic type includes high-density arterial sign area semantics and background semantics;

[0027] The fourth segmentation model is implemented based on two U-Net image segmentation networks, which are recorded as the corresponding first U-Net network and the second U-Net network; the fourth segmentation model is used to perform left and right brain ASPECTS partitioning and ventricular region segmentation processing on the current CT image sequence input by the model to obtain the corresponding fourth semantic map sequence, specifically: the fourth segmentation model is used to use the first CT image in the current CT image sequence input by the model as the corresponding first network input image, and the second CT image in the current CT image sequence as the corresponding second network input image; and the first U-Net network performs fourteen categories of left and right brain ASPECTS partitioning and ventricular region segmentation processing on the first network input image to obtain the corresponding fourth-first semantic map; and the second U-Net network performs six categories of left and right brain ASPECTS partitioning and ventricular region segmentation processing on the second network input image to obtain the corresponding fourth-second semantic map; and the obtained fourth-first and fourth-second semantic maps are sequentially sorted to form the corresponding fourth semantic map sequence output;

[0028] The current CT image sequence is formed by sorting two CT images, wherein the first CT image is an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei of the brain, and the second CT image is an axial plain scan CT image of the brain located at a level above the basal ganglia nuclei of the brain;

[0029] The fourth-first semantic map includes a plurality of fourth-first pixels; each of the fourth-first pixels corresponds to a fourth-first semantic type; the fourth-first semantic type includes fourteen categories of left and right brain ASPECTS partition semantics, ventricular region semantics and background semantics; the fourteen categories of left and right brain ASPECTS partition semantics include left / right brain C partition semantics, left / right brain L partition semantics, left / right brain IC partition semantics, left / right brain I partition semantics, left / right brain M1 partition semantics, left / right brain M2 partition semantics, and left / right brain M3 partition semantics;

[0030] The fourth-second semantic map includes a plurality of fourth-second pixels; each of the fourth-second pixels corresponds to a fourth-second semantic type; the fourth-second semantic type includes six categories of left and right brain ASPECTS partition semantics, ventricular region semantics and background semantics; the six categories of left and right brain ASPECTS partition semantics include left / right brain M4 region semantics, left / right brain M5 region semantics, and left / right brain M6 region semantics;

[0031] The first classification model includes a first feature extraction module and a first classification module; the first feature extraction module is implemented based on a ResNet network; the first classification module is implemented based on an MLP network; the first classification model is used to classify and identify the brain side information that produces brain tissue cytotoxic edema on the current CT image input by the model to obtain the corresponding predicted side, specifically: the first classification model is used to input the current CT image input by the model into the first feature extraction module for feature extraction processing to obtain the corresponding feature tensor; and input the feature tensor into the first classification module for classification prediction processing to obtain the corresponding prediction type; the prediction types include empty, left, right and bilateral;

[0032] The first scoring model is composed of ten parallel partition scoring models and one partition scoring merging module; each of the partition scoring models is composed of a left and right brain image sorting module, a left brain feature extraction module, a right brain feature extraction module, a left and right brain feature difference module, a left brain feature fusion module, a right brain feature fusion module, a left brain score prediction module, a right brain score prediction module and a left and right brain score merging module; the left and right brain score prediction modules are both implemented based on a binary classification prediction model;

[0033] The first scoring model is used to perform ASPECTS scoring on the current partition image pair sequence input by the model to obtain a corresponding partition score pair sequence, specifically: the first scoring model is used to input each partition image pair in the current partition image pair sequence input by the model into the corresponding partition scoring model, and each partition scoring model performs left and right brain ASPECTS scoring according to the corresponding partition image pair to obtain a corresponding partition score pair and send it to the partition scoring merging module; and the partition scoring merging module sequentially sorts all the obtained partition score pairs to form the corresponding partition score pair sequence;

[0034] Among them, each of the partition scoring models performs left and right brain ASPECTS scoring according to the corresponding partition image pair to obtain the corresponding partition scoring pair and send it to the partition scoring merging module, specifically: each of the partition scoring models inputs the corresponding partition image pair into the left and right brain image sorting module, and the left and right brain image sorting module extracts the corresponding left brain partition image and right brain partition image from the partition image pair and sends them to the corresponding left brain feature extraction module and the right brain feature extraction module; and the left brain feature extraction module performs imaging genomics feature extraction on the left brain partition image The corresponding left brain imaging omics features are extracted and sent to the left and right brain feature difference module and the left brain feature fusion module; and the right brain feature extraction module extracts the imaging omics features of the right brain partition image to obtain the corresponding right brain imaging omics features and sends them to the left and right brain feature difference module and the right brain feature fusion module; and the left and right brain feature difference module records the differential features of the left brain imaging omics features minus the right brain imaging omics features and the differential features of the right brain imaging omics features minus the left brain imaging omics features as the corresponding left-right brain imaging omics differential features and right-left brain images. The left-right brain imaging omics differential features are sent to the corresponding left brain feature fusion module, and the right-left brain imaging omics differential features are sent to the right brain feature fusion module; the left brain feature fusion module performs feature splicing processing on the left brain imaging omics features and the left-right brain imaging omics differential features to obtain the corresponding left brain splicing features and sends them to the left brain score prediction module; the right brain feature fusion module performs feature splicing processing on the right brain imaging omics features and the right-left brain imaging omics differential features to obtain the corresponding right brain splicing features and sends them to the left brain score prediction module. The right brain score prediction module sends the result; the left brain score prediction module performs binarized ASPECTS score prediction processing according to the left brain splicing features to obtain the corresponding left brain partition score and sends it to the left and right brain score merging module; the right brain score prediction module performs binarized ASPECTS score prediction processing according to the right brain splicing features to obtain the corresponding right brain partition score and sends it to the left and right brain score merging module; the left and right brain score merging module combines the left brain partition score and the right brain partition score into the corresponding partition score pair and sends it to the partition score merging module;

[0035] The partition image pair sequence is formed by sequentially sorting the ten current partition image pairs; each partition image pair corresponds to a type of partition in the ASPECTS partition; the partition image pair corresponds one-to-one to the partition scoring model in the first scoring model; each partition image is composed of the corresponding left brain partition image and the right brain partition image;

[0036] The partition score pair sequence is composed of ten partition score pairs arranged in sequence; the partition score pairs correspond one-to-one to the partition image pairs; each partition score pair is composed of the corresponding left brain partition value and the right brain partition value; the left and right brain partition values ​​are 0 or 1.

[0037] Preferably, the plain scan CT image library includes a plurality of first plain scan CT image groups; each of the first plain scan CT image groups corresponds to two plain scan CT acquisition images of an image acquisition object; the first plain scan CT image group includes a first acquisition object type, a first CT acquisition image, and a second CT acquisition image; the first acquisition object type is used to mark the characteristic type of the corresponding image acquisition object, including at least a healthy type without brain disease, a type with a history of old infarction of cerebral arterial vessels but no symptoms of ischemic stroke, a type of class A ischemic stroke with no history of old infarction of cerebral arterial vessels but with symptoms of ischemic stroke, and a type of class B ischemic stroke with a history of old infarction of cerebral arterial vessels and symptoms of ischemic stroke; the symptoms of ischemic stroke include the high-density arterial signs of cerebral vessels and the One or all of brain tissue cytotoxic edema; the first CT acquisition image is an axial brain plain scan CT image of the corresponding image acquisition object located at the level of the basal ganglia nuclei of the brain; the second CT acquisition image is an axial brain plain scan CT image of the corresponding image acquisition object located at a level above the basal ganglia nuclei of the brain; the acquisition method of the first and second CT acquisition images is: an axial brain CT plain scan is performed on the image acquisition object to obtain a corresponding plain scan CT image sequence of the time, and a CT image located at the level of the basal ganglia nuclei of the brain is extracted from the plain scan CT image sequence of the time as the corresponding first CT acquisition image, and a CT image located at a level above the basal ganglia nuclei of the brain is extracted from the plain scan CT image sequence of the time as the corresponding second CT acquisition image;

[0038] The first partition image set includes twenty first partition images, and the first to twentieth first partition images are the corresponding left / right brain C area image, left / right brain L area image, left / right brain IC area image, left / right brain I area image, left / right brain M1 area image, left / right brain M2 area image, left / right brain M3 area image, left / right brain M4 area image, left / right brain M5 area image and left / right brain M6 area image respectively;

[0039] The second subarea image set includes twenty second subarea images, and the first to the twentieth second subarea images are the corresponding left / right brain C area image, left / right brain L area image, left / right brain IC area image, left / right brain I area image, left / right brain M1 area image, left / right brain M2 area image, left / right brain M3 area image, left / right brain M4 area image, left / right brain M5 area image and left / right brain M6 area image respectively;

[0040] The first partition image pair sequence is formed by sequentially sorting ten first partition image pairs; the first to tenth first partition image pairs are respectively corresponding left and right brain C area image pairs, left and right brain L area image pairs, left and right brain IC area image pairs, left and right brain I area image pairs, left and right brain M1 area image pairs, left and right brain M2 area image pairs, left and right brain M3 area image pairs, left and right brain M4 area image pairs, left and right brain M5 area image pairs and left and right brain M6 area image pairs;

[0041] The first partition score pair sequence is formed by sequentially sorting ten first partition score pairs; the first partition score pairs correspond to the first partition image pairs one by one; the first first partition score pair is composed of corresponding left and right brain C distinction values; the second first partition score pair is composed of corresponding left and right brain L distinction values; the third first partition score pair is composed of corresponding left and right brain IC distinction values; the fourth first partition score pair is composed of corresponding left and right brain I distinction values; the fifth first partition score pair is composed of corresponding left and right brain M1 distinction values; the sixth first partition score pair is composed of corresponding left and right brain M2 distinction values; the seventh The first partition score pair consists of corresponding left and right brain M3 distinguishing values; the eighth first partition score pair consists of corresponding left and right brain M4 distinguishing values; the ninth first partition score pair consists of corresponding left and right brain M5 distinguishing values; the tenth first partition score pair consists of corresponding left and right brain M6 distinguishing values; all left / right brain partition score values ​​in the first partition score pair sequence are 0 or 1, and when any left / right brain partition score is 1, it indicates that the possibility of ischemic stroke occurring in the corresponding left / right brain partition is relatively high, and when any left / right brain partition score is 0, it indicates that the possibility of ischemic stroke occurring in the corresponding left / right brain partition is relatively low.

[0042] Preferably, the performing a comprehensive evaluation of the side information based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence to obtain the corresponding first evaluated side specifically includes:

[0043] Step 501: convert the first image semantic graph A of the third semantic graph group into 31 In the above, each pixel point a in the first semantic map corresponding to the semantics of the high-density arterial sign region 31as corresponding first-category points; and clustering all the first-category points based on a preset point clustering algorithm to obtain a plurality of first-category point sets; and marking corresponding high-density arterial sign areas on the first CT image of the first CT image sequence based on each of the first-category point sets to obtain corresponding first marked areas;

[0044] Wherein, the point clustering algorithm includes at least K-means clustering algorithm, DBSCAN clustering algorithm, and OPTICS clustering algorithm;

[0045] Step 502: convert the second image semantic graph A of the third semantic graph group into 32 In the above, each pixel point a of the second semantic map corresponding to the semantics of the high-density arterial sign region 32 as corresponding second-category points; and clustering all the second-category points based on the point clustering algorithm to obtain a plurality of second-category point sets; and marking corresponding high-density arterial sign areas on the second CT image of the first CT image sequence based on each of the second-category point sets to obtain corresponding second marked areas;

[0046] Step 503: convert the first image semantic graph A of the fourth semantic graph group into 41 In the above, all the first semantic map pixels a corresponding to each type of partition semantics of the fourteen types of left and right brain ASPECTS partition semantics 41 The fourteen third-class point sets are clustered into one class to form a corresponding third-class point set; and based on the obtained fourteen third-class point sets, corresponding left and right brain ASPECTS partition labels are performed on the first CT image to obtain corresponding fourteen first-labeled partitions; and seven left brain partitions in the fourteen first-labeled partitions form a corresponding first left brain partition set, and seven right brain partitions form a corresponding first right brain partition set;

[0047] Step 504: convert the second image semantic graph A of the fourth semantic graph group into 42 In the above, all the pixel points a of the second semantic map corresponding to each type of partition semantics of the six types of left and right brain ASPECTS partition semantics 42 The six fourth-class point sets are clustered into one class to form a corresponding fourth-class point set; and based on the obtained six fourth-class point sets, corresponding left and right brain ASPECTS partitions are marked on the second CT image to obtain corresponding six second-marked partitions; and three left brain partitions in the six second-marked partitions form a corresponding second left brain partition set, and three right brain partitions form a corresponding second right brain partition set; and the first and second left brain partition sets are merged to obtain a corresponding third left brain partition set, and the first and second right brain partition sets are merged to obtain a corresponding third right brain partition set;

[0048] Step 505, setting a corresponding first side-specific evaluation result based on the regional intersection relationship between the third left brain partition set and the third right brain partition set and all the first and second marked regions; and taking the third average precision rate of the third segmentation model obtained in the most recent model training as the first confidence level corresponding to the first side-specific evaluation result;

[0049] Wherein, the first side assessment results include empty, left side, right side and bilateral;

[0050] Step 506, and identify the first image prediction side and the second image prediction side of the first prediction side group; if both the first and second image prediction sides are empty, set the corresponding second side evaluation result to empty; if one of the first and second image prediction sides is the left side and the other is the left side or empty, set the corresponding second side evaluation result to the left side; if one of the first and second image prediction sides is the right side and the other is the right side or empty, set the corresponding second side evaluation result to the right side; if one of the first and second image prediction sides is the left side and the other is the right side, set the corresponding second side evaluation result to the bilateral side; and use the fifth precision rate obtained by the first classification model in the most recent model training as the second confidence corresponding to the second side evaluation result;

[0051] Wherein, the second side assessment results include empty, left side, right side and bilateral;

[0052] Step 507, and identify the switch state of the preset manual side evaluation switch; if the switch state of the manual side evaluation switch is in the open state, set the corresponding third side evaluation result to be empty, and set the third confidence level corresponding to the third side evaluation result to 0; if the switch state of the manual side evaluation switch is in the closed state, send the first CT image sequence to the preset first manual side evaluation interface, and receive the third side evaluation result and the corresponding third confidence level sent back by the first manual side evaluation interface;

[0053] Wherein, the switch state of the artificial side evaluation switch includes an open state and a closed state;

[0054] Step 508, identifying whether the first, second and third side-specific evaluation results are all the same; if they are all the same, taking any one of the first, second and third side-specific evaluation results as the corresponding current evaluation side and going to step 516; if they are not all the same, going to step 509;

[0055] Step 509, identifying whether the third confidence is 0; if the third confidence is 0, go to step 510; if the third confidence is not 0, go to step 511;

[0056] Step 510, identifying whether one of the first and second side-specific evaluation results is empty; if so, taking the other side-specific evaluation result that is not empty as the corresponding current evaluation side and going to step 516; if not, taking the side-specific evaluation result with the highest confidence between the first and second side-specific evaluation results as the corresponding current evaluation side and going to step 516;

[0057] Step 511, identifying a preset first evaluation mode; if the first evaluation mode is the first mode, go to step 512; if the first evaluation mode is the second mode, go to step 513; if the first evaluation mode is the third mode, go to step 514; if the first evaluation mode is the fourth mode, go to step 515;

[0058] Wherein, the first evaluation mode includes a first mode, a second mode, a third mode and a fourth mode;

[0059] Step 512, taking the side evaluation result with the highest confidence among the first, second and third side evaluation results as the corresponding current evaluation side and going to step 516;

[0060] Step 513, set four sets of ballot boxes initialized to be empty for the four side types of empty, left, right and both sides, and record them as the corresponding first, second, third and fourth ballot box sets; and when the first, second or third side evaluation result matches one of the four side types, add the current side evaluation result to the first, second, third or fourth ballot box set corresponding to the current side type; and use the total number of the added side evaluation results in the first, second, third and fourth ballot box sets finally obtained as the corresponding number of votes in the first, second, third and fourth ballot boxes; and add the first, second, third and fourth ballot box sets to the first, second, third and fourth ballot box sets. The maximum value among the votes in the four ballot boxes is taken as the corresponding maximum votes; and whether the number of ballot box sets corresponding to the maximum votes is unique is identified; if the number of ballot box sets corresponding to the maximum votes is unique, the side type corresponding to the maximum votes among the four side types of empty, left, right and double sides is taken as the corresponding current evaluation side and the process goes to step 516; if the number of ballot box sets corresponding to the maximum votes is not unique, the side evaluation result with the highest confidence among all the side evaluation results in all the ballot box sets corresponding to the maximum votes is taken as the corresponding current evaluation side and the process goes to step 516;

[0061] Step 514, taking the difference of the first, second or third confidence minus a preset baseline confidence threshold as the corresponding first, second or third differential confidence; and setting four differential confidence sets initialized to empty for the four side identification types of empty, left, right and bilateral as the corresponding first, second, third and fourth differential confidence sets; and when the first, second or third side identification evaluation result matches one of the four side identification types, adding the first, second or third differential confidence corresponding to the current side identification evaluation result to the first, second, third or fourth differential confidence set corresponding to the current side identification type; and calculating the sum of all differential confidences in the first, second, third or fourth differential confidence set finally obtained Obtain the corresponding sum of the first, second, third or fourth sets; and take the maximum value among the sums of the first, second, third and fourth sets as the corresponding maximum sum; and identify whether the number of differential confidence sets corresponding to the maximum sum is unique; if the number of differential confidence sets corresponding to the maximum sum is unique, then take the side type corresponding to the maximum sum among the four side types of empty, left, right and bilateral as the corresponding current evaluation side and go to step 516; if the number of differential confidence sets corresponding to the maximum sum is not unique, then take the side evaluation result with the largest confidence among all side evaluation results in all differential confidence sets corresponding to the maximum sum as the corresponding current evaluation side and go to step 516;

[0062] Step 515, taking the difference of the first, second or third confidence minus one of the baseline confidence thresholds as the corresponding fourth, fifth or sixth differential confidence; and setting four differential confidence sets initialized to empty for the four side identification types of empty, left, right and bilateral as the corresponding fifth, sixth, seventh and eighth differential confidence sets; and when the first, second or third side identification evaluation result matches one of the four side identification types, adding the fourth, fifth or sixth differential confidence corresponding to the current side identification result to the fifth, sixth, seventh or eighth differential confidence set corresponding to the current side identification type; and calculating the average of all differential confidences in the fifth, sixth, seventh or eighth differential confidence set finally obtained to the corresponding first, second, third or fourth set mean; and the maximum value among the means of the first, second, third and fourth sets is taken as the corresponding maximum mean; and whether the number of differential confidence sets corresponding to the maximum mean is unique is identified; if the number of differential confidence sets corresponding to the maximum mean is unique, then the side type corresponding to the maximum mean among the four side types of empty, left, right and bilateral is taken as the corresponding current evaluation side and the process goes to step 516; if the number of differential confidence sets corresponding to the maximum mean is not unique, then the side evaluation result with the largest confidence among all side evaluation results in all differential confidence sets corresponding to the maximum mean is taken as the corresponding current evaluation side and the process goes to step 516;

[0063] Step 516: output the current evaluation aspect finally obtained as the corresponding first evaluation aspect.

[0064] Furthermore, the first sideness evaluation result corresponding to the area intersection relationship between the third left brain partition set and the third right brain partition set and all the first and second marked areas is set specifically including:

[0065] If all left brain partitions in the third left brain partition set have no intersection with all the first and second marked areas, and all right brain partitions in the third right brain partition set have no intersection with all the first and second marked areas, then set the corresponding first sideness evaluation result to empty;

[0066] If at least one left brain partition in the third set of left brain partitions intersects with at least one of the first or second marked regions, and at least one right brain partition in the third set of right brain partitions intersects with at least one of the first or second marked regions, then setting the corresponding first sideness assessment result to bilateral;

[0067] If at least one left brain partition in the third set of left brain partitions has an intersection with at least one of the first or second marked regions, and all right brain partitions in the third set of right brain partitions have no intersection with all of the first and second marked regions, then the corresponding first side evaluation result is set to the left side;

[0068] If at least one right brain partition in the third right brain partition set has an intersection with at least one of the first or second marked areas, and all left brain partitions in the third left brain partition set have no intersection with all of the first and second marked areas, then the corresponding first side assessment result is set to the right side.

[0069] Preferably, the extracting of left and right brain ASPECTS subregion images of the first CT image sequence to obtain the corresponding first subregion image set specifically includes:

[0070] Extracting the left brain C region image, the left brain L region image, the left brain IC region image, the left brain I region image, the left brain M1 region image, the left brain M2 region image, the left brain M3 region image, the left brain M4 region image, the left brain M5 region image and the left brain M6 region image corresponding to the ten left brain regions of the third left brain region set on the first and second CT images of the first CT image sequence as the corresponding ten first region images;

[0071] and extracting right brain C region images, right brain L region images, right brain IC region images, right brain I region images, right brain M1 region images, right brain M2 region images, right brain M3 region images, right brain M4 region images, right brain M5 region images, and right brain M6 region images corresponding to the ten right brain partitions of the third right brain partition set on the first and second CT images of the first CT image sequence as the corresponding ten first partition images;

[0072] The obtained twenty first subarea images form a corresponding first subarea image set.

[0073] Preferably, after filtering is completed, based on the first and second semantic graph groups, each sub-area image in the first sub-area image set is subjected to a sulcus region and a cerebral artery old infarction region elimination process to obtain a corresponding second sub-area image set, specifically comprising:

[0074] The first image semantic graph A of the first semantic graph group 11 In the above, each pixel point a in the first semantic map corresponding to the semantics of the sulcus region 11The points of the fifth category are recorded as corresponding fifth category points; and based on a preset point clustering algorithm, all the fifth category points are clustered to obtain multiple fifth category point sets; and based on each of the fifth category point sets, corresponding sulcus area markings are performed on the first CT image to obtain corresponding third marking areas; the point clustering algorithm at least includes a K-means clustering algorithm, a DBSCAN clustering algorithm, and an OPTICS clustering algorithm;

[0075] and transform the second image semantic graph A of the first semantic graph group into 12 In the above, each pixel point a in the second semantic map corresponding to the semantics of the sulcus region 12 Recorded as the corresponding sixth-category points; and clustering all the sixth-category points based on the point clustering algorithm to obtain multiple sixth-category point sets; and marking the corresponding sulcus area on the second CT image based on each of the sixth-category point sets to obtain the corresponding fourth marked area;

[0076] and transform the first image semantic graph A of the second semantic graph group into 21 In the above, each pixel point a in the first semantic map corresponding to the semantics of the old infarction area 21 Recorded as the corresponding seventh type of points; and clustering all the seventh type of points based on the point clustering algorithm to obtain multiple seventh type of point sets; and marking the corresponding old infarction area on the first CT image based on each of the seventh type of point sets to obtain the corresponding fifth marked area;

[0077] The second image semantic graph A of the second semantic graph group 22 In the above, each pixel point a in the second semantic map corresponding to the semantics of the old infarction area 22 Recorded as the corresponding eighth-category points; and clustering all the eighth-category points based on the point clustering algorithm to obtain multiple eighth-category point sets; and marking the corresponding old infarction area on the second CT image based on each of the eighth-category point sets to obtain the corresponding sixth marking area;

[0078] And take each of the first sub-area images in the first sub-area image set as the corresponding current sub-area image in turn; and take the intersection area of ​​the current sub-area image with any of the third, fourth, fifth or sixth marked areas as the corresponding first intersection area; and count the total number of the first intersection areas obtained to obtain the corresponding first intersection total number; and identify the first intersection total number; if the first intersection total number is 0, take the current sub-area image as the corresponding second sub-area image; if the first intersection total number is greater than 0, delete the local area images covered by each of the first intersection areas in the current sub-area image and take the deleted current sub-area image as the corresponding second sub-area image;

[0079] All the obtained second subarea images form a corresponding second subarea image set.

[0080] Preferably, the overall score aggregation and affected side aggregation of the sequence based on the first assessment side and the first partition score to obtain a corresponding first summary report specifically includes:

[0081] Clustering all left brain partition values ​​with a score of 1 in the first partition score pair sequence into one class to form a corresponding first left brain partition value set; and clustering all right brain partition values ​​with a score of 1 in the first partition score pair sequence into one class to form a corresponding first right brain partition value set;

[0082] Identifying the first assessment aspect;

[0083] If the first assessment aspect is empty, the sum of the scores of the first left brain partition value set is calculated to obtain the corresponding first total score, and the sum of the scores of the first right brain partition value set is calculated to obtain the corresponding second total score; and the first and second total scores are compared, and if the first total score is greater than or equal to the second total score, the first total score is used as the corresponding current total score N c , and set the corresponding first affected side type to the left side, if the first total score is less than the second total score, then use the second total score as the corresponding current total score N c , and setting the corresponding first affected side type to the right side;

[0084] If the first assessment side is the left side, the sum of the scores of the first left brain partition score set is calculated and the calculation result is used as the corresponding current total score N c , and setting the corresponding first affected side type to the left side;

[0085] If the first assessment side is the right side, the sum of the scores of the first right brain partition score set is calculated and the calculation result is used as the corresponding current total score N c , and setting the corresponding first affected side type to the right side;

[0086] If the first assessment side is bilateral, the sum of the scores of the first left brain partition value set is calculated to obtain a corresponding third total score, and the sum of the scores of the first right brain partition value set is calculated to obtain a corresponding fourth total score; and the third and fourth total scores are compared, and if the third score is greater than or equal to the fourth total score, the third total score is used as the corresponding current total score N c If the third total score is less than the fourth total score, the second total score is used as the corresponding current total score N c ; and setting the corresponding first affected side type to bilateral;

[0087] Based on the preset ASPECTS score N max and the current total score N c Calculate and obtain the corresponding first total score, first total score = N max -N c ; The total score of ASPECTS is N max The default value is 10; the current total score N c and the first total score are both integers ranging from 0 to 10;

[0088] The first total score and the first affected side type obtained form the corresponding first summary report.

[0089] The second aspect of the embodiment of the present invention provides a device for implementing the processing method for predicting ASPECTS scores based on CT images described in the first aspect, the device comprising: a model building and training module, a data receiving module, an image segmentation and classification module, a side comprehensive evaluation module, a partition image denoising and de-interference module, a partition scoring module and an overall summary module;

[0090] The model construction and training module is used to construct an image semantic segmentation model for performing sulcus region segmentation on brain CT images, which is recorded as the corresponding first segmentation model; and to construct an image semantic segmentation model for performing cerebral arterial vascular old infarction region segmentation on brain CT images, which is recorded as the corresponding second segmentation model; and to construct an image semantic segmentation model for performing cerebral vascular high-density arterial sign region segmentation on brain CT images, which is recorded as the corresponding third segmentation model; and to construct an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on brain CT image sequences, which is recorded as the corresponding fourth segmentation model; and to construct a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images; and to construct a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs; and to collectively train the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on a preset plain scan CT image library;

[0091] The data receiving module is used to receive an axial plain scan brain CT image located at the level of the basal ganglia nuclei and an axial plain scan brain CT image located at a level above the basal ganglia nuclei as corresponding first CT images and second CT images after the collective model training is completed; and the first and second CT images form a corresponding first CT image sequence;

[0092] The image segmentation and classification module is used to perform image segmentation processing on the first CT image sequence by the first, second, third and fourth segmentation models respectively to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group, and to classify and identify the first CT image sequence by the first classification model to obtain the corresponding first predicted side group; the first predicted side group includes the first image predicted side and the second image predicted side; the first and second image predicted sides both include empty, left, right and bilateral;

[0093] The side comprehensive evaluation module is used to perform a comprehensive evaluation of side information based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence to obtain a corresponding first evaluation side; the first evaluation side includes empty, left, right and bilateral;

[0094] The subarea image denoising and de-interference module is used to extract left and right brain ASPECTS subarea images from the first CT image sequence to obtain a corresponding first subarea image set; and filter each subarea image in the first subarea image set based on a median filter; and after filtering, eliminate the sulcus region and the old infarction region of the cerebral artery vessels from each subarea image in the first subarea image set based on the first and second semantic graph groups to obtain a corresponding second subarea image set; and extract and sort ASPECTS subarea left and right brain CT image pairs based on the second subarea image set to obtain a corresponding first subarea image pair sequence;

[0095] The partition scoring module is used to perform ASPECTS scoring on the first partition image pair sequence using the first scoring model to obtain a corresponding first partition score pair sequence;

[0096] The overall summary module is used to summarize the overall score and the affected side of the sequence based on the first assessment side and the first partition score to obtain a corresponding first summary report; the first summary report includes a first total score and a first affected side type; the first total score is an integer between 0 and 10; the first affected side type includes left side, right side and bilateral side.

[0097] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0098] The processor is used to be coupled to the memory, read and execute instructions in the memory, so as to implement the method steps described in the first aspect above;

[0099] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0100] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed by a computer, the computer executes the instructions of the method described in the first aspect above.

[0101] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for predicting ASPECTS score based on CT images. As can be seen from the above content, the embodiment of the present invention constructs four image segmentation models (a first segmentation model for brain sulcus area segmentation, a second segmentation model for brain arterial old infarction area segmentation, a third segmentation model for brain vascular high-density arterial sign area segmentation, and a fourth segmentation model for left and right brain ASPECTS partition and ventricular area segmentation), a first classification model for classifying brain lateral information of brain tissue cytotoxic edema, and a first scoring model for ASPECTS scoring of ten groups of left and right brain partition CT images; and after all models are trained, a CT image sequence is composed of an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain and an axial brain plain scan CT image located above the level of the basal ganglia nuclei of the brain; and the first / second / third / fourth segmentation models are used to perform image classification on the CT image sequence respectively. The corresponding first / second / third / fourth semantic map groups are obtained by segmentation, and the first classification model is used to classify and identify the CT image sequence to obtain the corresponding predicted side group; and the side information is comprehensively evaluated based on the third / fourth semantic map groups, the predicted side group and the CT image sequence to obtain the corresponding assessed side; and the left and right brain ASPECTS partition images of the CT image sequence are extracted to obtain the corresponding partition image set; and each partition image in the partition image set is filtered based on the median filtering method, and after the filtering, the interference area (brain sulcus area and cerebral artery old infarction area) of each partition image is eliminated based on the first / second semantic map group; and the first scoring model is used to perform ASPECTS scoring on the partition image pair sequence to obtain the corresponding partition scoring pair sequence; finally, based on the assessed side and the partition scoring pair sequence, the overall score and the affected side summary are summarized to obtain the corresponding summary report. It can be seen that the embodiment of the present invention uses the fourth segmentation model to improve the boundary accuracy of the partition image, uses the median filter to reduce the noise of the partition image, uses the first / second segmentation model to eliminate two types of interference factor images in the partition image (brain sulcus area image and cerebral arterial old infarction area image), uses the third segmentation model + the first classification model and a side information comprehensive evaluation mechanism to improve the positioning accuracy of the affected side, and uses the batch processing mechanism of the first scoring model to improve the partition scoring efficiency; through the embodiment of the present invention, not only the overall prediction efficiency is further improved, but also the boundary accuracy of the partition image is improved, the noise and interference of the partition image are reduced, the positioning accuracy of the affected side is improved, and the prediction accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 A schematic diagram of a processing method for predicting ASPECTS scores based on CT images provided in Example 1 of the present invention;

[0103] Figure 2 A module structure diagram of the first classification model provided in the first embodiment of the present invention;

[0104] Figure 3 This is a module structure diagram of the first scoring model provided in the first embodiment of the present invention;

[0105] Figure 4 A module structure diagram of a processing device for predicting ASPECTS scores based on CT images provided in Embodiment 2 of the present invention;

[0106] Figure 5 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0107] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0108] Embodiment 1 of the present invention provides a processing method for predicting ASPECTS scores based on CT images, such as Figure 1 A schematic diagram of a processing method for predicting ASPECTS scores based on CT images provided in the first embodiment of the present invention is shown. The method mainly includes the following steps:

[0109] Step 1, construct an image semantic segmentation model for performing sulcus region segmentation on brain CT images, which is recorded as the corresponding first segmentation model; and construct an image semantic segmentation model for performing cerebral arterial vascular old infarction region segmentation on brain CT images, which is recorded as the corresponding second segmentation model; and construct an image semantic segmentation model for performing cerebral vascular high-density arterial sign region segmentation on brain CT images, which is recorded as the corresponding third segmentation model; and construct an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on brain CT image sequences, which is recorded as the corresponding fourth segmentation model; and construct a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images; and construct a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs; and collectively train the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on a preset plain scan CT image library;

[0110] Specifically, it includes: step 11, constructing an image semantic segmentation model for performing sulcus region segmentation on a brain CT image, which is recorded as a corresponding first segmentation model;

[0111] Here, the first segmentation model of the embodiment of the present invention is implemented based on the U-Net image segmentation network; the first segmentation model is used to perform sulcus region segmentation processing on the current CT image input by the model to obtain the corresponding first semantic map; wherein the current CT image here is an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei of the brain or an axial plain scan CT image of the brain located at a level above the basal ganglia nuclei of the brain; the first semantic map includes multiple first pixel points; each first pixel point corresponds to a first semantic type; the first semantic type includes sulcus region semantics and background semantics;

[0112] Step 12, constructing an image semantic segmentation model for segmenting the old infarction area of ​​the cerebral artery in the brain CT image, which is recorded as the corresponding second segmentation model;

[0113] Here, the second segmentation model of the embodiment of the present invention is implemented based on the U-Net image segmentation network; the second segmentation model is used to perform brain artery old infarction area segmentation processing on the current CT image input by the model to obtain a corresponding second semantic map; wherein the current CT image here is an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain or an axial brain plain scan CT image located at a level above the basal ganglia nuclei of the brain; the second semantic map includes multiple second pixel points; each second pixel point corresponds to a second semantic type; the second semantic type includes old infarction area semantics and background semantics;

[0114] Step 13, constructing an image semantic segmentation model for segmenting the brain vascular high-density arterial sign area of ​​the brain CT image, which is recorded as the corresponding third segmentation model;

[0115] Here, it should be noted that the high-density arterial sign of brain blood vessels mentioned here and the cytotoxic edema of brain tissue mentioned below are each a typical sign of ischemic stroke;

[0116] The third segmentation model of the embodiment of the present invention is implemented based on the U-Net image segmentation network; the third segmentation model is used to perform brain vascular high-density arterial sign area segmentation processing on the current CT image input by the model to obtain a corresponding third semantic map; wherein the current CT image here is an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain or an axial brain plain scan CT image located at a level above the basal ganglia nuclei of the brain; the third semantic map includes multiple third pixel points; each third pixel point corresponds to a third semantic type; the third semantic type includes high-density arterial sign area semantics and background semantics;

[0117] Step 14, constructing an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on the brain CT image sequence, which is recorded as the corresponding fourth segmentation model;

[0118] Here, it should be noted that the ASPECTS partitioning of the embodiment of the present invention is composed of ten types of brain tissue partitions, namely C partition, L partition, IC partition, I partition, M1 partition, M2 partition, M3 partition, M4 partition, M5 partition and M6 partition; C partition is the caudate nucleus area of ​​the brain; L partition is the lenticular nucleus area of ​​the brain; IC partition is the internal capsule area of ​​the brain; I partition is the insular cortex area of ​​the brain; M1 partition is the anterior cortical area of ​​the middle cerebral artery; M2 partition is the lateral cortical area of ​​the insular lobe of the middle cerebral artery; M3 partition is the posterior cortical area of ​​the middle cerebral artery; M4 partition is the middle cerebral artery cortical area above the M1 partition; M5 partition is the middle cerebral artery cortical area above the M2 partition; M6 partition is the middle cerebral artery cortical area above the M3 partition; various types of brain tissue partitions of the ASPECTS partitioning of the embodiment of the present invention are symmetrical between the left and right brains;

[0119] The fourth segmentation model of the embodiment of the present invention is implemented based on two U-Net image segmentation networks, which are recorded as the corresponding first U-Net network and the second U-Net network; the fourth segmentation model is used to perform left and right brain ASPECTS partitioning and ventricular region segmentation processing on the current CT image sequence input by the model to obtain the corresponding fourth semantic map sequence, specifically: the fourth segmentation model is used to use the first CT image in the current CT image sequence input by the model as the corresponding first network input image, and the second CT image in the current CT image sequence as the corresponding second network input image; and the first U-Net network performs fourteen categories of left and right brain ASPECTS partitioning and ventricular region segmentation processing on the first network input image to obtain the corresponding fourth-first semantic map; and the second U-Net network performs six categories of left and right brain ASPECTS partitioning and ventricular region segmentation processing on the second network input image to obtain the corresponding fourth-second semantic map; and the obtained fourth-first and fourth-second semantic maps are sequentially sorted to form a corresponding fourth semantic map sequence output;

[0120] The current CT image sequence is formed by sequencing two CT images, wherein the first CT image is an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei, and the second CT image is an axial plain scan CT image of the brain located at a level above the basal ganglia nuclei;

[0121] The fourth-first semantic map includes a plurality of fourth-first pixels; each fourth-first pixel corresponds to a fourth-first semantic type; the fourth-first semantic types include fourteen categories of left and right brain ASPECTS partition semantics, ventricular region semantics and background semantics; the fourteen categories of left and right brain ASPECTS partition semantics include left / right brain C partition semantics, left / right brain L partition semantics, left / right brain IC partition semantics, left / right brain I partition semantics, left / right brain M1 partition semantics, left / right brain M2 partition semantics, and left / right brain M3 partition semantics;

[0122] The fourth second semantic map includes a plurality of fourth second pixels; each fourth second pixel corresponds to a fourth second semantic type; the fourth second semantic type includes six categories of left and right brain ASPECTS partition semantics, ventricular region semantics and background semantics; the six categories of left and right brain ASPECTS partition semantics include left / right brain M4 region semantics, left / right brain M5 region semantics, and left / right brain M6 region semantics;

[0123] Step 15, constructing a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images;

[0124] Here, if Figure 2As shown in the module structure diagram of the first classification model provided in the first embodiment of the present invention, the first classification model of the embodiment of the present invention includes a first feature extraction module and a first classification module; the first feature extraction module is implemented based on the ResNet network; the first classification module is implemented based on the MLP network; the first classification model is used to classify and identify the brain side information that produces brain tissue cytotoxic edema on the current CT image input by the model to obtain the corresponding predicted side, specifically: the first classification model is used to input the current CT image input by the model into the first feature extraction module for feature extraction processing to obtain the corresponding feature tensor; and input the feature tensor into the first classification module for classification prediction processing to obtain the corresponding prediction type; wherein the prediction types include empty, left, right and bilateral;

[0125] Step 16, and constructing a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs;

[0126] Here, if Figure 3 As shown in the module structure diagram of the first scoring model provided in the first embodiment of the present invention, the first scoring model of the embodiment of the present invention is composed of ten parallel partition scoring models and one partition scoring merging module; each partition scoring model is composed of a left and right brain image sorting module, a left brain feature extraction module, a right brain feature extraction module, a left and right brain feature difference module, a left brain feature fusion module, a right brain feature fusion module, a left brain score prediction module, a right brain score prediction module and a left and right brain score merging module; the left and right brain feature extraction modules are both based on a solidified omics feature extraction program, or a dynamically settable omics feature extraction tool interface, and the omics feature extraction tool is a PyRadiomics software interface; the left and right brain score prediction modules are both based on a binary classification prediction model that performs a binary 0 / 1 binary classification prediction based on an input feature vector / tensor, and the left and right brain score prediction modules can also be based on a regression calculation model that performs a binary 0 / 1 prediction data regression calculation based on an input feature vector / tensor.

[0127] The first scoring model of the embodiment of the present invention is used to perform ASPECTS scoring on the current partition image pair sequence input by the model to obtain a corresponding partition score pair sequence, specifically: the first scoring model is used to input each partition image pair in the current partition image pair sequence input by the model into the corresponding partition scoring model, and each partition scoring model performs left and right brain ASPECTS scoring according to the corresponding partition image pair to obtain a corresponding partition score pair and send it to the partition scoring merging module; and the partition scoring merging module sequentially sorts all the obtained partition score pairs to form a corresponding partition score pair sequence;

[0128] The current partition image pair sequence of the embodiment of the present invention is formed by sequentially sorting ten partition image pairs; each partition image pair corresponds to a type of partition in the ASPECTS partition; the partition image pair corresponds to the partition scoring model in the first scoring model one by one; each partition image is composed of a corresponding left brain partition image and a right brain partition image;

[0129] The partition score pair sequence of the embodiment of the present invention is formed by sequentially sorting ten partition score pairs; the partition score pairs correspond to the partition image pairs one by one; each partition score pair is composed of a corresponding left brain partition value and a right brain partition value; the left and right brain partition values ​​are 0 or 1;

[0130] Each partition scoring model performs left and right brain ASPECTS scoring according to the corresponding partition image pair to obtain the corresponding partition score pair and send it to the partition score merging module, which specifically includes:

[0131] Step A1, each partition scoring model inputs the corresponding partition image pair into the left and right brain image sorting module, and the left and right brain image sorting module extracts the corresponding left brain partition image and right brain partition image from the partition image pair and sends them to the corresponding left brain feature extraction module and right brain feature extraction module;

[0132] Step A2, the left brain feature extraction module extracts radiomics features from the left brain partition image to obtain corresponding left brain radiomics features, and sends them to the left and right brain feature difference module and the left brain feature fusion module;

[0133] Step A3, the right brain feature extraction module extracts the imaging features of the right brain partition image to obtain the corresponding right brain imaging features and sends them to the left and right brain feature difference module and the right brain feature fusion module;

[0134] Step A4, the left and right brain feature difference module records the difference feature of the left brain imaging omics feature minus the right brain imaging omics feature and the difference feature of the right brain imaging omics feature minus the left brain imaging omics feature as the corresponding left-right brain imaging omics difference feature and right-left brain imaging omics difference feature, and sends the left-right brain imaging omics difference feature to the corresponding left brain feature fusion module, and sends the right-left brain imaging omics difference feature to the right brain feature fusion module;

[0135] Step A5: The left brain feature fusion module performs feature splicing processing on the left brain imaging features and the left-right brain imaging differential features to obtain the corresponding left brain splicing features and send them to the left brain score prediction module;

[0136] Step A6: The right brain feature fusion module performs feature splicing processing on the right brain imaging features and the right-left brain imaging differential features to obtain the corresponding right brain splicing features and send them to the right brain score prediction module;

[0137] Step A7, the left brain score prediction module performs binarized ASPECTS score prediction processing based on the left brain splicing features to obtain the corresponding left brain partition score and sends it to the left and right brain score merging module;

[0138] Step A8: The right brain score prediction module performs binarized ASPECTS score prediction processing based on the right brain splicing features to obtain the corresponding right brain partition score and sends it to the left and right brain score merging module;

[0139] Step A9: the left and right brain scoring merging module sends the corresponding partition scoring pairs composed of the left brain partition scoring and the right brain partition scoring to the partition scoring merging module;

[0140] Step 17, collective model training is performed on the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on the preset plain scan CT image library;

[0141] Among them, the plain scan CT image library of the embodiment of the present invention includes multiple first plain scan CT image groups; each first plain scan CT image group corresponds to two plain scan CT acquisition images of an image acquisition object; the first plain scan CT image group includes a first acquisition object type, a first CT acquisition image and a second CT acquisition image; the first acquisition object type is used to mark the characteristic type of the corresponding image acquisition object, at least including a healthy type without brain disease, a type with a history of old infarction of cerebral arterial vessels but no symptoms of ischemic stroke, a type of class A ischemic stroke with no history of old infarction of cerebral arterial vessels but with symptoms of ischemic stroke, and a type of class B ischemic stroke with a history of old infarction of cerebral arterial vessels and symptoms of ischemic stroke; here, the symptoms of ischemic stroke include high density of cerebral blood vessels one or all of the following: arterial sign and brain tissue cytotoxic edema; the first CT acquisition image is an axial brain plain scan CT image of the corresponding image acquisition object located at the level of the basal ganglia nuclei of the brain; the second CT acquisition image is an axial brain plain scan CT image of the corresponding image acquisition object located at a level above the basal ganglia nuclei of the brain; the acquisition method of the first and second CT acquisition images is: an axial brain CT plain scan is performed on the image acquisition object to obtain a corresponding plain scan CT image sequence of the time, and a CT image located at the level of the basal ganglia nuclei of the brain is extracted from the plain scan CT image sequence of the time as the corresponding first CT acquisition image, and a CT image located at a level above the basal ganglia nuclei of the brain is extracted from the plain scan CT image sequence of the time as the corresponding second CT acquisition image;

[0142] The current step 17 specifically includes:

[0143] Specifically include:

[0144] Step 171, taking each first CT acquisition image or second CT acquisition image in the plain scan CT image library as a corresponding first training image; marking the sulcus area and the background area on each first training image by manual marking or other tool marking methods, and creating a label semantic map based on the marking results to obtain a corresponding first label semantic map; and forming a corresponding first data record from each first training image and the corresponding first label semantic map; and forming a corresponding first data set from all the obtained first data records;

[0145] Here, the first data set obtained in the embodiment of the present invention includes multiple first data records; the first data record includes a first training image and a first label semantic map; the first label semantic map includes multiple first label map pixels; each first label map pixel corresponds to a first semantic type;

[0146] Step 172, and use each first CT acquisition image or second CT acquisition image in the plain scan CT image library as a corresponding second training image; and mark the old infarction area and background area of ​​the cerebral artery on each second training image by manual marking or other tool marking methods, and create a label semantic map based on the marking result to obtain a corresponding second label semantic map; and each second training image and the corresponding second label semantic map form a corresponding second data record; and all the obtained second data records form a corresponding second data set;

[0147] Here, the second data set obtained in the embodiment of the present invention includes a plurality of second data records; the second data record includes a second training image and a second label semantic map; the second label semantic map includes a plurality of second label map pixels; each second label map pixel corresponds to a second semantic type;

[0148] Step 173, each first CT acquisition image or second CT acquisition image in the plain scan CT image library is used as a corresponding third training image; a high-density arterial sign area and a background area of ​​cerebral blood vessels are marked on each third training image by manual marking or other tool marking methods, and a label semantic map is created based on the marking results to obtain a corresponding third label semantic map; each third training image and the corresponding third label semantic map form a corresponding third data record; and all the obtained third data records form a corresponding third data set;

[0149] Here, the third data set obtained in the embodiment of the present invention includes multiple third data records; the third data record includes a third training image and a third label semantic map; the third label semantic map includes multiple third label map pixels; each third label map pixel corresponds to a third semantic type;

[0150] Step 174, and use the first CT acquisition image and the second CT acquisition image of each first plain scan CT image group in the plain scan CT image library as the corresponding fourth-first training image and the fourth-second training image, and sequentially sort the fourth-first and the fourth-second training images corresponding to each first plain scan CT image group to form a corresponding fourth training image sequence; and perform sixteen types of semantic region marking corresponding to the fourth-first semantic type on the fourth-first training image of each fourth training image sequence through manual marking or other tool marking methods, and create a label semantic map based on the marking results to obtain a corresponding fourth-first label semantic map; and perform eight types of semantic region marking corresponding to the fourth-second semantic type on the fourth-second training image of each fourth training image sequence through manual marking or other tool marking methods, and create a label semantic map based on the marking results to obtain a corresponding fourth-second label semantic map; and sequentially sort the fourth-first and the fourth-second label semantic maps corresponding to each fourth training image sequence to form a corresponding fourth label semantic map sequence; and each fourth training image sequence and the corresponding fourth label semantic map sequence form a corresponding fourth data record; and all the obtained fourth data records form a corresponding fourth data set;

[0151] Here, the fourth data set obtained by the embodiment of the present invention includes multiple fourth data records; the fourth data record includes a fourth training image sequence and a fourth label semantic map sequence; the fourth training image sequence is formed by sequentially sorting the fourth-first training image and the fourth-second training image; the fourth label semantic map sequence is formed by sequentially sorting the fourth-first label semantic map and the fourth-second label semantic map; the fourth-first label semantic map includes multiple fourth-first label map pixels; each fourth-first label map pixel corresponds to a fourth-first semantic type; the fourth-second label semantic map includes multiple fourth-second label map pixels; each fourth-second label map pixel corresponds to a fourth-second semantic type;

[0152] Step 175, and use each first CT acquisition image or second CT acquisition image in the plain scan CT image library as a corresponding fifth training image; and identify whether brain tissue cytotoxic edema appears on each fifth training image by manual recognition or other tool recognition methods; if it is identified that brain tissue cytotoxic edema does not appear on the current fifth training image, set the corresponding first label predicted side to be empty; if it is identified that brain tissue cytotoxic edema appears on the current fifth training image, further identify the left and right brain regions where brain tissue cytotoxic edema appears by manual recognition or other tool recognition methods, if it is identified that brain tissue cytotoxic edema appears only in the left brain region, set the corresponding first label predicted side to the left, if it is identified that brain tissue cytotoxic edema appears only in the right brain region, set the corresponding first label predicted side to the right, if it is identified that brain tissue cytotoxic edema appears in both left and right brain regions, set the corresponding first label predicted side to bilateral; and each fifth training image and the corresponding first label predicted side form a corresponding fifth data record; and all the obtained fifth data records form a corresponding fifth data set;

[0153] Here, the fifth data set obtained in the embodiment of the present invention includes a plurality of fifth data records; the fifth data record includes a fifth training image and a first label predicted side; the first label predicted side includes empty, left, right, and both sides;

[0154] Step 176, and use each first plain scan CT image group in the plain scan CT image library as a corresponding current CT image group; and extract corresponding left and right brain C area image pairs, left and right brain L area image pairs, left and right brain IC area image pairs, left and right brain I area image pairs, left and right brain M1 area image pairs, left and right brain M2 area image pairs, left and right brain M3 area image pairs, left and right brain M4 area image pairs, left and right brain M5 area image pairs and left and right brain M6 area image pairs from the first and second CT acquisition images of the current CT image group through an artificial image extraction method or an image extraction method using other tools, and use each extracted image pair as a corresponding first training partition image pair, and sequentially sort the obtained ten first training partition image pairs to obtain the corresponding first A training partition image pair sequence; and setting the left and right brain partition scores corresponding to each first training partition image pair to 0 / 1 by means of a manual score setting method or a score setting method of other tools, and recording them as the corresponding first label left brain partition score and the first label right brain partition score; and forming a corresponding first label partition score pair by the first label left brain partition score and the first label right brain partition score of each first training partition image pair; and sequentially sorting all the obtained first label partition score pairs to obtain a corresponding first label partition score pair sequence; and forming a corresponding sixth data record by the first training partition image pair sequence and the first label partition score pair sequence; and forming a corresponding sixth data set by all the obtained sixth data records;

[0155] Here, the sixth data set obtained in the embodiment of the present invention includes a plurality of sixth data records; the sixth data record includes a first training partition image pair sequence and a first label partition score pair sequence;

[0156] Step 177, after obtaining the first, second, third, fourth, fifth and sixth data sets, perform a first model training on the first segmentation model based on the first data set; perform a second model training on the second segmentation model based on the second data set; perform a third model training on the third segmentation model based on the third data set; perform a fourth model training on the fourth segmentation model based on the fourth data set; perform a fifth model training on the first classification model based on the fifth data set; and perform a sixth model training on the first scoring model based on the sixth data set;

[0157] Specifically comprising: step 1771, performing a first model training on a first segmentation model based on a first data set;

[0158] Specifically include:

[0159] Step 1771-1, randomly dividing the first data set into training / evaluation subsets based on a preset first training-evaluation ratio to obtain a corresponding first training data set and a first evaluation data set;

[0160] Wherein, the first training evaluation ratio is a preset ratio; the first training data set and the first evaluation data set are both composed of a plurality of first data records; the ratio of the total number of records of the first training data set to the total number of records of the first evaluation data set satisfies the first training evaluation ratio;

[0161] Step 1771-2, extracting the first first data record of the first training data set as the corresponding current training record;

[0162] Step 1771-3, inputting the first training image of the current training record into the first segmentation model to perform brain sulcus region segmentation processing to obtain a corresponding first predicted semantic map;

[0163] Step 1771-4, bringing the first predicted semantic graph and the first label semantic graph of the current training record into a preset first model loss function to calculate and obtain a corresponding first loss value;

[0164] The loss function type of the first model loss function includes at least an L1 loss function, an L2 loss function, and a binary cross entropy loss function;

[0165] Step 1771-5, identifying whether the first loss value satisfies a preset first loss value range; if the first loss value satisfies the first loss value range, identifying whether the current training record is the last first data record of the first training data set, if so, turning to step 1771-6, if not, extracting the next first data record of the first training data set as a new current training record and returning to step 1771-3; if the first loss value does not satisfy the first loss value range, performing a round of parameter optimization on the first segmentation model in the direction of minimizing the first model loss function based on a preset first model parameter optimizer, and returning to step 1771-3 at the end of this round of parameter optimization;

[0166] The first loss value range is a preset loss value range; the first model parameter optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0167] Step 1771-6, perform a round of traversal of all first data records of the first evaluation data set; and during the traversal, use the currently traversed first data record as the corresponding current evaluation record; and input the first training image of the current evaluation record into the first segmentation model to perform sulcus region segmentation processing to obtain the corresponding second predicted semantic map; and according to the conventional binary classification evaluation method based on precision, recall and F-score, use the first label semantic map of the current evaluation record as the true value reference to calculate the binary classification precision and binary classification recall of the pixel-level semantics of the second predicted semantic map to obtain the corresponding first precision and first recall, and calculate the corresponding F-score based on the first precision and first recall to obtain the corresponding first F-score; and at the end of this round of traversal, perform an average calculation on all the first precisions obtained to obtain the corresponding first average precision, and perform an average calculation on all the first recalls obtained to obtain the corresponding first average recall, and perform an average calculation on all the first F-scores obtained to obtain the corresponding first average F-score;

[0168] Step 1771-7, identifying the first average precision, the first average recall and the first average F-score based on the preset first precision range, the first recall range and the first F-score range; if the first average precision does not satisfy the first precision range or the first average recall does not satisfy the first recall range or the first average F-score does not satisfy the first F-score range, returning to step 1771-1 to continue training; if the first average precision satisfies the first precision range and the first average recall satisfies the first recall range and the first average F-score satisfies the first F-score range, stopping training and confirming that the first model training is completed;

[0169] Here, the first precision range, the first recall range, and the first F-score range are a set of preset precision, recall, and F-score ranges;

[0170] Step 1772, performing a second model training on the second segmentation model based on the second data set;

[0171] Specifically include:

[0172] Step 1772-1, randomly dividing the second data set into training / evaluation subsets based on a preset second training-evaluation ratio to obtain a corresponding second training data set and a second evaluation data set;

[0173] The second training evaluation ratio is a preset ratio; both the second training data set and the second evaluation data set are composed of a plurality of second data records; the ratio of the total number of records in the second training data set to the total number of records in the second evaluation data set satisfies the second training evaluation ratio;

[0174] Step 1772-2, extracting the first second data record of the second training data set as the corresponding current training record;

[0175] Step 1772-3, inputting the second training image of the current training record into the second segmentation model to perform segmentation processing on the old infarction area of ​​the cerebral artery to obtain a corresponding third predicted semantic map;

[0176] Step 1772-4, bringing the third predicted semantic graph and the second label semantic graph of the current training record into a preset second model loss function to calculate and obtain a corresponding second loss value;

[0177] The loss function type of the second model loss function includes at least an L1 loss function, an L2 loss function, and a binary cross entropy loss function;

[0178] Step 1772-5, identifying whether the second loss value satisfies a preset second loss value range; if the second loss value satisfies the second loss value range, identifying whether the current training record is the last second data record of the second training data set, if so, turning to step 1772-6, if not, extracting the next second data record of the second training data set as a new current training record and returning to step 1772-3; if the second loss value does not satisfy the second loss value range, performing a round of parameter optimization on the second segmentation model in the direction of minimizing the second model loss function based on a preset second model parameter optimizer, and returning to step 1772-3 at the end of this round of parameter optimization;

[0179] The second loss value range is a preset loss value range; the second model parameter optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0180] Step 1772-6, perform a round of traversal of all second data records of the second evaluation data set; and during the traversal, use the second data record currently traversed as the corresponding current evaluation record; and input the second training image of the current evaluation record into the second segmentation model to perform segmentation processing on the old infarction area of ​​the cerebral artery to obtain the corresponding fourth predicted semantic map; and according to the conventional binary classification evaluation method based on precision, recall and F-score, use the second label semantic map of the current evaluation record as the true value reference to calculate the binary classification precision and binary classification recall of the pixel-level semantics of the fourth predicted semantic map to obtain the corresponding second precision and second recall, and calculate the corresponding F-score based on the second precision and the second recall to obtain the corresponding second F-score; and at the end of this round of traversal, perform an average calculation on all the obtained second precisions to obtain the corresponding second average precision, and perform an average calculation on all the obtained second recalls to obtain the corresponding second average recall, and perform an average calculation on all the obtained second F-scores to obtain the corresponding second average F-score;

[0181] Step 1772-7, identifying the second average precision, the second average recall and the second average F-score based on the preset second precision range, the second recall range and the second F-score range; if the second average precision does not satisfy the second precision range or the second average recall does not satisfy the second recall range or the second average F-score does not satisfy the second F-score range, returning to step 1772-1 to continue training; if the second average precision satisfies the second precision range and the second average recall satisfies the second recall range and the second average F-score satisfies the second F-score range, stopping training and confirming that the second model training is finished;

[0182] Here, the second precision range, the second recall range, and the second F-score range are a set of preset precision, recall, and F-score ranges;

[0183] Step 1773, performing third model training on the third segmentation model based on the third data set;

[0184] Specifically include:

[0185] Step 1773-1, randomly dividing the third data set into training / evaluation subsets based on a preset third training-evaluation ratio to obtain a corresponding third training data set and a third evaluation data set;

[0186] Wherein, the third training evaluation ratio is a preset ratio; the third training data set and the third evaluation data set are both composed of a plurality of third data records; the ratio of the total number of records of the third training data set to the total number of records of the third evaluation data set satisfies the third training evaluation ratio;

[0187] Step 1773-2, extracting the first third data record of the third training data set as the corresponding current training record;

[0188] Step 1773-3, inputting the third training image of the current training record into the third segmentation model to perform segmentation processing on the high-density arterial sign area of ​​the brain blood vessels to obtain the corresponding fifth predicted semantic map;

[0189] Step 1773-4, bringing the fifth predicted semantic graph and the third label semantic graph of the current training record into a preset third model loss function to calculate and obtain a corresponding third loss value;

[0190] The loss function types of the third model loss function include at least an L1 loss function, an L2 loss function, and a binary cross entropy loss function;

[0191] Step 1773-5, identifying whether the third loss value satisfies a preset third loss value range; if the third loss value satisfies the third loss value range, identifying whether the current training record is the last third data record of the third training data set, if so, turning to step 1773-6, if not, extracting the next third data record of the third training data set as the new current training record and returning to step 1773-3; if the third loss value does not satisfy the third loss value range, performing a round of parameter optimization on the third segmentation model in the direction of minimizing the third model loss function based on a preset third model parameter optimizer, and returning to step 1773-3 at the end of this round of parameter optimization;

[0192] The third loss value range is a preset loss value range; the third model parameter optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0193] Step 1773-6, perform a round of traversal of all third data records of the third evaluation data set; and during the traversal, use the third data record currently traversed as the corresponding current evaluation record; and input the third training image of the current evaluation record into the third segmentation model to perform brain vascular high-density arterial sign area segmentation processing to obtain the corresponding sixth predicted semantic map; and according to the conventional binary classification evaluation method based on precision, recall and F-score, use the third label semantic map of the current evaluation record as the true value reference to calculate the binary classification precision and binary classification recall of the pixel-level semantics of the sixth predicted semantic map to obtain the corresponding third precision and third recall, and calculate the corresponding F score based on the third precision and third recall to obtain the corresponding third F score; and at the end of this round of traversal, perform the average calculation on all the obtained third precisions to obtain the corresponding third average precision, and perform the average calculation on all the obtained third recalls to obtain the corresponding third average recall, and perform the average calculation on all the obtained third F scores to obtain the corresponding third average F score;

[0194] Step 1773-7, identifying the third average precision, the third average recall and the third average F-score based on the preset third precision range, the third recall range and the third F-score range; if the third average precision does not satisfy the third precision range or the third average recall does not satisfy the third recall range or the third average F-score does not satisfy the third F-score range, returning to step 1773-1 to continue training; if the third average precision satisfies the third precision range and the third average recall satisfies the third recall range and the third average F-score satisfies the third F-score range, stopping training and confirming that the third model training is completed;

[0195] Here, the third precision range, the third recall range, and the third F-score range are a set of preset precision, recall, and F-score ranges;

[0196] Step 1774, performing a fourth model training on a fourth segmentation model based on the fourth data set;

[0197] Specifically include:

[0198] Step 1774-1, randomly dividing the fourth data set into training / evaluation subsets based on a preset fourth training-evaluation ratio to obtain a corresponding fourth training data set and a fourth evaluation data set;

[0199] The fourth training evaluation ratio is a preset ratio; both the fourth training data set and the fourth evaluation data set are composed of a plurality of fourth data records; the ratio of the total number of records of the fourth training data set to the total number of records of the fourth evaluation data set satisfies the fourth training evaluation ratio;

[0200] Step 1774-2, extracting the first third data record of the fourth training data set as the corresponding current training record;

[0201] Step 1774-3, inputting the fourth training image sequence of the current training record into the fourth segmentation model to perform left and right brain ASPECTS partitioning and ventricular region segmentation processing to obtain the corresponding seventh prediction semantic map sequence; the seventh prediction semantic map sequence is formed by sequentially sorting the seventh-first prediction semantic map and the seventh-second prediction semantic map;

[0202] Step 1774-4, bringing the seventh predicted semantic graph sequence and the fourth label semantic graph sequence of the current training record into a preset fourth model loss function to calculate and obtain a corresponding fourth loss value;

[0203] in,

[0204] w1 and w2 are two preset weighting coefficients. is the preset fourth model loss function, is the preset loss function of the fourth model, Y1 is the seventh first prediction semantic graph of the seventh prediction semantic graph sequence, Y2 is the seventh second prediction semantic graph of the seventh prediction semantic graph sequence, is the fourth one-label semantic graph of the fourth label semantic graph sequence of the current training record, The fourth second label semantic graph of the fourth label semantic graph sequence of the current training record; the loss function types of the fourth first and fourth second model loss functions are the same, and the loss function types of the fourth first and fourth second model loss functions at least include L1 loss function, L2 loss function and binary cross entropy loss function;

[0205] Step 1774-5, identifying whether the fourth loss value satisfies a preset fourth loss value range; if the fourth loss value satisfies the fourth loss value range, identifying whether the current training record is the last fourth data record of the fourth training data set, if so, turning to step 1774-6, if not, extracting the next fourth data record of the fourth training data set as a new current training record and returning to step 1774-3; if the fourth loss value does not satisfy the fourth loss value range, performing a round of parameter optimization on the fourth segmentation model in the direction of minimizing the loss functions of the fourth first and fourth second models based on a preset fourth model parameter optimizer, and returning to step 1774-3 at the end of this round of parameter optimization;

[0206] The fourth loss value range is a preset loss value range; the fourth model parameter optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0207] Step 1774-6, perform a round of traversal of all fourth data records of the fourth evaluation data set; and during the traversal, use the currently traversed fourth data record as the corresponding current evaluation record; and input the fourth training image sequence of the current evaluation record into the fourth segmentation model for left and right brain ASPECTS partitioning and ventricular region segmentation processing to obtain the corresponding eighth predicted semantic map sequence; and according to the conventional multi-classification evaluation method based on precision, recall and F-score, use the fourth-first label semantic map of the fourth label semantic map sequence of the current evaluation record as the true value reference to calculate the multi-classification precision and multi-classification recall of the pixel-level semantics of the eighth-first predicted semantic map of the eighth predicted semantic map sequence to obtain the corresponding fourth-first precision and fourth-first recall, and calculate the corresponding F-score based on the fourth-first precision and fourth-first recall to obtain the corresponding fourth-first F-score; and according to the conventional multi-classification evaluation method based on precision, recall and F-score, use the fourth-second label semantic map of the fourth label semantic map sequence of the current evaluation record as the true value reference to calculate the eighth predicted The multi-classification precision and multi-classification recall of the pixel-level semantics of the eighth second prediction semantic graph of the semantic graph sequence are calculated to obtain the corresponding fourth second precision and fourth second recall, and the corresponding F score is calculated based on the fourth second precision and fourth second recall to obtain the corresponding fourth second F score; and the corresponding fourth precision is calculated based on the fourth first and fourth second precisions = a1×fourth first precision + a2×fourth second precision, where a1 and a2 are two preset weighting coefficients; and the corresponding fourth recall is calculated based on the fourth first and fourth second recalls = a1×fourth first recall + a2×fourth second recall; and the corresponding fourth F score is calculated based on the fourth first and fourth second F scores = a1×fourth first F score + a2×fourth second F score; and at the end of this round of traversal, all the fourth precisions obtained are averaged to obtain the corresponding fourth average precision, and all the fourth recalls obtained are averaged to obtain the corresponding fourth average recall, and all the fourth F scores obtained are averaged to obtain the corresponding fourth average F score;

[0208] The eighth predicted semantic graph sequence is formed by sequentially sorting the eighty-first predicted semantic graph and the eighty-second predicted semantic graph;

[0209] Step 1774-7, identifying the fourth average precision, the fourth average recall and the fourth average F-score based on the preset fourth precision range, the fourth recall range and the fourth F-score range; if the fourth average precision does not satisfy the fourth precision range or the fourth average recall does not satisfy the fourth recall range or the fourth average F-score does not satisfy the fourth F-score range, returning to step 1774-1 to continue training; if the fourth average precision satisfies the fourth precision range and the fourth average recall satisfies the fourth recall range and the fourth average F-score satisfies the fourth F-score range, stopping training and confirming that the fourth model training is completed;

[0210] Here, the fourth precision range, the fourth recall range, and the fourth F-score range are a set of preset precision, recall, and F-score ranges;

[0211] Step 1775, performing fifth model training on the first classification model based on the fifth data set;

[0212] Specifically include:

[0213] Step 1775-1, randomly dividing the fifth data set into training / evaluation subsets based on a preset fifth training-evaluation ratio to obtain a corresponding fifth training data set and a fifth evaluation data set;

[0214] Wherein, the fifth training evaluation ratio is a preset ratio; the fifth training data set and the fifth evaluation data set are both composed of a plurality of fifth data records; the ratio of the total number of records of the fifth training data set to the total number of records of the fifth evaluation data set satisfies the fifth training evaluation ratio;

[0215] Step 1775-2, extracting the first fifth data record of the fifth training data set as the corresponding current training record;

[0216] Step 1775-3, inputting the fifth training image of the current training record into the first classification model, and using the first classification model to classify and identify the brain side information of the brain tissue cytotoxic edema generated in the fifth training image of the current training record to obtain the corresponding first training predicted side;

[0217] Step 1775-4, bringing the first training prediction aspect and the first label prediction aspect of the current training record into the preset fifth model loss function to calculate and obtain the corresponding fifth loss value;

[0218] Among them, the loss function type of the fifth model loss function includes at least an L1 loss function, an L2 loss function, and a multi-classification cross entropy loss function;

[0219] Step 1775-5, identifying whether the fifth loss value satisfies the preset fifth loss value range; if the fifth loss value satisfies the fifth loss value range, identifying whether the current training record is the last fifth data record of the fifth training data set, if so, turning to step 1775-6, if not, extracting the next fifth data record of the fifth training data set as the new current training record and returning to step 1775-3; if the fifth loss value does not satisfy the fifth loss value range, performing a round of parameter optimization on the first classification model in the direction of minimizing the fifth model loss function based on the preset fifth model parameter optimizer, and returning to step 1775-3 at the end of this round of parameter optimization;

[0220] The fifth loss value range is a preset loss value range; the fifth model parameter optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0221] Step 1775-6, perform a round of traversal on all fifth data records of the fifth evaluation data set; and during the traversal, use the fifth data record currently traversed as the corresponding current evaluation record; and input the fifth training image of the current evaluation record into the first classification model, and use the first classification model to classify and identify the brain side information of brain tissue cytotoxic edema on the fifth training image of the current evaluation record to obtain the corresponding second training predicted side; and form a corresponding prediction-label data pair with the second training predicted side and the first label predicted side of the current evaluation record; and at the end of this round of traversal, form a corresponding prediction-label data pair set with all the obtained prediction-label data pairs;

[0222] Step 1775-7, according to the conventional multi-classification evaluation method based on precision, recall and F-score, the precision and recall of the set are calculated according to the prediction-label data to obtain the corresponding fifth precision and fifth recall; and the corresponding F-score is calculated based on the fifth precision and the fifth recall to obtain the corresponding fifth F-score;

[0223] Step 1775-8, identifying the fifth precision, the fifth recall and the fifth F-score based on the preset fifth precision range, the fifth recall range and the fifth F-score range; if the fifth precision does not satisfy the fifth precision range or the fifth recall does not satisfy the fifth recall range or the fifth F-score does not satisfy the fifth F-score range, returning to step 1775-1 to continue training; if the fifth precision satisfies the fifth precision range and the fifth recall satisfies the fifth recall range and the fifth F-score satisfies the fifth F-score range, stopping training and confirming that the fifth model training is completed;

[0224] Here, the fifth precision range, the fifth recall range, and the fifth F-score range are a set of preset precision, recall, and F-score ranges;

[0225] Step 1776, performing a sixth model training on the first scoring model based on the sixth data set;

[0226] Specifically include:

[0227] Step 1776-1, randomly dividing the sixth data set into training / evaluation subsets based on a preset sixth training-evaluation ratio to obtain a corresponding sixth training data set and a sixth evaluation data set;

[0228] The sixth training evaluation ratio is a preset ratio; the sixth training data set and the sixth evaluation data set are both composed of a plurality of sixth data records; the ratio of the total number of records in the sixth training data set to the total number of records in the sixth evaluation data set satisfies the sixth training evaluation ratio;

[0229] Step 1776-2, extracting the first sixth data record of the sixth training data set as the corresponding current training record;

[0230] Step 1776-3, input the first training partition image pair sequence of the current training record into the first scoring model for ASPECTS scoring to obtain the corresponding first training partition scoring pair sequence;

[0231] Step 1776-4, bringing the first training partition score pair sequence and the first label partition score pair sequence of the current training record into the preset sixth model loss function to calculate and obtain the corresponding sixth loss value;

[0232] Among them, the loss function types of the sixth model loss function include at least L1 loss function, L2 loss function and multi-classification cross entropy loss function;

[0233] Step 1776-5, identifying whether the sixth loss value satisfies the preset sixth loss value range; if the sixth loss value satisfies the sixth loss value range, identifying whether the current training record is the last sixth data record of the sixth training data set, if so, turning to step 1776-6, if not, extracting the next sixth data record of the sixth training data set as the new current training record and returning to step 1776-3; if the sixth loss value does not satisfy the sixth loss value range, performing a round of parameter optimization on the first scoring model in the direction of minimizing the sixth model loss function based on the preset sixth model parameter optimizer, and returning to step 1776-3 at the end of this round of parameter optimization;

[0234] The sixth loss value range is a preset loss value range; the sixth model parameter optimizer includes at least an SGD optimizer and an ADAM optimizer;

[0235] Step 1776-6, perform a round of traversal on all the sixth data records of the sixth evaluation data set; and during the traversal, use the sixth data record currently traversed as the corresponding current evaluation record; and input the first training partition image pair sequence of the current evaluation record into the first scoring model for ASPECTS scoring to obtain the corresponding second training partition scoring pair sequence; and form a corresponding prediction-label sequence pair from the second training partition scoring pair sequence and the first label partition scoring pair sequence of the current evaluation record; and at the end of this round of traversal, form a corresponding prediction-label sequence pair set from all the obtained prediction-label sequence pairs;

[0236] Step 1776-7, according to the conventional multi-classification evaluation method based on precision, recall and F-score, the precision and recall of the set are calculated according to the prediction-label sequence to obtain the corresponding sixth precision and sixth recall; and the corresponding F-score is calculated based on the sixth precision and the sixth recall to obtain the corresponding sixth F-score;

[0237] Step 1776-8, identifying the sixth precision, the sixth recall and the sixth F-score based on the preset sixth precision range, the sixth recall range and the sixth F-score range; if the sixth precision does not satisfy the sixth precision range or the sixth recall does not satisfy the sixth recall range or the sixth F-score does not satisfy the sixth F-score range, returning to step 1776-1 to continue training; if the sixth precision satisfies the sixth precision range and the sixth recall satisfies the sixth recall range and the sixth F-score satisfies the sixth F-score range, stopping training and confirming that the sixth model training is completed;

[0238] Here, the sixth precision range, the sixth recall range, and the sixth F-score range are a set of preset precision, recall, and F-score ranges;

[0239] Step 178, after the first, second, third, fourth, fifth and sixth model training are completed, confirm that the collective model training is completed.

[0240] Step 2, after the collective model training is completed, an axial plain scan brain CT image located at the level of the basal ganglia nuclei of the brain and an axial plain scan brain CT image located at the level above the basal ganglia nuclei of the brain are received and recorded as the corresponding first CT image and second CT image; and the first and second CT images form a corresponding first CT image sequence.

[0241] Step 3, using the first, second, third and fourth segmentation models to perform image segmentation processing on the first CT image sequence to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group respectively, and using the first classification model to classify and identify the first CT image sequence to obtain the corresponding first predicted side group;

[0242] Specifically, the first and second CT images of the first CT image sequence are respectively input into the first segmentation model to perform brain sulcus region segmentation processing to obtain the corresponding first image semantic map A. 11 and the second image semantic map A 12 Forming a corresponding first semantic graph group;

[0243] The first semantic graph group includes the first image semantic graph A 11 and the second image semantic map A 12 ; First image semantic map A 11 Including a plurality of first semantic image pixels a11 , each pixel a in the first semantic map 11 Corresponding to a first semantic type; second image semantic map A 12 Including a plurality of second semantic graph pixels a 12 , each pixel a in the second semantic map 12 corresponds to a first semantic type;

[0244] Step 32: Input the first and second CT images of the first CT image sequence into the second segmentation model to perform segmentation processing on the old infarction area of ​​the cerebral artery to obtain the corresponding first image semantic map A. 21 and the second image semantic map A 22 forming a corresponding second semantic graph group;

[0245] The second semantic graph group includes the first image semantic graph A 21 and the second image semantic map A 22 ; First image semantic map A 21 Including a plurality of first semantic image pixels a 21 , each pixel a in the first semantic map 21 Corresponding to a second semantic type; second image semantic map A 22 Including a plurality of second semantic graph pixels a 22 , each pixel a in the second semantic map 22 corresponds to a second semantic type;

[0246] Step 33: Input the first and second CT images of the first CT image sequence into the third segmentation model to perform high-density arterial sign area segmentation processing on the cerebral blood vessels to obtain the corresponding first image semantic map A. 31 and the second image semantic map A 32 Form a corresponding third semantic graph group;

[0247] The third semantic graph group includes the first image semantic graph A 31 and the second image semantic map A 32 ; First image semantic map A 31 Including a plurality of first semantic image pixels a 31 , each pixel a in the first semantic map 31 Corresponding to a third semantic type; second image semantic map A 32 Including a plurality of second semantic graph pixels a 32 , each pixel a in the second semantic map 32 corresponds to a third semantic type;

[0248] Step 34, input the first CT image sequence into the fourth segmentation model to perform left and right brain ASPECTS partitioning and ventricular region segmentation processing to obtain a corresponding semantic map sequence; and use the first and second semantic maps in the semantic map sequence obtained this time as the corresponding first image semantic map A in turn. 41 and the second image semantic map A 42 Form a corresponding fourth semantic graph group;

[0249] The fourth semantic graph group includes the first image semantic graph A 41 and the second image semantic map A 42 ; First image semantic map A 41 Including a plurality of first semantic image pixels a 41 , each pixel a in the first semantic map 41 Corresponding to a fourth semantic type; second image semantic map A 42 Including a plurality of second semantic graph pixels a 42 , each pixel a in the second semantic map 42 corresponds to a 42nd semantic type;

[0250] Step 35, inputting the first and second CT images of the first CT image sequence into the first classification model respectively, and using the first classification model to classify and identify the brain side information of brain tissue cytotoxic edema generated on the first and second CT images to obtain the corresponding first image predicted side and the second image predicted side to form a corresponding first predicted side group;

[0251] The first prediction side group includes a first image prediction side and a second image prediction side; both the first and second image prediction sides include empty, left, right and double sides.

[0252] Step 4, performing a comprehensive evaluation of the side information based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence to obtain a corresponding first evaluated side;

[0253] Among them, the first assessment side includes empty, left, right and bilateral;

[0254] Specifically, step 4-1: transforming the first image semantic map A of the third semantic map group into 31 In the figure, each pixel point a in the first semantic image corresponding to the semantics of the high-density arterial sign area 31 Recorded as corresponding first-category points; and clustering all first-category points based on a preset point clustering algorithm to obtain multiple first-category point sets; and marking corresponding high-density arterial sign areas on the first CT image of the first CT image sequence based on each first-category point set to obtain a corresponding first marked area;

[0255] Among them, the point clustering algorithm includes at least K-means clustering algorithm, DBSCAN clustering algorithm, and OPTICS clustering algorithm;

[0256] Step 4-2, and convert the second image semantic map A of the third semantic map group 32 In the figure, each pixel point a in the second semantic map corresponding to the semantics of the high-density arterial sign area 32 Recorded as corresponding second-category points; and clustering all the second-category points based on the point clustering algorithm to obtain multiple second-category point sets; and marking the corresponding high-density arterial sign area on the second CT image of the first CT image sequence based on each second-category point set to obtain the corresponding second marked area;

[0257] Step 4-3, and convert the first image semantic map A of the fourth semantic map group 41 In the figure, all pixels a in the first semantic graph corresponding to each type of partition semantics of the fourteen types of left and right brain ASPECTS partition semantics 41 Clustering into one category to form a corresponding third category point set; and based on the obtained fourteen third category point sets, corresponding left and right brain ASPECTS partition labels are performed on the first CT image to obtain corresponding fourteen first labeled partitions; and seven left brain partitions in the fourteen first labeled partitions form a corresponding first left brain partition set, and seven right brain partitions form a corresponding first right brain partition set;

[0258] Step 4-4, and convert the second image semantic map A of the fourth semantic map group 42 In the figure, all the pixels a in the second semantic map corresponding to the semantics of each type of the six types of left and right brain ASPECTS semantics 42 Cluster into one category to form a corresponding fourth category point set; and based on the obtained six fourth category point sets, corresponding left and right brain ASPECTS partition marks are performed on the second CT image to obtain corresponding six second labeled partitions; and three left brain partitions in the six second labeled partitions form a corresponding second left brain partition set, and three right brain partitions form a corresponding second right brain partition set; and the first and second left brain partition sets are merged to obtain a corresponding third left brain partition set, and the first and second right brain partition sets are merged to obtain a corresponding third right brain partition set;

[0259] Step 4-5, and setting a corresponding first side-specific evaluation result based on the regional intersection relationship between the third left brain partition set and the third right brain partition set and all the first and second marked areas; and using the third average precision rate obtained by the third segmentation model in the most recent model training as the first confidence level corresponding to the first side-specific evaluation result;

[0260] Specifically comprising: step 4-5-1, setting a corresponding first sideness evaluation result based on the region intersection relationship between the third left brain partition set and the third right brain partition set and all the first and second marked regions;

[0261] Among them, the first side assessment results include empty, left side, right side, and bilateral;

[0262] Specifically comprising: step 4-5-1-1, if all left brain partitions in the third left brain partition set have no intersection with all first and second marked areas, and all right brain partitions in the third right brain partition set have no intersection with all first and second marked areas, then setting the corresponding first sideness evaluation result to empty;

[0263] Step 4-5-1-2, if at least one left brain partition in the third left brain partition set has an intersection with at least one first or second marked area, and at least one right brain partition in the third right brain partition set has an intersection with at least one first or second marked area, then set the corresponding first sideness assessment result to bilateral;

[0264] Step 4-5-1-3, if at least one left brain partition in the third left brain partition set has an intersection with at least one first or second marked area, and all right brain partitions in the third right brain partition set have no intersection with all first and second marked areas, then set the corresponding first sideness assessment result to left;

[0265] Step 4-5-1-4, if at least one right brain partition in the third right brain partition set has an intersection with at least one first or second marked area, and all left brain partitions in the third left brain partition set have no intersection with all first and second marked areas, then set the corresponding first side evaluation result to the right side;

[0266] Step 4-5-2, and use the third average precision obtained by the third segmentation model in the most recent model training as the first confidence level corresponding to the first identification evaluation result;

[0267] Step 4-6, and identify the first image predicted side and the second image predicted side of the first predicted side group; if both the first and second image predicted sides are empty, set the corresponding second side evaluation result to empty; if one of the first and second image predicted sides is the left side and the other is the left side or empty, set the corresponding second side evaluation result to the left side; if one of the first and second image predicted sides is the right side and the other is the right side or empty, set the corresponding second side evaluation result to the right side; if one of the first and second image predicted sides is the left side and the other is the right side, set the corresponding second side evaluation result to the bilateral side; and use the fifth precision rate obtained by the first classification model in the most recent model training as the second confidence corresponding to the second side evaluation result;

[0268] Among them, the second side assessment results include empty, left side, right side and bilateral;

[0269] Step 4-7, and identify the switch state of the preset manual side evaluation switch; if the switch state of the manual side evaluation switch is in the open state, set the corresponding third side evaluation result to empty, and set the third confidence level corresponding to the third side evaluation result to 0; if the switch state of the manual side evaluation switch is in the closed state, send the first CT image sequence to the preset first manual side evaluation interface, and receive the third side evaluation result and the corresponding third confidence level sent back by the first manual side evaluation interface;

[0270] The switch state of the manual side evaluation switch includes an open state and a closed state;

[0271] Step 4-8, identifying whether the first, second and third side-identity evaluation results are all the same; if they are all the same, taking any one of the first, second and third side-identity evaluation results as the corresponding current evaluation side and going to step 4-16; if they are not all the same, going to step 4-9;

[0272] Step 4-9, identifying whether the third confidence is 0; if the third confidence is 0, go to step 4-10; if the third confidence is not 0, go to step 4-11;

[0273] Step 4-10, identifying whether one of the first and second side-by-side evaluation results is empty; if so, taking the other side-by-side evaluation result that is not empty as the corresponding current evaluation side and going to step 4-16; if not, taking the side-by-side evaluation result with the highest confidence between the first and second side-by-side evaluation results as the corresponding current evaluation side and going to step 4-16;

[0274] Step 4-11, identifying the preset first evaluation mode; if the first evaluation mode is the first mode, go to step 4-12; if the first evaluation mode is the second mode, go to step 4-13; if the first evaluation mode is the third mode, go to step 4-14; if the first evaluation mode is the fourth mode, go to step 4-15;

[0275] Wherein, the first evaluation mode includes a first mode, a second mode, a third mode and a fourth mode;

[0276] Step 4-12, taking the side evaluation result with the highest confidence among the first, second and third side evaluation results as the corresponding current evaluation side and going to step 4-16;

[0277] Step 4-13, set four ballot box sets initialized to be empty for the four side types of empty, left, right and both sides, and record them as the corresponding first, second, third and fourth ballot box sets; and when the first, second or third side evaluation result matches one of the four side types, add the current side evaluation result to the first, second, third or fourth ballot box set corresponding to the current side type; and use the total number of the added side evaluation results in the first, second, third and fourth ballot box sets finally obtained as the corresponding number of votes in the first, second, third and fourth ballot boxes; and add the first, second, third and fourth ballot box sets to the corresponding number of votes. and the maximum value of the votes in the fourth ballot box as the corresponding maximum votes; and identify whether the number of ballot box sets corresponding to the maximum votes is unique; if the number of ballot box sets corresponding to the maximum votes is unique, then the side type corresponding to the maximum votes among the four side types of empty, left, right and double-sided is taken as the corresponding current evaluation side and go to step 4-16; if the number of ballot box sets corresponding to the maximum votes is not unique, then the side evaluation result with the highest confidence among all side evaluation results in all ballot box sets corresponding to the maximum votes is taken as the corresponding current evaluation side and go to step 4-16;

[0278] Step 4-14, taking the difference of the first, second or third confidence minus a preset baseline confidence threshold as the corresponding first, second or third differential confidence; and setting four differential confidence sets initialized to empty for the four side identification types of empty, left, right and bilateral as the corresponding first, second, third and fourth differential confidence sets; and when the first, second or third side identification evaluation result matches one of the four side identification types, adding the first, second or third differential confidence corresponding to the current side identification result to the first, second, third or fourth differential confidence set corresponding to the current side identification type; and summing up all differential confidences in the first, second, third or fourth differential confidence set finally obtained Calculate the corresponding sum of the first, second, third or fourth sets; and take the maximum value among the sums of the first, second, third and fourth sets as the corresponding maximum sum; and identify whether the number of differential confidence sets corresponding to the maximum sum is unique; if the number of differential confidence sets corresponding to the maximum sum is unique, then take the side type corresponding to the maximum sum among the four side types of empty, left, right and bilateral as the corresponding current evaluation side and go to step 4-16; if the number of differential confidence sets corresponding to the maximum sum is not unique, then take the side evaluation result with the largest confidence among all side evaluation results in all differential confidence sets corresponding to the maximum sum as the corresponding current evaluation side and go to step 4-16;

[0279] Step 4-15, taking the difference of the first, second or third confidence minus a baseline confidence threshold as the corresponding fourth, fifth or sixth differential confidence; and setting four differential confidence sets initialized to empty for the four side identification types of empty, left, right and bilateral as the corresponding fifth, sixth, seventh and eighth differential confidence sets; and when the first, second or third side identification evaluation result matches one of the four side identification types, adding the fourth, fifth or sixth differential confidence corresponding to the current side identification result to the fifth, sixth, seventh or eighth differential confidence set corresponding to the current side identification type; and averaging all differential confidences in the fifth, sixth, seventh or eighth differential confidence set finally obtained. Calculate the corresponding first, second, third or fourth set mean; and take the maximum value among the first, second, third and fourth set means as the corresponding maximum mean; and identify whether the number of differential confidence sets corresponding to the maximum mean is unique; if the number of differential confidence sets corresponding to the maximum mean is unique, then take the side type corresponding to the maximum mean among the four side types of empty, left, right and bilateral as the corresponding current evaluation side and go to step 4-16; if the number of differential confidence sets corresponding to the maximum mean is not unique, then take the side evaluation result with the largest confidence among all side evaluation results in all differential confidence sets corresponding to the maximum mean as the corresponding current evaluation side and go to step 4-16;

[0280] Step 4-16, outputting the current evaluation aspect finally obtained as the corresponding first evaluation aspect.

[0281] Step 5, extracting left and right brain ASPECTS partition images from the first CT image sequence to obtain a corresponding first partition image set; filtering each partition image in the first partition image set based on a median filter method; and eliminating the sulcus region and the old infarction region of the cerebral artery from each partition image in the first partition image set based on the first and second semantic graph groups after filtering to obtain a corresponding second partition image set; and extracting and sorting left and right brain CT image pairs from the ASPECTS partition based on the second partition image set to obtain a corresponding first partition image pair sequence;

[0282] Specifically, it includes: step 51, performing left and right brain ASPECTS partition image extraction processing on the first CT image sequence to obtain a corresponding first partition image set;

[0283] The first partition image set includes twenty first partition images, and the first to twentieth first partition images are the corresponding left / right brain C area image, left / right brain L area image, left / right brain IC area image, left / right brain I area image, left / right brain M1 area image, left / right brain M2 area image, left / right brain M3 area image, left / right brain M4 area image, left / right brain M5 area image and left / right brain M6 area image respectively;

[0284] Specifically, it includes: step 511, extracting the left brain C area image, the left brain L area image, the left brain IC area image, the left brain I area image, the left brain M1 area image, the left brain M2 area image, the left brain M3 area image, the left brain M4 area image, the left brain M5 area image and the left brain M6 area image corresponding to the ten left brain partitions of the third left brain partition set on the first and second CT images of the first CT image sequence as the corresponding ten first partition images;

[0285] Step 512: extract the right brain C region image, the right brain L region image, the right brain IC region image, the right brain I region image, the right brain M1 region image, the right brain M2 region image, the right brain M3 region image, the right brain M4 region image, the right brain M5 region image and the right brain M6 region image corresponding to the ten right brain regions of the third right brain region set on the first and second CT images of the first CT image sequence as the corresponding ten first region images;

[0286] Step 513, the twenty first subarea images obtained form a corresponding first subarea image set;

[0287] Step 52, filtering each subarea image in the first subarea image set based on a median filtering method;

[0288] Step 53, after filtering is completed, based on the first and second semantic graph groups, each sub-area image in the first sub-area image set is processed to eliminate the sulcus area and the old infarction area of ​​the cerebral artery to obtain a corresponding second sub-area image set;

[0289] The second partition image set includes twenty second partition images, and the first to twentieth second partition images are the corresponding left / right brain C area image, left / right brain L area image, left / right brain IC area image, left / right brain I area image, left / right brain M1 area image, left / right brain M2 area image, left / right brain M3 area image, left / right brain M4 area image, left / right brain M5 area image and left / right brain M6 area image respectively;

[0290] Specifically, step 531 includes: 11 In the first semantic map, the pixel points a corresponding to the semantics of the sulcus region 11Recorded as the corresponding fifth-category points; and clustering all the fifth-category points based on the point clustering algorithm to obtain multiple fifth-category point sets; and marking the corresponding sulcus area on the first CT image based on each fifth-category point set to obtain the corresponding third marking area;

[0291] Step 532: the second image semantic graph A of the first semantic graph group 12 In the second semantic map, the pixel points a corresponding to the semantics of the sulcus region 12 Recorded as the corresponding sixth-category points; and clustering all the sixth-category points based on the point clustering algorithm to obtain multiple sixth-category point sets; and marking the corresponding sulcus area on the second CT image based on each sixth-category point set to obtain the corresponding fourth marking area;

[0292] Step 533: the first image semantic graph A of the second semantic graph group 21 In the first semantic map, the pixel points a corresponding to the semantics of the old infarction area 21 Recorded as the corresponding seventh-category points; and clustering all the seventh-category points based on the point clustering algorithm to obtain multiple seventh-category point sets; and marking the corresponding old infarction area on the first CT image based on each seventh-category point set to obtain the corresponding fifth marking area;

[0293] Step 534: the second image semantic graph A of the second semantic graph group 22 In the second semantic map, the pixel points a corresponding to the semantics of the old infarction area 22 Recorded as the corresponding eighth-category points; and clustering all the eighth-category points based on the point clustering algorithm to obtain multiple eighth-category point sets; and marking the corresponding old infarction area on the second CT image based on each eighth-category point set to obtain the corresponding sixth marking area;

[0294] Step 535, each first subarea image in the first subarea image set is sequentially used as the corresponding current subarea image; and the intersection area of ​​the current subarea image with any third, fourth, fifth or sixth marked area is used as the corresponding first intersection area; and the total number of the obtained first intersection areas is counted to obtain the corresponding first intersection total number; and the first intersection total number is identified; if the first intersection total number is 0, the current subarea image is used as a corresponding second subarea image; if the first intersection total number is greater than 0, the local area images covered by each first intersection area in the current subarea image are deleted and the deleted current subarea image is used as a corresponding second subarea image;

[0295] Step 536, forming a corresponding second subarea image set from all the obtained second subarea images;

[0296] Step 54, extracting and sorting ASPECTS partition left and right brain CT image pairs based on the second partition image set to obtain a corresponding first partition image pair sequence;

[0297] Among them, the first partition image pair sequence is formed by sequentially sorting ten first partition image pairs; the first to tenth first partition image pairs are the corresponding left and right brain C area image pairs, left and right brain L area image pairs, left and right brain IC area image pairs, left and right brain I area image pairs, left and right brain M1 area image pairs, left and right brain M2 area image pairs, left and right brain M3 area image pairs, left and right brain M4 area image pairs, left and right brain M5 area image pairs and left and right brain M6 area image pairs;

[0298] Specifically, it includes: step 541, forming a left and right brain C region image pair from two second partition images in the second partition image set corresponding to the left / right brain C region image as a corresponding first partition image pair;

[0299] Step 542, two second sub-area images corresponding to the left / right brain L area images in the second sub-area image set are combined into a left and right brain L area image pair and recorded as a corresponding first sub-area image pair;

[0300] Step 543, two second sub-area images corresponding to the left / right brain IC area images in the second sub-area image set are combined into a left and right brain IC area image pair and recorded as a corresponding first sub-area image pair;

[0301] Step 544, two second sub-area images corresponding to the left / right brain I area images in the second sub-area image set are combined into a left and right brain I area image pair and recorded as a corresponding first sub-area image pair;

[0302] Step 545, two second sub-area images corresponding to the left / right brain M1 area images in the second sub-area image set are combined into a left and right brain M1 area image pair and recorded as a corresponding first sub-area image pair;

[0303] Step 546, two second sub-area images corresponding to the left / right brain M2 area images in the second sub-area image set are combined into a left and right brain M2 area image pair and recorded as a corresponding first sub-area image pair;

[0304] Step 547, two second sub-area images corresponding to the left / right brain M3 area images in the second sub-area image set are combined into a left and right brain M3 area image pair and recorded as a corresponding first sub-area image pair;

[0305] Step 548, two second sub-area images corresponding to the left / right brain M4 area images in the second sub-area image set are combined into a left and right brain M4 area image pair and recorded as a corresponding first sub-area image pair;

[0306] Step 549, two second sub-area images corresponding to the left / right brain M5 area images in the second sub-area image set are combined into a left and right brain M5 area image pair and recorded as a corresponding first sub-area image pair;

[0307] Step 550, two second partition images corresponding to the left / right brain M6 area images in the second partition image set are combined into a left and right brain M6 area image pair and recorded as a corresponding first partition image pair; and the ten first partition image pairs obtained are sequentially sorted to form a corresponding first partition image pair sequence.

[0308] Step 6: Use the first scoring model to perform ASPECTS scoring on the first partition image pair sequence to obtain the corresponding first partition score pair sequence.

[0309] Here, the first partition score pair sequence output by the first scoring model of the embodiment of the present invention is composed of ten first partition score pairs sorted in sequence; the first partition score pairs in the output sequence, i.e., the first partition score pair sequence, correspond one to one with the first partition image pairs in the input sequence, i.e., the first partition image pair sequence; the first first partition score pair is composed of corresponding left and right brain C distinction values; the second first partition score pair is composed of corresponding left and right brain L distinction values; the third first partition score pair is composed of corresponding left and right brain IC distinction values; the fourth first partition score pair is composed of corresponding left and right brain I distinction values; the fifth first partition score pair is composed of corresponding left and right brain M1 distinction values; the sixth first partition score pair is composed of corresponding left and right brain M2 distinction values. The score pair consists of the corresponding left and right brain M2 distinguishing values; the seventh first partition score pair consists of the corresponding left and right brain M3 distinguishing values; the eighth first partition score pair consists of the corresponding left and right brain M4 distinguishing values; the ninth first partition score pair consists of the corresponding left and right brain M5 distinguishing values; the tenth first partition score pair consists of the corresponding left and right brain M6 distinguishing values; all left / right brain distinguishing values ​​in the first partition score pair sequence are 0 or 1. When any left / right brain distinguishing value is 1, it means that the possibility of ischemic stroke occurring in the corresponding left / right brain partition is relatively high. When any left / right brain distinguishing value is 0, it means that the possibility of ischemic stroke occurring in the corresponding left / right brain partition is relatively low.

[0310] Step 7, based on the first assessment side and the first partition score, the sequence is summarized as a whole and on the affected side to obtain a corresponding first summary report;

[0311] The first summary report includes a first total score and a first affected side type; the first total score is an integer ranging from 0 to 10; the first affected side type includes left side, right side and bilateral side;

[0312] Specifically, it includes: step 71, clustering all left brain partition values ​​with a score of 1 in the first partition score pair sequence into one class to form a corresponding first left brain partition value set; and clustering all right brain partition values ​​with a score of 1 in the first partition score pair sequence into one class to form a corresponding first right brain partition value set;

[0313] Step 72, identifying the first evaluation category;

[0314] Step 73: If the first assessment category is empty, the sum of the scores of the first left brain partition value set is calculated to obtain the corresponding first total score, and the sum of the scores of the first right brain partition value set is calculated to obtain the corresponding second total score; and the first and second total scores are compared. If the first total score is greater than or equal to the second total score, the first total score is used as the corresponding current total score N. c , and set the corresponding first affected side type to the left side. If the first total score is less than the second total score, the second total score is used as the corresponding current total score N c , and set the corresponding first affected side type to the right side;

[0315] Step 74: If the first assessed side is the left side, the sum of the scores of the first left brain partition score set is calculated and the calculated result is used as the corresponding current total score N c , and set the corresponding first affected side type to the left side;

[0316] Step 75: If the first assessed side is the right side, the sum of the scores of the first right brain partition score set is calculated and the calculated result is used as the corresponding current total score N c , and set the corresponding first affected side type to the right side;

[0317] Step 76, if the first assessment side is bilateral, the sum of the scores of the first left brain partition value set is calculated to obtain a corresponding third total score, and the sum of the scores of the first right brain partition value set is calculated to obtain a corresponding fourth total score; and the third and fourth total scores are compared, and if the third score is greater than or equal to the fourth total score, the third total score is used as the corresponding current total score N c If the third total score is less than the fourth total score, the second total score is used as the corresponding current total score N c ; and set the corresponding first affected side type to bilateral;

[0318] Step 77, and based on the preset ASPECTS score total score N max and the current total score N c Calculate and obtain the corresponding first total score;

[0319] Among them, the first total score = N max -N c ;

[0320] Here, the total score of ASPECTS is N max The default is 10; the current total score is N c and the first total score are both integers between 0 and 10;

[0321] Step 78: A corresponding first summary report is formed by the obtained first total score and the first affected side type.

[0322] Figure 4 The module structure diagram of a processing device for predicting ASPECTS scores based on CT images provided in the second embodiment of the present invention is a terminal device or server that implements the aforementioned method embodiment, and can also be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 4 As shown, the device includes: a model building and training module 201, a data receiving module 202, an image segmentation and classification module 203, a side comprehensive evaluation module 204, a partition image denoising and de-interference module 205, a partition scoring module 206 and an overall summary module 207.

[0323] The model building and training module 201 is used to build an image semantic segmentation model for performing sulcus region segmentation on brain CT images, which is recorded as the corresponding first segmentation model; and to build an image semantic segmentation model for performing cerebral arterial vascular old infarction region segmentation on brain CT images, which is recorded as the corresponding second segmentation model; and to build an image semantic segmentation model for performing cerebral vascular high-density arterial sign region segmentation on brain CT images, which is recorded as the corresponding third segmentation model; and to build an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on brain CT image sequences, which is recorded as the corresponding fourth segmentation model; and to build a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images; and to build a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs; and to collectively train the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on a preset plain scan CT image library.

[0324] The data receiving module 202 is used to receive an axial plain scan brain CT image located at the level of the basal ganglia nuclei and an axial plain scan brain CT image located above the basal ganglia nuclei as the corresponding first CT image and second CT image after the collective model training is completed; and the first and second CT images form a corresponding first CT image sequence.

[0325] The image segmentation and classification module 203 is used to perform image segmentation processing on the first CT image sequence by the first, second, third and fourth segmentation models respectively to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group, and to classify and identify the first CT image sequence by the first classification model to obtain the corresponding first predicted side group; the first predicted side group includes the first image predicted side and the second image predicted side; the first and second image predicted sides both include empty, left, right and bilateral.

[0326] The side comprehensive evaluation module 204 is used to perform a comprehensive evaluation of the side information based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence to obtain the corresponding first evaluated side; the first evaluated side includes empty, left, right and bilateral.

[0327] The partition image denoising and de-interference module 205 is used to extract the left and right brain ASPECTS partition images of the first CT image sequence to obtain the corresponding first partition image set; and filter each partition image in the first partition image set based on the median filtering method; and after the filtering, eliminate the sulcus area and the old infarction area of ​​the cerebral artery vessels of each partition image in the first partition image set based on the first and second semantic graph groups to obtain the corresponding second partition image set; and extract and sort the ASPECTS partition left and right brain CT image pairs based on the second partition image set to obtain the corresponding first partition image pair sequence.

[0328] The partition scoring module 206 is used to perform ASPECTS scoring on the first partition image pair sequence using a first scoring model to obtain a corresponding first partition score pair sequence.

[0329] The overall summary module 207 is used to summarize the overall score and the affected side of the sequence based on the first assessment side and the first partition score to obtain a corresponding first summary report; the first summary report includes a first total score and a first affected side type; the first total score is an integer between 0 and 10; the first affected side type includes left side, right side and bilateral side.

[0330] An embodiment of the present invention provides a processing device for predicting ASPECTS scores based on CT images, which can execute the method steps in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0331] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the model building and training module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-mentioned module is determined. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in software form.

[0332] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0333] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the above method embodiments are generated. The above computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0334] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device may be a terminal device or a server for implementing the method of the aforementioned embodiment, or may be a terminal device or a server for implementing the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 5 As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303. Various instructions may be stored in the memory 302 to complete various processing functions and implement the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment of the present invention also includes: a power supply 304, a system bus 305 and a communication port 306. The system bus 305 is used to realize the communication connection between components. The above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0335] exist Figure 5The system bus 305 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM), and may also include non-volatile memory (Non-Volatile Memory), such as at least one disk storage.

[0336] The above-mentioned processor can be a general-purpose processor, including a central processing unit CPU, a network processor (Network Processor, NP), a graphics processing unit (Graphics Processing Unit, GPU), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0337] It should be noted that an embodiment of the present invention further provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the method and processing process provided in the above embodiments.

[0338] An embodiment of the present invention further provides a chip for executing instructions, wherein the chip is used to execute the processing steps described in the aforementioned method embodiment.

[0339] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for predicting ASPECTS score based on CT images. As can be seen from the above content, the embodiment of the present invention constructs four image segmentation models (a first segmentation model for brain sulcus area segmentation, a second segmentation model for brain arterial old infarction area segmentation, a third segmentation model for brain vascular high-density arterial sign area segmentation, and a fourth segmentation model for left and right brain ASPECTS partition and ventricular area segmentation), a first classification model for classifying brain lateral information of brain tissue cytotoxic edema, and a first scoring model for ASPECTS scoring of ten groups of left and right brain partition CT images; and after all models are trained, a CT image sequence is composed of an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain and an axial brain plain scan CT image located above the level of the basal ganglia nuclei of the brain; and the first / second / third / fourth segmentation models are used to perform image classification on the CT image sequence respectively. The corresponding first / second / third / fourth semantic map groups are obtained by segmentation, and the first classification model is used to classify and identify the CT image sequence to obtain the corresponding predicted side group; and the side information is comprehensively evaluated based on the third / fourth semantic map groups, the predicted side group and the CT image sequence to obtain the corresponding assessed side; and the left and right brain ASPECTS partition images of the CT image sequence are extracted to obtain the corresponding partition image set; and each partition image in the partition image set is filtered based on the median filtering method, and after the filtering, the interference area (brain sulcus area and cerebral artery old infarction area) of each partition image is eliminated based on the first / second semantic map group; and the first scoring model is used to perform ASPECTS scoring on the partition image pair sequence to obtain the corresponding partition scoring pair sequence; finally, based on the assessed side and the partition scoring pair sequence, the overall score and the affected side summary are summarized to obtain the corresponding summary report. It can be seen that the embodiment of the present invention uses the fourth segmentation model to improve the boundary accuracy of the partition image, uses the median filter to reduce the noise of the partition image, uses the first / second segmentation model to eliminate two types of interference factor images in the partition image (brain sulcus area image and cerebral arterial old infarction area image), uses the third segmentation model + the first classification model and a side information comprehensive evaluation mechanism to improve the positioning accuracy of the affected side, and uses the batch processing mechanism of the first scoring model to improve the partition scoring efficiency; through the embodiment of the present invention, not only the overall prediction efficiency is further improved, but also the boundary accuracy of the partition image is improved, the noise and interference of the partition image are reduced, the positioning accuracy of the affected side is improved, and the prediction accuracy is improved.

[0340] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0341] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0342] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting ASPECTS scores based on CT images, characterized in that: The method comprises: Construct an image semantic segmentation model for performing sulcus region segmentation on brain CT images, which is recorded as the corresponding first segmentation model; construct an image semantic segmentation model for performing cerebral arterial vascular old infarction region segmentation on brain CT images, which is recorded as the corresponding second segmentation model; construct an image semantic segmentation model for performing cerebral vascular high-density arterial sign region segmentation on brain CT images, which is recorded as the corresponding third segmentation model; construct an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on brain CT image sequences, which is recorded as the corresponding fourth segmentation model; construct a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images; construct a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs; and collectively train the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on a preset plain scan CT image library; After the collective model training is completed, an axial plain scan brain CT image located at the level of the basal ganglia nuclei and an axial plain scan brain CT image located at a level above the basal ganglia nuclei are received and recorded as corresponding first CT images and second CT images; and the first and second CT images form a corresponding first CT image sequence; The first, second, third and fourth segmentation models respectively perform image segmentation processing on the first CT image sequence to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group, and the first classification model classifies and identifies the first CT image sequence to obtain the corresponding first predicted side group; the first predicted side group includes the first image predicted side and the second image predicted side; the first and second image predicted sides both include empty, left, right and bilateral; Based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence, a comprehensive evaluation of the side information is performed to obtain a corresponding first evaluated side; the first evaluated side includes empty, left, right and bilateral; The first CT image sequence is subjected to left and right brain ASPECTS partition image extraction processing to obtain a corresponding first partition image set; and each partition image in the first partition image set is filtered based on a median filtering method; and after filtering, each partition image in the first partition image set is subjected to brain sulcus area and cerebral arterial vessel old infarction area elimination processing based on the first and second semantic graph groups to obtain a corresponding second partition image set; and based on the second partition image set, ASPECTS partition left and right brain CT image pairs are extracted and sorted to obtain a corresponding first partition image pair sequence; Performing ASPECTS scoring on the first partition image pair sequence by the first scoring model to obtain a corresponding first partition score pair sequence; Based on the first assessment side and the first partition score, the sequence is summarized in overall score and affected side to obtain a corresponding first summary report; the first summary report includes a first total score and a first affected side type; the first total score is an integer between 0 and 10; the first affected side type includes left side, right side and bilateral side.

2. The method for predicting ASPECTS scores based on CT images according to claim 1, characterized in that: The high-density arterial sign of brain blood vessels and the cytotoxic edema of brain tissue are each a typical symptom of ischemic stroke; The ASPECTS partition is composed of ten types of brain tissue partitions, namely C partition, L partition, IC partition, I partition, M1 partition, M2 partition, M3 partition, M4 partition, M5 partition and M6 partition; the C partition is the caudate nucleus area of ​​the brain; the L partition is the lenticular nucleus area of ​​the brain; the IC partition is the internal capsule area of ​​the brain; the I partition is the insular cortex area of ​​the brain; the M1 partition is the anterior cortical area of ​​the middle cerebral artery; the M2 partition is the lateral cortical area of ​​the insular lobe of the middle cerebral artery; the M3 partition is the posterior cortical area of ​​the middle cerebral artery; the M4 partition is the middle cerebral artery cortical area above the M1 partition; the M5 partition is the middle cerebral artery cortical area above the M2 partition; the M6 ​​partition is the middle cerebral artery cortical area above the M3 partition; the various types of brain tissue partitions of the ASPECTS partition are symmetrical between the left and right brains.

3. The method for predicting ASPECTS scores based on CT images according to claim 2, characterized in that: The first segmentation model is implemented based on the U-Net image segmentation network; the first segmentation model is used to perform sulcus region segmentation processing on the current CT image input by the model to obtain the corresponding first semantic map; the current CT image is an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei of the brain or an axial plain scan CT image of the brain located at a level above the basal ganglia nuclei of the brain; the first semantic map includes a plurality of first pixel points; each of the first pixel points corresponds to a first semantic type; the first semantic type includes sulcus region semantics and background semantics; The second segmentation model is implemented based on the U-Net image segmentation network; the second segmentation model is used to perform brain artery old infarction area segmentation processing on the current CT image input by the model to obtain the corresponding second semantic map; the current CT image is an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain or an axial brain plain scan CT image located at a level above the basal ganglia nuclei of the brain; the second semantic map includes a plurality of second pixel points; each of the second pixel points corresponds to a second semantic type; the second semantic type includes old infarction area semantics and background semantics; The third segmentation model is implemented based on the U-Net image segmentation network; the third segmentation model is used to perform brain vascular high-density arterial sign area segmentation processing on the current CT image input by the model to obtain a corresponding third semantic map; the current CT image is an axial brain plain scan CT image located at the level of the basal ganglia nuclei of the brain or an axial brain plain scan CT image located at a level above the basal ganglia nuclei of the brain; the third semantic map includes a plurality of third pixel points; each of the third pixel points corresponds to a third semantic type; the third semantic type includes high-density arterial sign area semantics and background semantics; The fourth segmentation model is implemented based on two U-Net image segmentation networks, which are recorded as the corresponding first U-Net network and the second U-Net network; the fourth segmentation model is used to perform left and right brain ASPECTS partitioning and ventricular region segmentation processing on the current CT image sequence input by the model to obtain the corresponding fourth semantic map sequence, specifically: the fourth segmentation model is used to use the first CT image in the current CT image sequence input by the model as the corresponding first network input image, and the second CT image in the current CT image sequence as the corresponding second network input image; and the first U-Net network performs fourteen categories of left and right brain ASPECTS partitioning and ventricular region segmentation processing on the first network input image to obtain the corresponding fourth-first semantic map; and the second U-Net network performs six categories of left and right brain ASPECTS partitioning and ventricular region segmentation processing on the second network input image to obtain the corresponding fourth-second semantic map; and the obtained fourth-first and fourth-second semantic maps are sequentially sorted to form the corresponding fourth semantic map sequence output; The current CT image sequence is formed by sorting two CT images, wherein the first CT image is an axial plain scan CT image of the brain located at the level of the basal ganglia nuclei of the brain, and the second CT image is an axial plain scan CT image of the brain located at a level above the basal ganglia nuclei of the brain; The fourth-first semantic map includes a plurality of fourth-first pixels; each of the fourth-first pixels corresponds to a fourth-first semantic type; the fourth-first semantic type includes fourteen categories of left and right brain ASPECTS partition semantics, ventricular region semantics and background semantics; the fourteen categories of left and right brain ASPECTS partition semantics include left / right brain C partition semantics, left / right brain L partition semantics, left / right brain IC partition semantics, left / right brain I partition semantics, left / right brain M1 partition semantics, left / right brain M2 partition semantics, and left / right brain M3 partition semantics; The fourth second semantic map includes a plurality of fourth second pixels; each of the fourth second pixels corresponds to a fourth second semantic type; the fourth second semantic type includes six categories of left and right brain ASPECTS partition semantics, ventricular region semantics and background semantics; the six categories of left and right brain ASPECTS partition semantics include left / right brain M4 region semantics, left / right brain M5 region semantics, and left / right brain M6 region semantics; The first classification model includes a first feature extraction module and a first classification module; the first feature extraction module is implemented based on a ResNet network; the first classification module is implemented based on an MLP network; the first classification model is used to classify and identify the brain side information that produces brain tissue cytotoxic edema on the current CT image input by the model to obtain the corresponding predicted side, specifically: the first classification model is used to input the current CT image input by the model into the first feature extraction module for feature extraction processing to obtain the corresponding feature tensor; and input the feature tensor into the first classification module for classification prediction processing to obtain the corresponding prediction type; the prediction types include empty, left, right and bilateral; The first scoring model is composed of ten parallel partition scoring models and one partition scoring merging module; each of the partition scoring models is composed of a left and right brain image sorting module, a left brain feature extraction module, a right brain feature extraction module, a left and right brain feature difference module, a left brain feature fusion module, a right brain feature fusion module, a left brain score prediction module, a right brain score prediction module and a left and right brain score merging module; the left and right brain score prediction modules are both implemented based on a binary classification prediction model; The first scoring model is used to perform ASPECTS scoring on the current partition image pair sequence input by the model to obtain a corresponding partition score pair sequence, specifically: the first scoring model is used to input each partition image pair in the current partition image pair sequence input by the model into the corresponding partition scoring model, and each partition scoring model performs left and right brain ASPECTS scoring according to the corresponding partition image pair to obtain a corresponding partition score pair and send it to the partition scoring merging module; and the partition scoring merging module sequentially sorts all the obtained partition score pairs to form the corresponding partition score pair sequence; Among them, each of the partition scoring models performs left and right brain ASPECTS scoring according to the corresponding partition image pair to obtain the corresponding partition scoring pair and send it to the partition scoring merging module, specifically: each of the partition scoring models inputs the corresponding partition image pair into the left and right brain image sorting module, and the left and right brain image sorting module extracts the corresponding left brain partition image and right brain partition image from the partition image pair and sends them to the corresponding left brain feature extraction module and the right brain feature extraction module; and the left brain feature extraction module performs imaging genomics feature extraction on the left brain partition image The corresponding left brain imaging omics features are extracted and sent to the left and right brain feature difference module and the left brain feature fusion module; and the right brain feature extraction module extracts the imaging omics features of the right brain partition image to obtain the corresponding right brain imaging omics features and sends them to the left and right brain feature difference module and the right brain feature fusion module; and the left and right brain feature difference module records the differential features of the left brain imaging omics features minus the right brain imaging omics features and the differential features of the right brain imaging omics features minus the left brain imaging omics features as the corresponding left-right brain imaging omics differential features and right-left brain images. The left-right brain imaging omics differential features are sent to the corresponding left brain feature fusion module, and the right-left brain imaging omics differential features are sent to the right brain feature fusion module; the left brain feature fusion module performs feature splicing processing on the left brain imaging omics features and the left-right brain imaging omics differential features to obtain the corresponding left brain splicing features and sends them to the left brain score prediction module; the right brain feature fusion module performs feature splicing processing on the right brain imaging omics features and the right-left brain imaging omics differential features to obtain the corresponding right brain splicing features and sends them to the left brain score prediction module. The right brain score prediction module sends the result; the left brain score prediction module performs binarized ASPECTS score prediction processing according to the left brain splicing features to obtain the corresponding left brain partition score and sends it to the left and right brain score merging module; the right brain score prediction module performs binarized ASPECTS score prediction processing according to the right brain splicing features to obtain the corresponding right brain partition score and sends it to the left and right brain score merging module; the left and right brain score merging module combines the left brain partition score and the right brain partition score into the corresponding partition score pair and sends it to the partition score merging module; The current partition image pair sequence is formed by sequentially sorting ten partition image pairs; each partition image pair corresponds to a type of partition in the ASPECTS partition; the partition image pairs correspond one-to-one to the partition scoring models in the first scoring model; each partition image is composed of the corresponding left brain partition image and the right brain partition image; The partition score pair sequence is composed of ten partition score pairs arranged in sequence; the partition score pairs correspond one-to-one to the partition image pairs; each partition score pair is composed of the corresponding left brain partition value and the right brain partition value; the left and right brain partition values ​​are 0 or 1.

4. The method for predicting ASPECTS scores based on CT images according to claim 3, characterized in that: The plain scan CT image library includes a plurality of first plain scan CT image groups; each of the first plain scan CT image groups corresponds to two plain scan CT acquisition images of an image acquisition object; the first plain scan CT image group includes a first acquisition object type, a first CT acquisition image, and a second CT acquisition image; the first acquisition object type is used to mark the characteristic type of the corresponding image acquisition object, including at least a healthy type without brain disease, a type with a history of old infarction of cerebral arterial vessels but no symptoms of ischemic stroke, a type of class A ischemic stroke with no history of old infarction of cerebral arterial vessels but with symptoms of ischemic stroke, and a type of class B ischemic stroke with a history of old infarction of cerebral arterial vessels and symptoms of ischemic stroke; the symptoms of ischemic stroke include the high-density arterial signs of the cerebral vessels and the brain group one or all of the following: tissue cell toxic edema; the first CT acquisition image is an axial brain plain scan CT image of the corresponding image acquisition object located at the level of the basal ganglia nuclei of the brain; the second CT acquisition image is an axial brain plain scan CT image of the corresponding image acquisition object located at a level above the basal ganglia nuclei of the brain; the acquisition method of the first and second CT acquisition images is: an axial brain CT plain scan is performed on the image acquisition object to obtain a corresponding plain scan CT image sequence of the time, and a CT image located at the level of the basal ganglia nuclei of the brain is extracted from the plain scan CT image sequence of the time as the corresponding first CT acquisition image, and a CT image located at a level above the basal ganglia nuclei of the brain is extracted from the plain scan CT image sequence of the time as the corresponding second CT acquisition image; The first semantic graph group includes a first image semantic graph A 11 and the second image semantic map A 12 ; The first image semantic map A 11 Including a plurality of first semantic image pixels a 11 , each pixel point a in the first semantic map 11 Corresponding to the first semantic type; the second image semantic map A 12 Including a plurality of second semantic graph pixels a 12 , each pixel a of the second semantic map 12 corresponding to one of the first semantic types; The second semantic graph group includes a first image semantic graph A 21 and the second image semantic map A 22 ; The first image semantic map A 21 Including a plurality of first semantic image pixels a 21 , each pixel point a in the first semantic map 21 Corresponding to the second semantic type; the second image semantic map A 22 Including a plurality of second semantic graph pixels a 22 , each pixel a of the second semantic map 22 corresponding to one of the second semantic types; The third semantic graph group includes the first image semantic graph A 31 and the second image semantic map A 32 ; The first image semantic map A 31 Including a plurality of first semantic image pixels a 31 , each pixel point a in the first semantic map 31 Corresponding to the third semantic type; the second image semantic map A 32 Including a plurality of second semantic graph pixels a 32 , each pixel a of the second semantic map 32 corresponding to one of the third semantic types; The fourth semantic graph group includes the first image semantic graph A 41 and the second image semantic map A 42 ; The first image semantic map A 41 Including a plurality of first semantic image pixels a 41 , each pixel point a in the first semantic map 41 Corresponding to the fourth first semantic type; the second image semantic map A 42 Including a plurality of second semantic graph pixels a 42 , each pixel a of the second semantic map 42 corresponding to the forty-second semantic type; The first partition image set includes twenty first partition images, and the first to twentieth first partition images are the corresponding left / right brain C area image, left / right brain L area image, left / right brain IC area image, left / right brain I area image, left / right brain M1 area image, left / right brain M2 area image, left / right brain M3 area image, left / right brain M4 area image, left / right brain M5 area image and left / right brain M6 area image respectively; The second subarea image set includes twenty second subarea images, and the first to the twentieth second subarea images are the corresponding left / right brain C area image, left / right brain L area image, left / right brain IC area image, left / right brain I area image, left / right brain M1 area image, left / right brain M2 area image, left / right brain M3 area image, left / right brain M4 area image, left / right brain M5 area image and left / right brain M6 area image respectively; The first partition image pair sequence is formed by sequentially sorting ten first partition image pairs; the first to tenth first partition image pairs are respectively corresponding left and right brain C area image pairs, left and right brain L area image pairs, left and right brain IC area image pairs, left and right brain I area image pairs, left and right brain M1 area image pairs, left and right brain M2 area image pairs, left and right brain M3 area image pairs, left and right brain M4 area image pairs, left and right brain M5 area image pairs and left and right brain M6 area image pairs; The first partition score pair sequence is formed by sequentially sorting ten first partition score pairs; the first partition score pairs correspond to the first partition image pairs one by one; the first first partition score pair is composed of corresponding left and right brain C distinction values; the second first partition score pair is composed of corresponding left and right brain L distinction values; the third first partition score pair is composed of corresponding left and right brain IC distinction values; the fourth first partition score pair is composed of corresponding left and right brain I distinction values; the fifth first partition score pair is composed of corresponding left and right brain M1 distinction values; the sixth first partition score pair is composed of corresponding left and right brain M2 distinction values; the seventh The first partition score pair consists of corresponding left and right brain M3 distinguishing values; the eighth first partition score pair consists of corresponding left and right brain M4 distinguishing values; the ninth first partition score pair consists of corresponding left and right brain M5 distinguishing values; the tenth first partition score pair consists of corresponding left and right brain M6 distinguishing values; all left / right brain partition score values ​​in the first partition score pair sequence are 0 or 1, and when any left / right brain partition score is 1, it indicates that the possibility of ischemic stroke occurring in the corresponding left / right brain partition is relatively high, and when any left / right brain partition score is 0, it indicates that the possibility of ischemic stroke occurring in the corresponding left / right brain partition is relatively low.

5. The method for predicting ASPECTS scores based on CT images according to claim 4, characterized in that: The step of comprehensively evaluating the side information based on the third semantic graph group, the fourth semantic graph group, the first predicted side group, and the first CT image sequence to obtain the corresponding first evaluated side specifically includes: Step 501: convert the first image semantic graph A of the third semantic graph group into 31 In the above, each pixel point a in the first semantic map corresponding to the semantics of the high-density arterial sign region 31 as corresponding first-category points; and clustering all the first-category points based on a preset point clustering algorithm to obtain a plurality of first-category point sets; and marking corresponding high-density arterial sign areas on the first CT image of the first CT image sequence based on each of the first-category point sets to obtain corresponding first marked areas; Wherein, the point clustering algorithm at least includes K-means clustering algorithm, DBSCAN clustering algorithm, and OPTI CS clustering algorithm; Step 502: convert the second image semantic graph A of the third semantic graph group into 32 In the above, each pixel point a of the second semantic map corresponding to the semantics of the high-density arterial sign region 32 as corresponding second-category points; and clustering all the second-category points based on the point clustering algorithm to obtain a plurality of second-category point sets; and marking corresponding high-density arterial sign areas on the second CT image of the first CT image sequence based on each of the second-category point sets to obtain corresponding second marked areas; Step 503: convert the first image semantic graph A of the fourth semantic graph group into 41 In the above, all the first semantic map pixels a corresponding to each type of partition semantics of the fourteen types of left and right brain ASPECTS partition semantics 41 The fourteen third-class point sets are clustered into one class to form a corresponding third-class point set; and based on the obtained fourteen third-class point sets, corresponding left and right brain ASPECTS partition labels are performed on the first CT image to obtain corresponding fourteen first-labeled partitions; and seven left brain partitions in the fourteen first-labeled partitions form a corresponding first left brain partition set, and seven right brain partitions form a corresponding first right brain partition set; Step 504: convert the second image semantic graph A of the fourth semantic graph group into 42 In the above, all the pixel points a of the second semantic map corresponding to each type of partition semantics of the six types of left and right brain ASPECTS partition semantics 42 The six fourth-class point sets are clustered into one class to form a corresponding fourth-class point set; and based on the obtained six fourth-class point sets, corresponding left and right brain ASPECTS partitions are marked on the second CT image to obtain corresponding six second-marked partitions; and three left brain partitions in the six second-marked partitions form a corresponding second left brain partition set, and three right brain partitions form a corresponding second right brain partition set; and the first and second left brain partition sets are merged to obtain a corresponding third left brain partition set, and the first and second right brain partition sets are merged to obtain a corresponding third right brain partition set; Step 505, setting a corresponding first side-specific evaluation result based on the regional intersection relationship between the third left brain partition set and the third right brain partition set and all the first and second marked regions; and taking the third average precision rate of the third segmentation model obtained in the most recent model training as the first confidence level corresponding to the first side-specific evaluation result; Wherein, the first side assessment results include empty, left side, right side and bilateral; Step 506, and identify the first image prediction side and the second image prediction side of the first prediction side group; if both the first and second image prediction sides are empty, set the corresponding second side evaluation result to empty; if one of the first and second image prediction sides is the left side and the other is the left side or empty, set the corresponding second side evaluation result to the left side; if one of the first and second image prediction sides is the right side and the other is the right side or empty, set the corresponding second side evaluation result to the right side; if one of the first and second image prediction sides is the left side and the other is the right side, set the corresponding second side evaluation result to the bilateral side; and use the fifth precision rate obtained by the first classification model in the most recent model training as the second confidence corresponding to the second side evaluation result; Wherein, the second side assessment results include empty, left side, right side and bilateral; Step 507, and identify the switch state of the preset manual side evaluation switch; if the switch state of the manual side evaluation switch is in the open state, set the corresponding third side evaluation result to be empty, and set the third confidence level corresponding to the third side evaluation result to 0; if the switch state of the manual side evaluation switch is in the closed state, send the first CT image sequence to the preset first manual side evaluation interface, and receive the third side evaluation result and the corresponding third confidence level sent back by the first manual side evaluation interface; Wherein, the switch state of the artificial side evaluation switch includes an open state and a closed state; Step 508, identifying whether the first, second and third side-specific evaluation results are all the same; if they are all the same, taking any one of the first, second and third side-specific evaluation results as the corresponding current evaluation side and going to step 516; if they are not all the same, going to step 509; Step 509, identifying whether the third confidence is 0; if the third confidence is 0, go to step 510; if the third confidence is not 0, go to step 511; Step 510, identifying whether one of the first and second side-specific evaluation results is empty; if so, taking the other side-specific evaluation result that is not empty as the corresponding current evaluation side and going to step 516; if not, taking the side-specific evaluation result with the highest confidence between the first and second side-specific evaluation results as the corresponding current evaluation side and going to step 516; Step 511, identifying a preset first evaluation mode; if the first evaluation mode is the first mode, go to step 512; if the first evaluation mode is the second mode, go to step 513; if the first evaluation mode is the third mode, go to step 514; if the first evaluation mode is the fourth mode, go to step 515; Wherein, the first evaluation mode includes a first mode, a second mode, a third mode and a fourth mode; Step 512, taking the side evaluation result with the highest confidence among the first, second and third side evaluation results as the corresponding current evaluation side and going to step 516; Step 513, set four sets of ballot boxes initialized to be empty for the four side types of empty, left, right and both sides, and record them as the corresponding first, second, third and fourth ballot box sets; and when the first, second or third side evaluation result matches one of the four side types, add the current side evaluation result to the first, second, third or fourth ballot box set corresponding to the current side type; and use the total number of the added side evaluation results in the first, second, third and fourth ballot box sets finally obtained as the corresponding number of votes in the first, second, third and fourth ballot boxes; and add the first, second, third and fourth ballot box sets to the first, second, third and fourth ballot box sets. The maximum value among the votes in the four ballot boxes is taken as the corresponding maximum votes; and whether the number of ballot box sets corresponding to the maximum votes is unique is identified; if the number of ballot box sets corresponding to the maximum votes is unique, the side type corresponding to the maximum votes among the four side types of empty, left, right and double sides is taken as the corresponding current evaluation side and the process goes to step 516; if the number of ballot box sets corresponding to the maximum votes is not unique, the side evaluation result with the highest confidence among all the side evaluation results in all the ballot box sets corresponding to the maximum votes is taken as the corresponding current evaluation side and the process goes to step 516; Step 514, taking the difference of the first, second or third confidence minus a preset baseline confidence threshold as the corresponding first, second or third differential confidence; and setting four differential confidence sets initialized to empty for the four side identification types of empty, left, right and bilateral as the corresponding first, second, third and fourth differential confidence sets; and when the first, second or third side identification evaluation result matches one of the four side identification types, adding the first, second or third differential confidence corresponding to the current side identification evaluation result to the first, second, third or fourth differential confidence set corresponding to the current side identification type; and calculating the sum of all differential confidences in the first, second, third or fourth differential confidence set finally obtained Obtain the corresponding sum of the first, second, third or fourth sets; and take the maximum value among the sums of the first, second, third and fourth sets as the corresponding maximum sum; and identify whether the number of differential confidence sets corresponding to the maximum sum is unique; if the number of differential confidence sets corresponding to the maximum sum is unique, then take the side type corresponding to the maximum sum among the four side types of empty, left, right and bilateral as the corresponding current evaluation side and go to step 516; if the number of differential confidence sets corresponding to the maximum sum is not unique, then take the side evaluation result with the largest confidence among all side evaluation results in all differential confidence sets corresponding to the maximum sum as the corresponding current evaluation side and go to step 516; Step 515, taking the difference of the first, second or third confidence minus one of the baseline confidence thresholds as the corresponding fourth, fifth or sixth differential confidence; and setting four differential confidence sets initialized to empty for the four side identification types of empty, left, right and bilateral as the corresponding fifth, sixth, seventh and eighth differential confidence sets; and when the first, second or third side identification evaluation result matches one of the four side identification types, adding the fourth, fifth or sixth differential confidence corresponding to the current side identification result to the fifth, sixth, seventh or eighth differential confidence set corresponding to the current side identification type; and calculating the average of all differential confidences in the fifth, sixth, seventh or eighth differential confidence set finally obtained to the corresponding first, second, third or fourth set mean; and the maximum value among the means of the first, second, third and fourth sets is taken as the corresponding maximum mean; and whether the number of differential confidence sets corresponding to the maximum mean is unique is identified; if the number of differential confidence sets corresponding to the maximum mean is unique, then the side type corresponding to the maximum mean among the four side types of empty, left, right and bilateral is taken as the corresponding current evaluation side and the process goes to step 516; if the number of differential confidence sets corresponding to the maximum mean is not unique, then the side evaluation result with the largest confidence among all side evaluation results in all differential confidence sets corresponding to the maximum mean is taken as the corresponding current evaluation side and the process goes to step 516; Step 516: output the current evaluation aspect finally obtained as the corresponding first evaluation aspect.

6. The method for predicting ASPECTS scores based on CT images according to claim 5, characterized in that: The first sideness evaluation result corresponding to the area intersection relationship between the third left brain partition set and the third right brain partition set and all the first and second marked areas is specifically set as follows: If all left brain partitions in the third left brain partition set have no intersection with all the first and second marked areas, and all right brain partitions in the third right brain partition set have no intersection with all the first and second marked areas, then set the corresponding first sideness evaluation result to empty; If at least one left brain partition in the third set of left brain partitions intersects with at least one of the first or second marked regions, and at least one right brain partition in the third set of right brain partitions intersects with at least one of the first or second marked regions, then setting the corresponding first sideness assessment result to bilateral; If at least one left brain partition in the third set of left brain partitions has an intersection with at least one of the first or second marked regions, and all right brain partitions in the third set of right brain partitions have no intersection with all of the first and second marked regions, then the corresponding first side evaluation result is set to the left side; If at least one right brain partition in the third right brain partition set has an intersection with at least one of the first or second marked areas, and all left brain partitions in the third left brain partition set have no intersection with all of the first and second marked areas, then the corresponding first side assessment result is set to the right side.

7. The method for predicting ASPECTS scores based on CT images according to claim 5, characterized in that: The step of performing left and right brain ASPECTS subarea image extraction processing on the first CT image sequence to obtain a corresponding first subarea image set specifically includes: Extracting the left brain C region image, the left brain L region image, the left brain IC region image, the left brain I region image, the left brain M1 region image, the left brain M2 region image, the left brain M3 region image, the left brain M4 region image, the left brain M5 region image and the left brain M6 region image corresponding to the ten left brain regions of the third left brain region set on the first and second CT images of the first CT image sequence as the corresponding ten first region images; and extracting right brain C region images, right brain L region images, right brain IC region images, right brain I region images, right brain M1 region images, right brain M2 region images, right brain M3 region images, right brain M4 region images, right brain M5 region images, and right brain M6 region images corresponding to the ten right brain partitions of the third right brain partition set on the first and second CT images of the first CT image sequence as the corresponding ten first partition images; The obtained twenty first subarea images form a corresponding first subarea image set.

8. The method for predicting ASPECTS scores based on CT images according to claim 4, characterized in that: After filtering, based on the first and second semantic graph groups, each sub-area image in the first sub-area image set is subjected to a sulcus region and a cerebral artery old infarction region elimination process to obtain a corresponding second sub-area image set, specifically comprising: The first image semantic graph A of the first semantic graph group 11 In the above, each pixel point a in the first semantic map corresponding to the semantics of the sulcus region 11 The points of the fifth category are recorded as corresponding fifth category points; and based on a preset point clustering algorithm, all the fifth category points are clustered to obtain multiple fifth category point sets; and based on each of the fifth category point sets, corresponding sulcus regions are marked on the first CT image to obtain corresponding third marked regions; the point clustering algorithm at least includes a K-means clustering algorithm, a DBSCAN clustering algorithm, and an OPTI CS clustering algorithm; and transform the second image semantic graph A of the first semantic graph group into 12 In the above, each pixel point a in the second semantic map corresponding to the semantics of the sulcus region 12 Recorded as the corresponding sixth-category points; and clustering all the sixth-category points based on the point clustering algorithm to obtain multiple sixth-category point sets; and marking the corresponding sulcus area on the second CT image based on each of the sixth-category point sets to obtain the corresponding fourth marked area; and transform the first image semantic graph A of the second semantic graph group into 21 In the above, each pixel point a in the first semantic map corresponding to the semantics of the old infarction area 21 Recorded as the corresponding seventh type of points; and clustering all the seventh type of points based on the point clustering algorithm to obtain multiple seventh type of point sets; and marking the corresponding old infarction area on the first CT image based on each of the seventh type of point sets to obtain the corresponding fifth marked area; The second image semantic graph A of the second semantic graph group 22 In the above, each pixel point a in the second semantic map corresponding to the semantics of the old infarction area 22 Recorded as the corresponding eighth-category points; and clustering all the eighth-category points based on the point clustering algorithm to obtain multiple eighth-category point sets; and marking the corresponding old infarction area on the second CT image based on each of the eighth-category point sets to obtain the corresponding sixth marking area; And take each of the first sub-area images in the first sub-area image set as the corresponding current sub-area image in turn; and take the intersection area of ​​the current sub-area image with any of the third, fourth, fifth or sixth marked areas as the corresponding first intersection area; and count the total number of the first intersection areas obtained to obtain the corresponding first intersection total number; and identify the first intersection total number; if the first intersection total number is 0, take the current sub-area image as the corresponding second sub-area image; if the first intersection total number is greater than 0, delete the local area images covered by each of the first intersection areas in the current sub-area image and take the deleted current sub-area image as the corresponding second sub-area image; All the obtained second subarea images form a corresponding second subarea image set.

9. The method for predicting ASPECTS scores based on CT images according to claim 4, characterized in that: The step of summarizing the overall scores and the affected side scores of the sequences based on the first assessment side and the first partition score to obtain a corresponding first summary report specifically includes: Clustering all left brain partition values ​​with a score of 1 in the first partition score pair sequence into one class to form a corresponding first left brain partition value set; and clustering all right brain partition values ​​with a score of 1 in the first partition score pair sequence into one class to form a corresponding first right brain partition value set; Identifying the first assessment aspect; If the first assessment aspect is empty, the sum of the scores of the first left brain partition value set is calculated to obtain the corresponding first total score, and the sum of the scores of the first right brain partition value set is calculated to obtain the corresponding second total score; and the first and second total scores are compared, and if the first total score is greater than or equal to the second total score, the first total score is used as the corresponding current total score N c , and set the corresponding first affected side type to the left side, if the first total score is less than the second total score, then use the second total score as the corresponding current total score N c , and setting the corresponding first affected side type to the right side; If the first assessment side is the left side, the sum of the scores of the first left brain partition score set is calculated and the calculation result is used as the corresponding current total score N c , and setting the corresponding first affected side type to the left side; If the first assessment side is the right side, the sum of the scores of the first right brain partition score set is calculated and the calculation result is used as the corresponding current total score N c , and setting the corresponding first affected side type to the right side; If the first assessment side is bilateral, the sum of the scores of the first left brain partition value set is calculated to obtain a corresponding third total score, and the sum of the scores of the first right brain partition value set is calculated to obtain a corresponding fourth total score; and the third and fourth total scores are compared, and if the third score is greater than or equal to the fourth total score, the third total score is used as the corresponding current total score N c If the third total score is less than the fourth total score, the second total score is used as the corresponding current total score N c ; and setting the corresponding first affected side type to bilateral; Based on the preset ASPECTS score N max and the current total score N c Calculate and obtain the corresponding first total score, first total score = N max -N c ; The total score of ASPECTS is N max The default value is 10; the current total score N c and the first total score are both integers ranging from 0 to 10; The first total score and the first affected side type obtained form the corresponding first summary report.

10. A device for executing the processing method for predicting ASPECTS score based on CT images according to any one of claims 1 to 9, characterized in that: The device comprises: a model building and training module, a data receiving module, an image segmentation and classification module, a side comprehensive evaluation module, a partition image denoising and de-interference module, a partition scoring module and an overall summary module; The model construction and training module is used to construct an image semantic segmentation model for performing sulcus region segmentation on brain CT images, which is recorded as the corresponding first segmentation model; and to construct an image semantic segmentation model for performing cerebral arterial vascular old infarction region segmentation on brain CT images, which is recorded as the corresponding second segmentation model; and to construct an image semantic segmentation model for performing cerebral vascular high-density arterial sign region segmentation on brain CT images, which is recorded as the corresponding third segmentation model; and to construct an image semantic segmentation model for performing left and right brain ASPECTS partitioning and ventricular region segmentation on brain CT image sequences, which is recorded as the corresponding fourth segmentation model; and to construct a first classification model for classifying and identifying brain side information that produces brain tissue cytotoxic edema on brain CT images; and to construct a first scoring model for performing ASPECTS scoring on ten groups of ASPECTS partitioned left and right brain CT image pairs; and to collectively train the first, second, third and fourth segmentation models, the first classification model and the first scoring model based on a preset plain scan CT image library; The data receiving module is used to receive an axial plain scan brain CT image located at the level of the basal ganglia nuclei and an axial plain scan brain CT image located at a level above the basal ganglia nuclei as corresponding first CT images and second CT images after the collective model training is completed; and the first and second CT images form a corresponding first CT image sequence; The image segmentation and classification module is used to perform image segmentation processing on the first CT image sequence by the first, second, third and fourth segmentation models respectively to obtain the corresponding first semantic map group, second semantic map group, third semantic map group and fourth semantic map group, and to classify and identify the first CT image sequence by the first classification model to obtain the corresponding first predicted side group; the first predicted side group includes the first image predicted side and the second image predicted side; the first and second image predicted sides both include empty, left, right and bilateral; The side comprehensive evaluation module is used to perform a comprehensive evaluation of side information based on the third semantic map group, the fourth semantic map group, the first predicted side group and the first CT image sequence to obtain a corresponding first evaluation side; the first evaluation side includes empty, left, right and bilateral; The subarea image denoising and de-interference module is used to extract left and right brain ASPECTS subarea images from the first CT image sequence to obtain a corresponding first subarea image set; and filter each subarea image in the first subarea image set based on a median filter; and after filtering, eliminate the sulcus region and the old infarction region of the cerebral artery vessels from each subarea image in the first subarea image set based on the first and second semantic graph groups to obtain a corresponding second subarea image set; and extract and sort ASPECTS subarea left and right brain CT image pairs based on the second subarea image set to obtain a corresponding first subarea image pair sequence; The partition scoring module is used to perform ASPECTS scoring on the first partition image pair sequence using the first scoring model to obtain a corresponding first partition score pair sequence; The overall summary module is used to summarize the overall score and the affected side of the sequence based on the first assessment side and the first partition score to obtain a corresponding first summary report; the first summary report includes a first total score and a first affected side type; the first total score is an integer between 0 and 10; the first affected side type includes left side, right side and bilateral side.

11. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 9; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is enabled to execute the method according to any one of claims 1 to 9.

Images