A processing method and device for N-stage classification prediction based on CT images

Abstract

Embodiments of the present application relate to a processing method and device for N-stage classification prediction based on CT images, the method comprising: receiving CT images and a prediction mode; performing semantic segmentation on the CT images to obtain gastric tumor lesion region points and perigastric lymph node region points based on a first semantic segmentation model; when the prediction mode is a first mode, extracting imageomics features of the gastric tumor lesion region and all perigastric lymph node regions based on the semantic segmentation map and the CT images to obtain a lesion feature set and a lymph feature set, and performing N-stage classification prediction based on the first prediction model according to the lesion feature set and the lymph feature set; when the prediction mode is a second mode, performing gastric tumor lesion and perigastric lymph node region graph synthesis based on the semantic segmentation map and the CT images, and performing N-stage classification prediction based on the second prediction model according to the synthesized graph; and outputting the corresponding prediction type. Through the present application, the prediction real-time performance and prediction efficiency can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a processing method and device for N-stage classification prediction based on CT images. BACKGROUND

[0002] The N-staging rule is a gastric cancer staging rule, which classifies based on the metastasis state of gastric tumors to perigastric lymph nodes. Common N-staging types of the rule include N0 type, N1 type, N2 type, N3a type and N3b type. Currently, medical experts can manually identify the N-staging type based on computed tomography (CT) images. However, it is obvious that this manual identification method has obvious timeliness defects due to the limitation of the working time limit of experts, and the processing efficiency is very low especially for batch CT image identification. SUMMARY

[0003] The present application is aimed at the defects of the prior art, and provides a processing method and device for N-stage classification prediction based on CT images, an electronic device and a computer readable storage medium. The present application first performs semantic segmentation processing on the received CT images based on a semantic segmentation model to obtain corresponding semantic segmentation maps of the gastric tumor lesion area points and the perigastric lymph node area points. Then, two optional prediction modes are provided for prediction: 1) in the first mode state, the imageomic features of the gastric tumor lesion area and the imageomic features of all perigastric lymph node areas are extracted according to the semantic segmentation map and the CT image, and then a first prediction model implemented based on a machine learning classification model is used to perform N-stage classification prediction according to the extracted first lesion feature set and the first lymph feature set to obtain the corresponding prediction type (N0 type, N1 type, N2 type, N3a type and N3b type); 2) in the second mode state, a corresponding synthesis map is obtained by synthesizing the gastric tumor lesion and perigastric lymph node area graph according to the semantic segmentation map and the CT image, and then a second prediction model implemented based on a deep learning classification model is used to perform N-stage classification prediction according to the synthesis map to obtain the prediction type (N0 type, N1 type, N2 type, N3a type and N3b type). Through the present application, the prediction real-time performance and prediction efficiency can be improved.

[0004] To achieve the above-mentioned purpose, the first aspect of the present application embodiment provides a processing method for N-stage classification prediction based on CT images, which comprises:

[0005] receiving a first CT image and a first prediction mode; the first prediction mode includes a first mode and a second mode;

[0006] perform semantic segmentation processing on the first CT image to obtain a corresponding first semantic segmentation map;

[0007] When the first prediction mode is the first mode, imageomics features of a gastric tumor lesion region and imageomics features of all perigastric lymph node regions are extracted according to the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set; and a first prediction type is predicted according to the first prediction model based on the first lesion feature set and the first lymph feature set, wherein the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type;

[0008] When the first prediction mode is the second mode, a first synthesis map is obtained by performing gastric tumor lesion and perigastric lymph node region image synthesis processing on the first semantic segmentation map and the first CT image; and the first prediction type is predicted according to the second prediction model based on the first synthesis map.

[0009] The first prediction type is output as a prediction result corresponding to the first CT image.

[0010] Preferably, the first CT image is a three-dimensional abdominal CT image with complete gastric and perigastric lymph node region three-dimensional information; the shape of the first CT image is D0xH0xW0xC0, D0, H0 and W0 are respectively the depth, height and width of the three-dimensional size of the first CT image, and C0 is the feature channel dimension of the first CT image, C0=1; the first CT image is composed of D0xH0xW0 first voxel points; each first voxel point corresponds to a first voxel point feature vector with a length of C0; the first voxel point feature vector includes a first voxel point pixel value.

[0011] The shape of the first semantic segmentation graph is D1xH1xW1xC1, D1, H1, and W1 are the depth, height, and width of the three-dimensional size of the first semantic segmentation graph respectively, C1 is the feature channel dimension of the first semantic segmentation graph, D1=D0, H1=H0, W1=W0, C1=3; the first semantic segmentation graph is composed of D1xH1xW1 second voxel points; each second voxel point corresponds to a second voxel point feature vector with a length of C1; the second voxel point feature vector includes a first lesion point semantic, a first lymph point semantic, and a first background point semantic; the first lesion point semantic, the first lymph point semantic, and the first background point semantic are all binary semantics, taking values of 0 or 1, and taking a value of 1 as true and a value of 0 as false; only one of the first lesion point semantic, the first lymph point semantic, and the first background point semantic in each second voxel point feature vector is 1;

[0012] The shape of the first synthetic graph is D2xH2xW2xC2, D2, H2, and W2 are the depth, height, and width of the three-dimensional size of the first synthetic graph respectively, C2 is the feature channel dimension of the first synthetic graph, D2<D0, H2<H0, W2<W0, C2=C0+C1; the first synthetic graph is composed of D2xH2xW2 third voxel points; each third voxel point corresponds to a third voxel point feature vector with a length of C2; the third voxel point feature vector includes a second voxel point pixel value, a second lesion point semantic, a second lymph point semantic, and a second background point semantic; the second lesion point semantic, the second lymph point semantic, and the second background point semantic are all binary semantics, taking values of 0 or 1, and taking a value of 1 as true and a value of 0 as false; only one of the second lesion point semantic, the second lymph point semantic, and the second background point semantic in each third voxel point feature vector is 1;

[0013] The first lesion feature set includes a lesion area first-order feature set, a lesion area shape feature set, and a lesion area texture feature set; the lesion area first-order feature set includes a lesion area gray value median, a lesion area gray value average, a lesion area gray value minimum, a lesion area gray value maximum, a lesion area gray value standard deviation, a lesion area gray value skewness, and a lesion area gray value distribution peak value; the lesion area shape feature set includes a lesion area surface area, a lesion area volume, a lesion area volume ratio, a lesion area sphericity, a lesion area compactness type, and a lesion area diameter; the lesion area texture feature set includes a lesion area gray value co-occurrence matrix, a lesion area gray value run length matrix, a lesion area gray value size region matrix, a lesion area neighborhood gray value difference matrix, and a lesion area gray value correlation matrix;

[0014] The first lymph feature set comprises a plurality of first lymph node region feature sets; each of the first lymph node region feature sets corresponds to a perigastric lymph node region; the first lymph node region feature set comprises a first lymph node region distance, a first lymph node region first-order feature set, a first lymph node region shape feature set, and a first lymph node region texture feature set; the first lymph node region distance is a straight-line distance from a center point of the corresponding perigastric lymph node region to a center point of a gastric tumor lesion region; the first lymph node region first-order feature set comprises a lymph node region gray value median, a lymph node region gray value average, a lymph node region gray value minimum, a lymph node region gray value maximum, a lymph node region gray value standard deviation, a lymph node region gray value skewness, and a lymph node region gray value distribution peak; the first lymph node region shape feature set comprises a lymph node surface area, a lymph node volume, a lymph node volume ratio, a lymph node sphericity, a lymph node compactness type, and a lymph node diameter; and the first lymph node region texture feature set comprises a lymph node region gray value co-occurrence matrix, a lymph node region gray value run-length matrix, a lymph node region gray value size region matrix, a lymph node region neighborhood gray value difference matrix, and a lymph node region gray value correlation matrix.

[0015] Preferably, the first semantic segmentation model comprises a first U-Net segmentation network, a second U-Net segmentation network, and a first fusion module; the first and second U-Net segmentation networks are in parallel; the input ends of the first and second U-Net segmentation networks are connected to the model input end of the first semantic segmentation model, and the output ends thereof are connected to the input ends of the first fusion module; and the first and second U-Net segmentation networks are both implemented based on the model structure of a U-Net model.

[0016] The first prediction model is implemented based on a machine learning classification model; and the types of the machine learning classification model at least include a classification prediction model implemented based on an RF model, a classification prediction model implemented based on an SVM model, a classification prediction model implemented based on a GBDT model, and a classification prediction model implemented based on a GBDT+LR model.

[0017] The second prediction model is implemented based on a deep learning classification model; and the types of the deep learning classification model at least include a classification prediction model implemented based on a ResNet model and a classification prediction model implemented based on a Transformer model.

[0018] Preferably, the first CT image is subjected to gastric tumor lesion region point and perigastric lymph node region point semantic segmentation processing based on the first semantic segmentation model to obtain a corresponding first semantic segmentation map, specifically comprising:

[0019] input the first CT image into the first U-Net segmentation network; perform lesion point semantic segmentation on the voxel points of the gastric tumor lesion region on the first CT image by the first U-Net segmentation network to obtain a corresponding first one-segment semantic graph; the shape of the first one-segment semantic graph is D 11 ×H 11 ×W 11 ×C 11 , D 11 , H 11 , W 11 are the depth, height and width of the three-dimensional size of the first one-segment semantic graph respectively, C 11 is the feature channel dimension of the first one-segment semantic graph, D 11 =D0, H 11 =H0, W 11 =W0, C 11 =2; the first one-segment semantic graph is composed of D 11 ×H 11 ×W 11 fourth voxel points; each fourth voxel point corresponds to a fourth voxel point feature vector with a length of C 11 ; the fourth voxel point feature vector includes a third lesion point semantic and a third background point semantic; the third lesion point semantic and the third background point semantic are both binary semantics, taking values of 0 or 1, and taking a value of 1 as true and a value of 0 as false; only one of the third lesion point semantic and the third background point semantic of each fourth voxel point feature vector is 1;

[0020] and input the first CT image into the second U-Net segmentation network; perform lymph node semantic segmentation on the voxel points of all perigastric lymph node regions on the first CT image by the second U-Net segmentation network to obtain a corresponding first two-segment semantic graph; the shape of the first two-segment semantic graph is D 12 ×H 12 ×W 12 ×C 12 , D 12 , H 12 , W 12 are the depth, height and width of the three-dimensional size of the first two-segment semantic graph respectively, C 12 is the feature channel dimension of the first two-segment semantic graph, D 12 =D0, H 12 =H0, W 12 =W0, C 12 =2; the first two-segment semantic graph is composed of D 12 ×H 12 ×W 12A fifth voxel point is composed; each of the fifth voxel points corresponds to a fifth voxel point feature vector with a length of C 12 The fifth voxel point feature vector includes a third lymph node semantic and a fourth background point semantic; the third lymph node semantic and the fourth background point semantic are both binary semantics, with values of 0 or 1, and a value of 1 is true and a value of 0 is false; only one of the third lymph node semantic and the fourth background point semantic of each fifth voxel point feature vector is 1;

[0021] The first one segmentation semantic graph and the first two segmentation semantic graph are input into the first fusion module; the first semantic segmentation graph with all zero semantics is initialized by the first fusion module; all the second voxel points on the first semantic segmentation graph are traversed; during the traversal, the current traversed second voxel point is taken as a corresponding current voxel point, and the fourth voxel point and the fifth voxel point corresponding to the current voxel point on the first one segmentation semantic graph and the first two segmentation semantic graph are recorded as first and second corresponding voxel points; when the third lesion point semantic of the first corresponding voxel point is 1, the first lesion point semantic of the current voxel point is reset to 1; when the third lymph node semantic of the second corresponding voxel point is 1, the first lymph node semantic of the current voxel point is reset to 1; when the third and fourth background point semantics of the first and second corresponding voxel points are both 1, the first background point semantic of the current voxel point is reset to 1; and when the traversal ends, the first semantic segmentation graph after the reset is taken as the processing result of the first semantic segmentation model and output.

[0022] Preferably, the imageomic features of the gastric tumor lesion region and the imageomic features of all perigastric lymph node regions are extracted according to the first semantic segmentation graph and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set, which specifically includes:

[0023] The imageomic features of the gastric tumor lesion region are extracted according to the first semantic segmentation graph and the first CT image to obtain the first lesion feature set;

[0024] The imageomic features of all perigastric lymph node regions are extracted according to the first semantic segmentation graph and the first CT image to obtain the first lymph feature set.

[0025] Further, the imageomic features of the gastric tumor lesion region are extracted according to the first semantic segmentation graph and the first CT image to obtain the first lesion feature set, which specifically includes:

[0026] record the second voxel points with the first semantic segmentation graph as corresponding first lesion voxel points; and record the first voxel points corresponding to each of the first lesion voxel points on the first CT image as corresponding second lesion voxel points;

[0027] record the maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value and the minimum width value in all the second lesion voxel points to obtain corresponding first maximum depth value, first minimum depth value, first maximum height value, first minimum height value, first maximum width value and first minimum width value; and cut out a three-dimensional CT subgraph with a depth range between the first maximum depth value and the first minimum depth value, a height range between the first maximum height value and the first minimum height value, and a width range between the first maximum width value and the first minimum width value on the first CT image as a corresponding first lesion area subgraph;

[0028] perform gray scale conversion on the first lesion area subgraph to obtain a corresponding first lesion area gray scale graph;

[0029] calculate the gray scale median value, the gray scale average value, the gray scale minimum value, the gray scale maximum value, the gray scale standard deviation, the gray scale skewness and the gray scale distribution peak value of the first lesion area gray scale graph to obtain the lesion area gray scale median value, the lesion area gray scale average value, the lesion area gray scale minimum value, the lesion area gray scale maximum value, the lesion area gray scale standard deviation, the lesion area gray scale skewness and the lesion area gray scale distribution peak value, which form a corresponding first-order feature set of the lesion area;

[0030] record a three-dimensional structure composed of all the second lesion voxel points on the first lesion area gray scale graph as a corresponding first tumor structure; and calculate the surface area, the volume, the volume ratio, the sphericity, the compactness type and the diameter of the first tumor structure to obtain the lesion area surface area, the lesion area volume, the lesion area volume ratio, the lesion area sphericity, the lesion area compactness type and the lesion area diameter, which form a corresponding shape feature set of the lesion area;

[0031] calculate the gray scale co-occurrence matrix, the gray scale run length matrix, the gray scale size region matrix, the neighborhood gray scale difference matrix and the gray scale correlation matrix of the first lesion area gray scale graph to obtain the lesion area gray scale co-occurrence matrix, the lesion area gray scale run length matrix, the lesion area gray scale size region matrix, the lesion area neighborhood gray scale difference matrix and the lesion area gray scale correlation matrix, which form a corresponding texture feature set of the lesion area;

[0032] And the first lesion feature set corresponding to the first lesion feature set is composed of the first-order feature set of the lesion area, the shape feature set of the lesion area and the texture feature set of the lesion area.

[0033] Further, the first lymph node feature set is obtained by performing feature extraction processing on the image features of all perigastric lymph node regions according to the first semantic segmentation map and the first CT image, specifically including:

[0034] The second voxel points on the first semantic segmentation map with the first lymph point semantics are recorded as the first lymph voxel points corresponding to each first lymph voxel point; and the first voxel points on the first CT image corresponding to each first lymph voxel point are recorded as the second lymph voxel points corresponding to each first lymph voxel point;

[0035] And the first lymph node feature set is obtained by performing feature extraction processing on the image features of all perigastric lymph node regions according to the first semantic segmentation map and the first CT image, specifically including:

[0036] And the first lymph node feature set is obtained by performing feature extraction processing on the image features of all perigastric lymph node regions according to the first semantic segmentation map and the first CT image, specifically including:

[0037] And the first lymph node feature set is obtained by performing feature extraction processing on the image features of all perigastric lymph node regions according to the first semantic segmentation map and the first CT image, specifically including:

[0038] And the first lymph node feature set is obtained by performing feature extraction processing on the image features of all perigastric lymph node regions according to the first semantic segmentation map and the first CT image, specifically including:

[0039] and the three-dimensional structure composed of all the second lymph node voxels on each of the first lymph node region gray scale images is recorded as a corresponding first lymph structure; and the surface area, volume, volume ratio, sphericity, compactness type and diameter of the first lymph structure are calculated to obtain the corresponding lymph node surface area, lymph node volume, lymph node volume ratio, lymph node sphericity, lymph node compactness type and lymph node diameter, which constitute the corresponding first lymph node region shape feature set;

[0040] and the gray level co-occurrence matrix, gray level run length matrix, gray level size zone matrix, neighborhood gray level difference matrix and gray level correlation matrix of each of the first lymph node region gray scale images are calculated to obtain the corresponding lymph node region gray level co-occurrence matrix, lymph node region gray level run length matrix, lymph node region gray level size zone matrix, lymph node region neighborhood gray level difference matrix and lymph node region gray level correlation matrix, which constitute the corresponding first lymph node region texture feature set;

[0041] and each of the first perigastric lymph node region subgraph corresponds to the first lymph node region distance, first lymph node region first-order feature set, first lymph node region shape feature set and first lymph node region texture feature set, which constitute a corresponding first lymph node region feature set; and all the first lymph node region feature sets obtained constitute a corresponding first lymph feature set.

[0042] Preferably, the first prediction type is obtained by performing N stage classification prediction processing on the first lesion feature set and the first lymph feature set based on the first prediction model, specifically including:

[0043] each of the first lesion feature set and the first lymph feature set constitutes a corresponding first lesion-lymph feature set; and all the first lesion-lymph feature sets are sorted in order from near to far according to the first lymph node region distance in the first lymph node region feature set in each of the first lesion-lymph feature sets to constitute a corresponding first feature set sequence;

[0044] the first feature set sequence is input into the first prediction model; the metastasis feature of gastric cancer tumor metastasis to all perigastric lymph node regions is identified by the first prediction model according to the imaging features of the first feature set sequence, and the five N stage types of gastric cancer tumor are classified and predicted based on the metastasis feature identification result to obtain a corresponding first prediction vector; the first prediction vector includes five first prediction probabilities; each of the first prediction probabilities corresponds to one of the N stage types; the N stage types include N0 type, N1 type, N2 type, N3a type and N3b type;

[0045] The N-stage type corresponding to the maximum probability value in the first prediction vector is taken as the corresponding first prediction type.

[0046] Preferably, the first synthesis graph is obtained by performing a stomach tumor lesion and perigastric lymph node region graph synthesis processing on the first semantic segmentation graph and the first CT image, and specifically includes the following steps.

[0047] The second voxel points with the first lesion point semantics or the first lymph point semantics of 1 on the first semantic segmentation graph are recorded as corresponding first-class voxel points, and the first voxel points corresponding to each first-class voxel point on the first CT image are recorded as corresponding second-class voxel points.

[0048] The maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value, and the minimum width value of all the second-class voxel points are counted to obtain corresponding third maximum depth value, third minimum depth value, third maximum height value, third minimum height value, third maximum width value, and third minimum width value.

[0049] The subgraphs with the depth range between the third maximum depth value and the third minimum depth value, the height range between the third maximum height value and the third minimum height value, and the width range between the third maximum width value and the third minimum width value on the first semantic segmentation graph and the first CT image are cut out as corresponding first segmentation subgraph and first CT subgraph; the three-dimensional sizes of the first segmentation subgraph and the first CT subgraph are consistent.

[0050] The depth, height, and width of the three-dimensional size of the first segmentation subgraph or the first CT subgraph are taken as corresponding depth D2, height H2, and width W2; the sum of the feature channel dimension C0 of the first CT subgraph and the feature channel dimension C1 of the first segmentation subgraph is taken as corresponding feature channel dimension C2; and the first segmentation subgraph and the first CT subgraph are spliced in a voxel point feature vector splicing manner to obtain a first synthesis graph with a shape of D2×H2×W2×C2.

[0051] Preferably, the first prediction type is obtained by performing an N-stage classification prediction processing on the first synthesis graph based on a second prediction model, and specifically includes the following steps.

[0052] inputting the first synthesis graph into the second prediction model; identifying, by the second prediction model, metastasis features of gastric cancer tumor metastasis to all perigastric lymph node regions according to image features of the first synthesis graph, and classifying and predicting five N stage types of the gastric cancer tumor based on the metastasis feature identification result to obtain a corresponding second prediction vector; the second prediction vector includes five second prediction probabilities; each second prediction probability corresponds to one N stage type; the N stage types include N0 type, N1 type, N2 type, N3a type and N3b type;

[0053] taking the N stage type corresponding to the second prediction probability with the maximum probability value in the second prediction vector as the corresponding first prediction type.

[0054] The second aspect of the embodiment of the present application provides a device for implementing the processing method for N stage classification and prediction based on CT images according to the first aspect.

[0055] The data receiving module is used to receive a first CT image and a first prediction mode; the first prediction mode includes a first mode and a second mode.

[0056] The semantic segmentation module is used to perform semantic segmentation processing on the first CT image based on a first semantic segmentation model to obtain a corresponding first semantic segmentation graph.

[0057] The first prediction model processing module is used to, when the first prediction mode is the first mode, perform feature extraction processing on imageomics features of a gastric tumor lesion region and imageomics features of all perigastric lymph node regions according to the first semantic segmentation graph and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set; and perform N stage classification and prediction processing based on a first prediction model according to the first lesion feature set and the first lymph feature set to obtain a corresponding first prediction type; the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type.

[0058] The second prediction model processing module is used to, when the first prediction mode is the second mode, perform gastric tumor lesion and perigastric lymph node region graph synthesis processing according to the first semantic segmentation graph and the first CT image to obtain a corresponding first synthesis graph; and perform N stage classification and prediction processing based on a second prediction model according to the first synthesis graph to obtain the corresponding first prediction type.

[0059] The prediction output module is used to output the obtained first prediction type as a prediction result corresponding to the first CT image.

[0060] The third aspect of the embodiment of the present application provides an electronic device, comprising a memory, a processor and a transceiver;

[0061] The processor is used for coupling with the memory, reading and executing instructions in the memory to realize the method steps of the first aspect;

[0062] The transceiver is coupled with the processor, and the transceiver is controlled by the processor to perform message transmission and reception.

[0063] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores computer instructions, when the computer instructions are executed by a computer, the computer instructions make the computer execute the instructions of the method of the first aspect.

[0064] The embodiment of the present application provides a processing method and device for N-stage classification prediction based on CT images, electronic equipment and computer readable storage medium. From the above content, it can be known that the embodiment of the present application first performs semantic segmentation processing on the received CT images based on a semantic segmentation model to obtain corresponding semantic segmentation graphs of the gastric tumor lesion region points and the perigastric lymph node region points; then two optional prediction modes are provided for prediction: 1) in the first mode state, first, the imageomic features of the gastric tumor lesion region and the imageomic features of all perigastric lymph node regions are extracted according to the semantic segmentation graph and the CT images, and then a first prediction model based on a machine learning classification model is used to perform N-stage classification prediction according to the extracted first lesion feature set and the first lymph feature set to obtain corresponding prediction types (N0 type, N1 type, N2 type, N3a type and N3b type); 2) in the second mode state, first, a corresponding synthesis graph is obtained by synthesizing the gastric tumor lesion and the perigastric lymph node region graph according to the semantic segmentation graph and the CT images, and then a second prediction model based on a deep learning classification model is used to perform N-stage classification prediction according to the synthesis graph to obtain the prediction types (N0 type, N1 type, N2 type, N3a type and N3b type). Through the embodiment of the present application, the prediction real-time performance and the prediction efficiency are improved. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 A processing method for N-stage classification prediction based on CT images is provided for the first embodiment of the present application;

[0066] Figure 2 A module structure diagram of a processing device for N-stage classification prediction based on CT images is provided for the second embodiment of the present application;

[0067] Figure 3A structural schematic diagram of an electronic device provided for the third embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0069] The first embodiment of the present application provides a processing method for N-stage classification prediction based on CT images, as shown in the following figure. Figure 1 The first embodiment of the present application provides a processing method for N-stage classification prediction based on CT images, as shown in the following figure.

[0070] Step 1, receiving a first CT image and a first prediction mode;

[0071] The first CT image is a three-dimensional abdominal CT image with complete stomach and all lymph node regions around the stomach three-dimensional information; the shape of the first CT image is D0xH0xW0xC0, D0, H0 and W0 are the depth, height and width of the three-dimensional size of the first CT image respectively, C0 is the feature channel dimension of the first CT image, C0=1; the first CT image is composed of D0xH0xW0 first voxel points; each first voxel point corresponds to a first voxel point feature vector with a length of C0; the first voxel point feature vector includes the first voxel point pixel value; the first prediction mode includes a first mode and a second mode.

[0072] Step 2, performing stomach tumor lesion region point and perigastric lymph node region point semantic segmentation processing on the first CT image based on a first semantic segmentation model to obtain a corresponding first semantic segmentation map;

[0073] The first semantic segmentation model includes a first U-Net segmentation network, a second U-Net segmentation network and a first fusion module; the first and second U-Net segmentation networks are parallel; the input ends of the first and second U-Net segmentation networks are connected with the model input end of the first semantic segmentation model, and the output ends are connected with the input ends of the first fusion module; the first and second U-Net segmentation networks are realized based on the model structure of the U-Net model; here, the U-Net model is a visual model commonly used for processing semantic segmentation tasks;

[0074] The shape of the first semantic segmentation map is D1xH1xW1xC1, D1, H1, and W1 are the depth, height, and width of the three-dimensional size of the first semantic segmentation map respectively, C1 is the feature channel dimension of the first semantic segmentation map, D1=D0, H1=H0, W1=W0, and C1=3; the first semantic segmentation map is composed of D1xH1xW1 second voxel points; each second voxel point corresponds to a second voxel point feature vector with a length of C1; the second voxel point feature vector includes a first lesion point semantic, a first lymph node semantic, and a first background point semantic; the first lesion point semantic, the first lymph node semantic, and the first background point semantic are all binary semantics, and the values are 0 or 1, and the value 1 is true and the value 0 is false; only one of the first lesion point semantic, the first lymph node semantic, and the first background point semantic of each second voxel point feature vector is 1;

[0075] The specific steps of the current step 2 include:

[0076] Step 21, input the first CT image into the first U-Net segmentation network; perform lesion point semantic segmentation on the voxel points of the gastric tumor lesion region on the first CT image by the first U-Net segmentation network to obtain a corresponding first one-segment semantic map;

[0077] The shape of the first one-segment semantic map is D 11 ×H 11 ×W 11 ×C 11 , D 11 , H 11 , and W 11 are the depth, height, and width of the three-dimensional size of the first one-segment semantic map respectively, C 11 is the feature channel dimension of the first one-segment semantic map, D 11 =D0, H 11 =H0, W 11 =W0, and C 11 =2; the first one-segment semantic map is composed of D 11 ×H 11 ×W 11 fourth voxel points; each fourth voxel point corresponds to a fourth voxel point feature vector with a length of C 11 ; the fourth voxel point feature vector includes a third lesion point semantic and a third background point semantic; the third lesion point semantic and the third background point semantic are both binary semantics, and the values are 0 or 1, and the value 1 is true and the value 0 is false; only one of the third lesion point semantic and the third background point semantic of each fourth voxel point feature vector is 1;

[0078] Step 22, input the first CT image into the second U-Net segmentation network; and perform lymph node semantic segmentation on all the voxel points of the perigastric lymph node region in the first CT image by the second U-Net segmentation network to obtain a corresponding first two segmentation semantic map;

[0079] wherein, the shape of the first two segmentation semantic map is D 12 ×H 12 ×W 12 ×C 12 , D 12 , H 12 , W 12 are the three-dimensional size of the first two segmentation semantic map, respectively, depth, height and width, C 12 is the feature channel dimension of the first two segmentation semantic map, D 12 =D0, H 12 =H0, W 12 =W0, C 12 =2; the first two segmentation semantic map is composed of D 12 ×H 12 ×W 12 fifth voxel points; each fifth voxel point corresponds to a fifth voxel point feature vector with a length of C 12 ; the fifth voxel point feature vector includes a third lymph node semantic and a fourth background point semantic; the third lymph node semantic and the fourth background point semantic are both binary semantics, taking values of 0 or 1, and taking a value of 1 as true and a value of 0 as false; there is only one of the third lymph node semantic and the fourth background point semantic in each fifth voxel point feature vector with a value of 1;

[0080] Step 23, input the first one segmentation semantic map and the first two segmentation semantic map into the first fusion module; initialize a first semantic segmentation map with a full zero semantic by the first fusion module; and traverse all the second voxel points on the first semantic segmentation map; and during the traversal, take the currently traversed second voxel point as a corresponding current voxel point, and take the fourth voxel point and the fifth voxel point corresponding to the current voxel point on the first one segmentation semantic map and the first two segmentation semantic map as corresponding first and second corresponding voxel points; and when the third lesion point semantic of the first corresponding voxel point is 1, reset the first lesion point semantic of the current voxel point to 1; and when the third lymph node semantic of the second corresponding voxel point is 1, reset the first lymph node semantic of the current voxel point to 1; and when the third and fourth background point semantics of the first and second corresponding voxel points are both 1, reset the first background point semantic of the current voxel point to 1; and when the traversal is completed, output the reset first semantic segmentation map as the processing result of the first semantic segmentation model.

[0081] Step 3, when the first prediction mode is the first mode, performing feature extraction processing on the image features of the gastric tumor lesion area and the image features of all perigastric lymph node areas according to the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set; and performing N stage classification prediction processing according to the first lesion feature set and the first lymph feature set based on the first prediction model to obtain a corresponding first prediction type;

[0082] The first prediction type includes an N0 type, an N1 type, an N2 type, an N3a type and an N3b type.

[0083] Specifically, step 31 comprises: when the first prediction mode is the first mode, performing feature extraction processing on the image features of the gastric tumor lesion area and the image features of all perigastric lymph node areas according to the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set.

[0084] Specifically, step 311 comprises: performing feature extraction processing on the image features of the gastric tumor lesion area according to the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set.

[0085] The first lesion feature set includes a lesion area first-order feature set, a lesion area shape feature set and a lesion area texture feature set; the lesion area first-order feature set includes a lesion area gray value median, a lesion area gray value average, a lesion area gray value minimum, a lesion area gray value maximum, a lesion area gray value standard deviation, a lesion area gray value skewness and a lesion area gray value distribution peak value; the lesion area shape feature set includes a lesion area surface area, a lesion area volume, a lesion area volume ratio, a lesion area sphericity, a lesion area compactness type and a lesion area diameter; and the lesion area texture feature set includes a lesion area gray value co-occurrence matrix, a lesion area gray value run length matrix, a lesion area gray value size region matrix, a lesion area neighborhood gray value difference matrix and a lesion area gray value correlation matrix.

[0086] Specifically, step 3111 comprises: recording each second voxel point with a first lesion point semantic of 1 on the first semantic segmentation map as a corresponding first lesion voxel point; and recording each first voxel point corresponding to each first lesion voxel point on the first CT image as a corresponding second lesion voxel point.

[0087] Step 3112, and the maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value and the minimum width value in all second lesion voxel points are counted to obtain the corresponding first maximum depth value, the first minimum depth value, the first maximum height value, the first minimum height value, the first maximum width value and the first minimum width value; and the three-dimensional CT subgraph with the depth range between the first maximum depth value and the first minimum depth value, the height range between the first maximum height value and the first minimum height value, and the width range between the first maximum width value and the first minimum width value on the first CT image is cut out as the corresponding first lesion area subgraph;

[0088] Step 3113, and the first lesion area subgraph is converted into a gray scale graph to obtain the corresponding first lesion area gray scale graph;

[0089] Step 3114, and the gray scale median value, the gray scale average value, the gray scale minimum value, the gray scale maximum value, the gray scale standard deviation, the gray scale skewness and the gray scale distribution peak value of the first lesion area gray scale graph are calculated to obtain the lesion area gray scale median value, the lesion area gray scale average value, the lesion area gray scale minimum value, the lesion area gray scale maximum value, the lesion area gray scale standard deviation, the lesion area gray scale skewness and the lesion area gray scale distribution peak value, which form the corresponding first-order feature set of the lesion area;

[0090] Here, the calculation method for calculating the first-order features (gray scale median value, gray scale average value, gray scale minimum value, gray scale maximum value, gray scale standard deviation, gray scale skewness and gray scale distribution peak value) of the three-dimensional gray scale graph is a disclosed first-order feature calculation method of imageomics features, and the implementation details can be confirmed by related technical literature, which will not be repeated here;

[0091] Step 3115, and the three-dimensional structure composed of all second lesion voxel points on the first lesion area gray scale graph is recorded as the corresponding first tumor structure; and the surface area, the volume, the volume ratio, the sphericity, the compactness type and the diameter of the first tumor structure are calculated to obtain the lesion area surface area, the lesion area volume, the lesion area volume ratio, the lesion area sphericity, the lesion area compactness type and the lesion area diameter, which form the corresponding shape feature set of the lesion area;

[0092] Here, the calculation method for calculating the geometric features (surface area, volume, volume ratio, sphericity, compactness type and diameter) of the specified three-dimensional structure in the three-dimensional gray scale graph is a disclosed geometric feature calculation method of imageomics features, and the implementation details can be confirmed by related technical literature, which will not be repeated here;

[0093] Step 3116, and the gray level co-occurrence matrix, the gray level run length matrix, the gray level size zone matrix, the neighborhood gray level difference matrix and the gray level correlation matrix of the first lesion area gray level image are calculated to obtain the corresponding lesion area gray level co-occurrence matrix, the lesion area gray level run length matrix, the lesion area gray level size zone matrix, the lesion area neighborhood gray level difference matrix and the lesion area gray level correlation matrix to form the corresponding lesion area texture feature set;

[0094] Here, the calculation method for calculating the texture features (gray level co-occurrence matrix, gray level run length matrix, gray level size zone matrix, neighborhood gray level difference matrix and gray level correlation matrix) in the three-dimensional gray level image is a publicly disclosed texture feature calculation method of imageomics features, and the implementation details can be confirmed by relevant technical documents, which will not be repeated here.

[0095] Step 3117, and the first lesion area first-order feature set, the first lesion area shape feature set and the first lesion area texture feature set are combined to form the corresponding first lesion feature set.

[0096] Step 312, according to the first semantic segmentation image and the first CT image, the imageomics features of all the perigastric lymph node regions are extracted to obtain the corresponding first lymph feature set;

[0097] The first lymph feature set includes a plurality of first lymph node area feature sets; each first lymph node area feature set corresponds to a perigastric lymph node region; the first lymph node area feature set includes a first lymph node area distance, a first lymph node area first-order feature set, a first lymph node area shape feature set and a first lymph node area texture feature set; the first lymph node area distance is the straight line distance from the center point of the corresponding perigastric lymph node region to the center point of the gastric tumor lesion region; the first lymph node area first-order feature set includes a lymph node area gray level median value, a lymph node area gray level average value, a lymph node area gray level minimum value, a lymph node area gray level maximum value, a lymph node area gray level standard deviation, a lymph node area gray level skewness and a lymph node area gray level distribution peak value; the first lymph node area shape feature set includes a lymph node surface area, a lymph node volume, a lymph node volume ratio, a lymph node sphericity, a lymph node compactness type and a lymph node diameter; and the first lymph node area texture feature set includes a lymph node area gray level co-occurrence matrix, a lymph node area gray level run length matrix, a lymph node area gray level size zone matrix, a lymph node area neighborhood gray level difference matrix and a lymph node area gray level correlation matrix.

[0098] Specifically, step 3121, each second voxel point with a first lymph point semantic of 1 on the first semantic segmentation image is recorded as a corresponding first lymph voxel point; and each first voxel point corresponding to each first lymph voxel point on the first CT image is recorded as a corresponding second lymph voxel point.

[0099] Step 3122, and point cluster all the second lymph nodes to get a plurality of first point group set corresponding to the class; and each first point group set as the corresponding current point group set; and the maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value and the minimum width value of all the second lymph nodes in the current point group set are counted to get the corresponding second maximum depth value, the second minimum depth value, the second maximum height value, the second minimum height value, the second maximum width value and the second minimum width value; and the first CT image on the depth range between the second maximum depth value and the second minimum depth value, the height range between the second maximum height value and the second minimum height value, the width range between the second maximum width value and the second minimum width value Three-dimensional CT subgraph is cut out as the corresponding first perigastric lymph node subgraph;

[0100] Among them, the first point group set includes a plurality of second lymph nodes;

[0101] Step 3123, and calculate the straight line distance from the center point of each first perigastric lymph node subgraph to the center point of the first lesion subgraph to get the corresponding first lymph node area distance;

[0102] Step 3124, and perform gray scale conversion on each first perigastric lymph node subgraph to get the corresponding first lymph node gray scale graph;

[0103] Step 3125, and calculate the gray value, gray average value, gray minimum value, gray maximum value, gray standard deviation, gray skewness and gray distribution peak of each first lymph node gray scale graph to get the corresponding lymph node gray value, lymph node gray average value, lymph node gray minimum value, lymph node gray maximum value, lymph node gray standard deviation, lymph node gray skewness and lymph node gray distribution peak Peak value, the first lymph node first feature set corresponding to the first lymph node first feature set is composed of;

[0104] Step 3126, and record each first lymph node gray scale graph as a three-dimensional structure composed of all second lymph nodes as the corresponding first lymph structure; and calculate the surface area, volume, volume ratio, sphericity, compactness type and diameter of the first lymph structure to get the corresponding lymph node surface area, lymph node volume, lymph node volume ratio, lymph node sphericity, lymph node compactness type and lymph node diameter Lymph node diameter, the shape feature set of the first lymph node area corresponding to the shape feature set of the first lymph node area is composed of;

[0105] Step 3127, and the gray level co-occurrence matrix, the gray level run length matrix, the gray level size zone matrix, the neighborhood gray level difference matrix and the gray level correlation matrix of each first lymph node region gray scale image are calculated to obtain the corresponding lymph node region gray level co-occurrence matrix, the lymph node region gray level run length matrix, the lymph node region gray level size zone matrix, the lymph node region neighborhood gray level difference matrix and the lymph node region gray level correlation matrix, which constitute the corresponding first lymph node region texture feature set;

[0106] Step 3128, and the first lymph node region distance, the first lymph node region first-order feature set, the first lymph node region shape feature set and the first lymph node region texture feature set of each first perigastric lymph node region sub-image constitute a corresponding first lymph node region feature set; and all the obtained first lymph node region feature sets constitute a corresponding first lymph node feature set collection;

[0107] Step 32, and the first prediction type is obtained by performing N stage classification prediction processing based on the first prediction model according to the first lesion feature set and the first lymph node feature set collection;

[0108] The first prediction model is implemented based on a machine learning classification model; the types of the machine learning classification model at least include a classification prediction model implemented based on an RF model, a classification prediction model implemented based on an SVM model, a classification prediction model implemented based on a GBDT model and a classification prediction model implemented based on a GBDT+LR model;

[0109] Specifically, it includes: step 321, combining the first lesion feature set with each first lymph node region feature set of the first lymph node feature set collection to form a corresponding first lesion-lymph feature set; and sorting all the first lesion-lymph feature sets in the order from near to far according to the first lymph node region distance of the first lymph node region feature set in each first lesion-lymph feature set to form a corresponding first feature set sequence;

[0110] Step 322, inputting the first feature set sequence into the first prediction model; and identifying the metastasis feature of the gastric cancer tumor to all perigastric lymph node regions by the first prediction model according to the imaging feature of the first feature set sequence, and classifying the five N stage types of the gastric cancer tumor based on the metastasis feature identification result to obtain a corresponding first prediction vector;

[0111] The first prediction vector includes five first prediction probabilities; each first prediction probability corresponds to an N stage type; the N stage type includes an N0 type, an N1 type, an N2 type, an N3a type and an N3b type;

[0112] Step 323, taking the N stage type corresponding to the first prediction probability with the maximum probability value in the first prediction vector as the corresponding first prediction type.

[0113] Step 4, when the first prediction mode is the second mode, performing a gastric tumor lesion and perigastric lymph node region map synthesis processing according to the first semantic segmentation map and the first CT image to obtain a corresponding first synthesis map; and performing N-stage classification prediction processing according to the first synthesis map based on the second prediction model to obtain a corresponding first prediction type;

[0114] Specifically comprising: step 41, when the first prediction mode is the second mode, performing a gastric tumor lesion and perigastric lymph node region map synthesis processing according to the first semantic segmentation map and the first CT image to obtain a corresponding first synthesis map;

[0115] Wherein, the shape of the first synthesis map is D2×H2×W2×C2, D2, H2 and W2 are respectively the depth, height and width of the three-dimensional size of the first synthesis map, and C2 is the feature channel dimension of the first synthesis map, D2<D0, H2<H0, W2<W0, and C2=C0+C1; the first synthesis map is composed of D2×H2×W2 third voxel points; each third voxel point corresponds to a third voxel point feature vector with a length of C2; the third voxel point feature vector includes a second voxel point pixel value, a second lesion point semantic, a second lymph point semantic and a second background point semantic; the second lesion point semantic, the second lymph point semantic and the second background point semantic are all binary semantics, taking values of 0 or 1, and taking a value of 1 as true and a value of 0 as false; only one of the second lesion point semantic, the second lymph point semantic and the second background point semantic of each third voxel point feature vector is 1;

[0116] Specifically comprising: step 411, recording the second voxel points with each first lesion point semantic or first lymph point semantic of 1 on the first semantic segmentation map as corresponding first type voxel points; and recording the first voxel points corresponding to each first type voxel point on the first CT image as corresponding second type voxel points;

[0117] Step 412, and statistically obtaining the third maximum depth value, the third minimum depth value, the third maximum height value, the third minimum height value, the third maximum width value and the third minimum width value corresponding to the maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value and the minimum width value in all second type voxel points;

[0118] Step 413, and cutting out the sub-maps with the depth range between the third maximum depth value and the third minimum depth value, the height range between the third maximum height value and the third minimum height value, and the width range between the third maximum width value and the third minimum width value on the first semantic segmentation map and the first CT image as corresponding first segmentation sub-maps and first CT sub-maps;

[0119] Wherein, the three-dimensional sizes of the first segmentation sub-maps and the first CT sub-maps are consistent;

[0120] Step 414, and take the depth, height and width of the three-dimensional size of the first segmentation subgraph or the first CT subgraph as the corresponding depth D2, height H2 and width W2; and take the sum of the feature channel dimension C0 of the first CT subgraph and the feature channel dimension C1 of the first segmentation subgraph as the corresponding feature channel dimension C2; and splice the first segmentation subgraph and the first CT subgraph in the voxel point feature vector splicing manner to obtain a first synthesis graph with the shape of D2xH2xW2xC2;

[0121] Step 42, and perform N stage classification prediction processing on the first synthesis graph based on the second prediction model to obtain a corresponding first prediction type;

[0122] The second prediction model is implemented based on a deep learning classification model, and the types of the deep learning classification model at least include a classification prediction model implemented based on a ResNet model and a classification prediction model implemented based on a Transformer model.

[0123] Specifically, it includes: step 421, inputting the first synthesis graph into the second prediction model; identifying the metastasis feature of the gastric cancer tumor to all perigastric lymph node regions by the second prediction model based on the image features of the first synthesis graph, and performing classification prediction on the five N stage types of the gastric cancer tumor based on the metastasis feature identification result to obtain a corresponding second prediction vector;

[0124] The second prediction vector includes five second prediction probabilities; each second prediction probability corresponds to an N stage type; and the N stage type includes an N0 type, an N1 type, an N2 type, an N3a type and an N3b type.

[0125] Step 422, taking the N stage type corresponding to the second prediction probability with the maximum probability value in the second prediction vector as the corresponding first prediction type.

[0126] Step 5, taking the obtained first prediction type as the prediction result corresponding to the first CT image and outputting it.

[0127] Here, the embodiments of the present application can perform gastric cancer N stage classification prediction on the input CT image based on two types of optional prediction models through the above steps 1-5 and output the corresponding prediction result.

[0128] As can be seen from the above steps 1-5, the embodiments of the present application use three types of models to complete the overall prediction processing process, which are: a first semantic segmentation model, a first prediction model and a second prediction model. It should be noted that before using these three types of models, the first semantic segmentation model, the first prediction model and the second prediction model need to be trained respectively. The processing steps of training these three types of models will be described below.

[0129] 1) The processing steps for training the first semantic segmentation model are as follows:

[0130] Step A1, a large number of three-dimensional abdominal CT images of gastric cancer patients diagnosed as various types of N stage gastric cancer are collected as corresponding sample images to form a first image set;

[0131] Among them, the first image set includes a plurality of first sample images;

[0132] Step A2, and based on artificial labeling or other machine labeling methods, the gastric tumor lesion area on each first sample image is labeled to obtain a corresponding first labeled image; and based on the gastric tumor lesion area labeling information of the first labeled image, a first labeled semantic graph with the same three-dimensional size as the first labeled image and a feature channel dimension of 2 is constructed according to the shape characteristics of the first segmentation semantic graph output by the first U-Net segmentation network.

[0133] Step A3, and based on artificial labeling or other machine labeling methods, all perigastric lymph node regions on each first sample image are labeled to obtain a corresponding second labeled image; and based on the labeling information of all perigastric lymph node regions on the second labeled image, a second labeled semantic graph with the same three-dimensional size as the second labeled image and a feature channel dimension of 2 is constructed according to the shape characteristics of the first two segmentation semantic graphs output by the second U-Net segmentation network.

[0134] Step A4, and each first sample image and the corresponding first labeled semantic graph are combined to form a corresponding first training data record; and each first sample image and the corresponding second labeled semantic graph are combined to form a corresponding second training data record; all first training data records are combined to form a corresponding first training data set; and all second training data records are combined to form a corresponding second training data set.

[0135] Step A5, and the first U-Net segmentation network is trained based on the first training data set.

[0136] Specifically, it includes the following steps:

[0137] Step A52, input the first sample image of the current training data record into the first U-Net segmentation network; perform lesion point semantic segmentation on the voxel points of the gastric tumor lesion area on the input first sample image by the first U-Net segmentation network to obtain a corresponding first predicted segmentation semantic graph;

[0138] Step A53, the first predicted segmentation semantic graph and the first labeled semantic graph of the current training data record are input into the preset first model loss function to obtain a corresponding first loss value;

[0139] Here, the first model loss function is implemented based on an L1 loss function or an L2 loss function;

[0140] Step A54, whether the first loss value meets a preset first loss value range is identified; if the first loss value meets the first loss value range, whether the current training data record is the last first training data record of the first training data set is identified, if yes, step A55 is transferred to, if no, the next first training data record of the first training data set is extracted as a new current training data record and step A52 is returned to continue training; if the first loss value does not meet the first loss value range, the model parameters of the first U-Net segmentation network are optimized for one round based on a preset first model parameter optimizer, and step A52 is returned to continue training at the end of the round of optimization;

[0141] Here, the first loss value range is a preset loss value range; and the first model parameter optimizer at least includes an SDG optimizer;

[0142] Step A55, the model parameters of the first U-Net segmentation network are solidified, and it is confirmed that the model training of the first U-Net segmentation network is ended;

[0143] Step A6, the second U-Net segmentation network is trained based on the second training data set;

[0144] Specifically, it includes: step A61, the first second training data record of the second training data set is taken as a corresponding current training data record;

[0145] Step A62, the first sample image of the current training data record is input into the second U-Net segmentation network; all the voxel points of the perigastric lymph node regions on the first sample image input this time are subjected to lymph point semantic segmentation by the second U-Net segmentation network to obtain a corresponding second predicted segmentation semantic map;

[0146] Step A63, the second predicted segmentation semantic map and the second label semantic map of the current training data record are brought into a preset second model loss function to obtain a corresponding second loss value;

[0147] Here, the second model loss function is implemented based on an L1 loss function or an L2 loss function;

[0148] Step A64, identify whether the second loss value meets the preset second loss value range; if the second loss value meets the second loss value range, identify whether the current training data record is the last second training data record of the second training data set, if yes, go to step A65, if no, extract the next second training data record of the second training data set as a new current training data record and return to step A62 for training; if the second loss value does not meet the second loss value range, perform one round of optimization on the model parameters of the second U-Net segmentation network based on the preset second model parameter optimizer, and return to step A62 for training at the end of the round of optimization;

[0149] Here, the second loss value range is a pre-set loss value range; the second model parameter optimizer at least includes an SDG optimizer;

[0150] Step A65, solidify the model parameters of the second U-Net segmentation network, and confirm that the model training of the second U-Net segmentation network is completed;

[0151] Step A7, after confirming that the model training of the first and second U-Net segmentation networks is completed, confirm that the model training of the first semantic segmentation model is completed.

[0152] 2) The processing steps for training the first prediction model are as follows:

[0153] Step B1, take each first sample image in the aforementioned first image set as a corresponding current sample image; take the first and second label semantic maps corresponding to the current sample image as the corresponding current lesion area label semantic map and the current perigastric lymph node area label semantic map; extract the imageomic features of the gastric tumor lesion area from the first sample image according to the current lesion area label semantic map and the first sample image to obtain a corresponding first sample lesion feature set; process the imageomic features of all perigastric lymph node areas according to the current perigastric lymph node area label semantic map and the first sample image to obtain a corresponding first sample lymph feature set; and form a corresponding first sample lesion-lymph feature set by combining each first sample lymph node area feature set in the first sample lesion feature set and the first sample lymph feature set; sort all first sample lesion-lymph feature sets in order of distance from the lymph node area to form a corresponding first sample feature set sequence; set a corresponding first label vector based on the N stage type of the gastric cancer patient corresponding to the current sample image; and form a corresponding third training data record by combining the first sample feature set sequence corresponding to the current sample image and the first label vector;

[0154] The first sample lymphatic feature set is composed of a plurality of first sample lymph node region feature sets; the first label vector includes five first label probabilities; each first label probability corresponds to an N stage type; and the first label probability corresponding to the N stage type of the gastric cancer patient corresponding to the current sample image in the first label vector is 1, and the remaining four first label probabilities are 0.

[0155] Step B2, and the corresponding third training data set is composed of all the third training data records obtained;

[0156] Step B3, and a plurality of hyperparameter combinations corresponding to the first prediction model are obtained by performing full combination on the hyperparameters of the first prediction model; and a corresponding first prediction model is obtained by setting the hyperparameters of the first prediction model based on each hyperparameter combination;

[0157] Step B4, and the first evaluation score is obtained by training and evaluating each first prediction model based on the third training data set in a k-fold cross-validation manner;

[0158] Here, k is a positive integer greater than or equal to 3; the model training and evaluation in the k-fold cross-validation manner is a public technique, which will not be described further here; it should be noted that the objective function or loss function used in the training of the model is realized based on the cross-entropy objective function or cross-entropy loss function for multi-classification, and the evaluation method used in the evaluation of the model is the F1 score evaluation method based on the multi-classification confusion matrix;

[0159] Step B5, the model full parameter set corresponding to the first prediction model corresponding to the highest first evaluation score is taken as the optimal model full parameter set; and the parameter of the first prediction model is solidified based on the optimal model full parameter set, and the model training of the first prediction model is confirmed to be completed.

[0160] 3) The processing steps of training the second prediction model are briefly described as follows:

[0161] Step C1, each first sample image in the aforementioned first image set is taken as a corresponding current sample image; and the first and second label semantic maps corresponding to the current sample image are taken as a corresponding current lesion region label semantic map and a current perigastric lymph node region label semantic map; and a gastric tumor lesion and perigastric lymph node region map synthesis processing is performed according to the current lesion region label semantic map, the current perigastric lymph node region label semantic map and the current sample image to obtain a corresponding first sample synthesis map; a corresponding second label vector is set based on the N stage type of the gastric cancer patient corresponding to the current sample image; and a corresponding fourth training data record is composed of the first sample synthesis map corresponding to the current sample image and the second label vector;

[0162] The second label vector includes five second label probabilities; each second label probability corresponds to an N stage type; the second label probability corresponding to the N stage type of the gastric cancer patient corresponding to the current sample image in the second label vector is 1, and the other four second label probabilities are 0;

[0163] Step C2, and the corresponding fourth training data set is composed of all the fourth training data records obtained;

[0164] Step C3, the first fourth training data record of the fourth training data set is taken as the corresponding current training data record;

[0165] Step C4, the first sample synthetic image of the current training data record is input into the second prediction model; the metastasis feature of the gastric cancer tumor to all perigastric lymph node regions is identified according to the image features of the input first sample synthetic image by the second prediction model, and the five N stage types of the gastric cancer tumor are classified and predicted based on the metastasis feature identification result to obtain the corresponding first training prediction vector;

[0166] The first training prediction vector includes five first training prediction probabilities; each first training prediction probability corresponds to an N stage type;

[0167] Step C5, the first training prediction vector and the second label vector of the current training data record are input into the preset third model loss function to obtain the corresponding third loss value;

[0168] Here, the third model loss function is realized based on the cross-entropy loss function of multi-classification;

[0169] Step C6, whether the third loss value meets the preset third loss value range is identified; if the third loss value meets the third loss value range, whether the current training data record is the last third training data record of the third training data set is identified, if yes, it goes to Step C7, if not, the next third training data record of the third training data set is extracted as a new current training data record and returns to Step C4 for continuous training; if the third loss value does not meet the third loss value range, the model parameters of the second prediction model are optimized for one round based on the preset third model parameter optimizer, and at the end of this round of optimization, it returns to Step C4 for continuous training;

[0170] Here, the third loss value range is a pre-set loss value range; the third model parameter optimizer at least includes an SDG optimizer;

[0171] Step C7, the model parameters of the second prediction model are solidified, and the model training of the second prediction model is confirmed to be completed.

[0172] Figure 2A module structure diagram of a processing device for N stage classification prediction based on CT images is provided for the second embodiment of the present application. The device is a terminal device or a server for implementing the foregoing method embodiments, or a device capable of enabling the foregoing terminal device or server to implement the foregoing method embodiments, such as a device or a chip system of the foregoing terminal device or server. As shown in Figure 2 the device includes a data receiving module 201, a semantic segmentation module 202, a first prediction model processing module 203, a second prediction model processing module 204, and a prediction output module 205.

[0173] The data receiving module 201 is configured to receive a first CT image and a first prediction mode. The first prediction mode includes a first mode and a second mode.

[0174] The semantic segmentation module 202 is configured to perform semantic segmentation processing on a gastric tumor lesion area point and a perigastric lymph node area point of the first CT image based on a first semantic segmentation model to obtain a corresponding first semantic segmentation map.

[0175] The first prediction model processing module 203 is configured to, when the first prediction mode is the first mode, perform feature extraction processing on an imageomics feature of the gastric tumor lesion area and an imageomics feature of all perigastric lymph node areas according to the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set; and perform N stage classification prediction processing based on a first prediction model according to the first lesion feature set and the first lymph feature set to obtain a corresponding first prediction type. The first prediction type includes an N0 type, an N1 type, an N2 type, an N3a type, and an N3b type.

[0176] The second prediction model processing module 204 is configured to, when the first prediction mode is the second mode, perform gastric tumor lesion and perigastric lymph node area graph synthesis processing according to the first semantic segmentation map and the first CT image to obtain a corresponding first synthesis graph; and perform N stage classification prediction processing based on a second prediction model according to the first synthesis graph to obtain a corresponding first prediction type.

[0177] The prediction output module 205 is configured to output the obtained first prediction type as a prediction result corresponding to the first CT image.

[0178] The processing device for N stage classification prediction based on CT images provided by the embodiment of the present application can execute the method steps in the foregoing method embodiments, and has similar implementation principles and technical effects, which will not be described here.

[0179] It should be noted that the division of the various modules of the above apparatus is only a logical functional division, and all or part of them can be integrated into one physical entity or physically separated when actually implemented. These modules can all be implemented in the form of software invoked by a processing element; all can be implemented in the form of hardware; or some modules can be implemented in the form of software invoked by a processing element, and some modules can be implemented in the form of hardware. For example, the data receiving module can be a separately established processing element, or can be implemented in a certain chip of the above apparatus, in addition, it can also be stored in the form of program code in the memory of the above apparatus, and the function of the above determining module is invoked and executed by a certain processing element of the above apparatus. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together or independently implemented. The processing element described herein can be an integrated circuit with signal processing capability. In the implementation process, each step of the above method or each module can be completed by the integrated logic circuit of the hardware in the processor element or the instruction in the form of software.

[0180] For example, the above modules can be one or more integrated circuits configured to implement the above method, 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), etc. For another example, when a certain module above is implemented in the form of program code invoked by a processing element, the processing element can be a general purpose processor, such as a central processing unit (CPU) or other processor that can invoke program code. For another example, these modules can be integrated together to implement in the form of system on a chip (SOC).

[0181] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The aforementioned computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the aforementioned computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The aforementioned computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0182] Figure 3 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be a terminal device or server implementing the methods of the aforementioned embodiments, or it can be a terminal device or server connected to the aforementioned terminal device or server implementing the methods of the aforementioned embodiments. Figure 3 As shown, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for performing various processing functions and implementing the processing steps described in the foregoing embodiments. Preferably, the electronic device involved in the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0183] exist Figure 3The system bus 305 mentioned above 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 the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as a client, a read-write library and a read-only library). The memory can contain a Random Access Memory (RAM), and can also include a Non-Volatile Memory, such as at least one disk memory.

[0184] The processor mentioned above can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a Graphics Processing Unit (GPU), etc.; 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.

[0185] It should be noted that the embodiments of the present application also provide a computer readable storage medium, which stores instructions, and when the instructions run on a computer, the computer executes the method and processing procedure provided in the above embodiments.

[0186] The embodiments of the present application also provide a chip for running instructions, which is used to execute the processing steps described in the foregoing method embodiments.

[0187] The embodiment of the present application provides a processing method and device for N stage classification prediction based on CT images, electronic equipment and computer readable storage medium. According to the above content, the embodiment of the present application first performs semantic segmentation processing on the received CT image based on a semantic segmentation model to obtain a corresponding semantic segmentation graph of the gastric tumor lesion area point and the perigastric lymph node area point; then two optional prediction modes are provided for prediction: 1) in the first mode state, first, the image features of the gastric tumor lesion area and the image features of all perigastric lymph node areas are extracted according to the semantic segmentation graph and the CT image, and then a first prediction model based on a machine learning classification model is used to perform N stage classification prediction according to the extracted first lesion feature set and the first lymph feature set to obtain a corresponding prediction type (N0 type, N1 type, N2 type, N3a type and N3b type); 2) in the second mode state, first, a corresponding synthesis graph is obtained by synthesizing the gastric tumor lesion and the perigastric lymph node area graph according to the semantic segmentation graph and the CT image, and then a second prediction model based on a deep learning classification model is used to perform N stage classification prediction according to the synthesis graph to obtain a prediction type (N0 type, N1 type, N2 type, N3a type and N3b type). Through the embodiment of the present application, the prediction real-time performance and the prediction efficiency are improved.

[0188] Those skilled in the art should further appreciate that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, each example has been described in general terms above as including functional components and steps. Whether such functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0189] The steps of the method or algorithm described in connection with the embodiments disclosed herein can be implemented in hardware, software executed by a processor, or a combination of both. The software module can 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.

[0190] The above specific implementation is further detailed for the purpose, technical solution and beneficial effects of the present application. It should be understood that the above description is only a specific implementation of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A processing method for N-stage classification prediction based on CT images, characterized in that, The method comprises: receiving a first CT image and a first prediction mode; the first prediction mode comprises a first mode and a second mode; performing semantic segmentation on the first CT image based on a first semantic segmentation model to obtain a corresponding first semantic segmentation map; when the first prediction mode is the first mode, performing feature extraction on the imageomics features of the gastric tumor lesion area and the imageomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set; and performing N-stage classification prediction based on a first prediction model and the first lesion feature set and the first lymph feature set to obtain a corresponding first prediction type; the first prediction type comprises N0 type, N1 type, N2 type, N3a type and N3b type; when the first prediction mode is the second mode, performing gastric tumor lesion and perigastric lymph node area graph synthesis based on the first semantic segmentation map and the first CT image to obtain a corresponding first synthesis graph; and performing N-stage classification prediction based on a second prediction model and the first synthesis graph to obtain a corresponding first prediction type; outputting the obtained first prediction type as the prediction result corresponding to the first CT image; wherein the first CT image is a three-dimensional abdominal CT image with complete gastric and perigastric lymph node area three-dimensional information; the first semantic segmentation model comprises a first U-Net segmentation network, a second U-Net segmentation network and a first fusion module; the first and second U-Net segmentation networks are parallel; the input ends of the first and second U-Net segmentation networks are connected with the model input end of the first semantic segmentation model, and the output ends are connected with the input ends of the first fusion module; the first and second U-Net segmentation networks are both based on the model structure of the U-Net model; the first prediction model is based on a machine learning classification model; the types of the machine learning classification model at least include a classification prediction model based on an RF model, a classification prediction model based on an SVM model, a classification prediction model based on a GBDT model and a classification prediction model based on a GBDT+LR model; the second prediction model is based on a deep learning classification model; the types of the deep learning classification model at least include a classification prediction model based on a ResNet model and a classification prediction model based on a Transformer model.

2. The processing method for N-stage classification prediction based on a CT image according to claim 1, characterized in that, The shape of the first CT image is D0xH0xW0xC0, D0, H0, and W0 are the depth, height, and width of the three-dimensional size of the first CT image respectively, C0 is the feature channel dimension of the first CT image, and C0=1; the first CT image is composed of D0xH0xW0 first voxel points; each first voxel point corresponds to a first voxel point feature vector with a length of C0; the first voxel point feature vector includes a first voxel point pixel value; The shape of the first semantic segmentation map is D1xH1xW1xC1, D1, H1, and W1 are the depth, height, and width of the three-dimensional size of the first semantic segmentation map respectively, C1 is the feature channel dimension of the first semantic segmentation map, D1=D0, H1=H0, W1=W0, and C1=3; the first semantic segmentation map is composed of D1xH1xW1 second voxel points; each second voxel point corresponds to a second voxel point feature vector with a length of C1; the second voxel point feature vector includes a first lesion point semantic, a first lymph point semantic, and a first background point semantic; The first lesion point semantic, the first lymph point semantic, and the first background point semantic are all binary semantics, taking values of 0 or 1, and are true when taking a value of 1 and false when taking a value of 0; among the first lesion point semantic, the first lymph point semantic, and the first background point semantic of each second voxel point feature vector, only one has a value of 1; The shape of the first synthetic map is D2xH2xW2xC2, D2, H2, and W2 are the depth, height, and width of the three-dimensional size of the first synthetic map respectively, C2 is the feature channel dimension of the first synthetic map, D2<D0, H2<H0, W2<W0, and C2=C0+C1; the first synthetic map is composed of D2xH2xW2 third voxel points; each third voxel point corresponds to a third voxel point feature vector with a length of C2; the third voxel point feature vector includes a second voxel point pixel value, a second lesion point semantic, a second lymph point semantic, and a second background point semantic; the second lesion point semantic, the second lymph point semantic, and the second background point semantic are all binary semantics, taking values of 0 or 1, and are true when taking a value of 1 and false when taking a value of 0; among the second lesion point semantic, the second lymph point semantic, and the second background point semantic of each third voxel point feature vector, only one has a value of 1; The first lesion feature set comprises a first-order feature set of a lesion region, a shape feature set of the lesion region, and a texture feature set of the lesion region; the first-order feature set of the lesion region comprises a median value of gray scale of the lesion region, an average value of gray scale of the lesion region, a minimum value of gray scale of the lesion region, a maximum value of gray scale of the lesion region, a standard deviation of gray scale of the lesion region, a skewness of gray scale of the lesion region, and a peak value of gray scale distribution of the lesion region; the shape feature set of the lesion region comprises a surface area of the lesion region, a volume of the lesion region, a volume ratio of the lesion region, a sphericity of the lesion region, a compactness type of the lesion region, and a diameter of the lesion region; and the texture feature set of the lesion region comprises a gray scale co-occurrence matrix of the lesion region, a gray scale run length matrix of the lesion region, a gray scale size region matrix of the lesion region, a neighborhood gray scale difference matrix of the lesion region, and a gray scale correlation matrix of the lesion region. The first lymph node feature set comprises a plurality of first lymph node region feature sets; each of the first lymph node region feature sets corresponds to a perigastric lymph node region. The first lymph node region feature set comprises a first lymph node region distance, a first-order feature set of the first lymph node region, a shape feature set of the first lymph node region, and a texture feature set of the first lymph node region; the first lymph node region distance is a straight-line distance from a center point of the corresponding perigastric lymph node region to a center point of a gastric tumor lesion region. The first-order feature set of the first lymph node region comprises a median value of gray scale of the lymph node region, an average value of gray scale of the lymph node region, a minimum value of gray scale of the lymph node region, a maximum value of gray scale of the lymph node region, a standard deviation of gray scale of the lymph node region, a skewness of gray scale of the lymph node region, and a peak value of gray scale distribution of the lymph node region; the shape feature set of the first lymph node region comprises a surface area of the lymph node, a volume of the lymph node, a volume ratio of the lymph node, a sphericity of the lymph node, a compactness type of the lymph node, and a diameter of the lymph node; and the texture feature set of the first lymph node region comprises a gray scale co-occurrence matrix of the lymph node region, a gray scale run length matrix of the lymph node region, a gray scale size region matrix of the lymph node region, a neighborhood gray scale difference matrix of the lymph node region, and a gray scale correlation matrix of the lymph node region. 3.The processing method for N-stage classification prediction based on CT images according to claim 2, characterized in that, The first CT image is subjected to gastric tumor lesion region point and perigastric lymph node region point semantic segmentation processing based on the first semantic segmentation model to obtain a corresponding first semantic segmentation image, specifically comprising: The first CT image is input into the first U-Net segmentation network; the first U-Net segmentation network is used for performing lesion point semantic segmentation on the voxel points of the gastric tumor lesion region on the first CT image to obtain a corresponding first one-segment semantic map; the shape of the first one-segment semantic map is D 11 ×H 11 ×W 11 ×C 11 , D 11 , H 11 , W 11 are respectively the depth, height and width of the three-dimensional size of the first one-segment semantic map, C 11 is the feature channel dimension of the first one-segment semantic map, D 11 =D0, H 11 =H0, W 11 =W0, C 11 =2; the first one-segment semantic map is composed of D 11 ×H 11 ×W 11 fourth voxel points; each fourth voxel point corresponds to a fourth voxel point feature vector with a length of C 11 ; the fourth voxel point feature vector includes a third lesion point semantic and a third background point semantic; the third lesion point semantic and the third background point semantic are both binary semantics, and the values are 0 or 1, and the value 1 is true and the value 0 is false; only one of the third lesion point semantic and the third background point semantic of each fourth voxel point feature vector is 1; and input the first CT image into the second U-Net segmentation network; perform lymph node semantic segmentation on all voxel points of the perigastric lymph node region on the first CT image by the second U-Net segmentation network to obtain a corresponding first two segmentation semantic graph; the shape of the first two segmentation semantic graph is D 12 x H 12 x W 12 x C 12 , D 12 , H 12 , W 12 respectively are the depth, height and width of the three-dimensional size of the first two segmentation semantic graph, C 12 is the feature channel dimension of the first two segmentation semantic graph, D 12 =D0, H 12 =H0, W 12 =W0, C 12 =2; the first two segmentation semantic graph is composed of D 12 x H 12 x W 12 fifth voxel points; each fifth voxel point corresponds to a fifth voxel point feature vector with a length of C 12 ; the fifth voxel point feature vector includes a third lymph node semantic and a fourth background point semantic; the third lymph node semantic and the fourth background point semantic are both binary semantics, taking values of 0 or 1, and taking a value of 1 as true and a value of 0 as false; only one of the third lymph node semantic and the fourth background point semantic of each fifth voxel point feature vector is 1; and the first two segmentation semantic maps are input into the first fusion module; the first semantic segmentation map is initialized as a full zero semantic by the first fusion module; all the second voxel points on the first semantic segmentation map are traversed; during the traversal, the current traversed second voxel point is taken as a corresponding current voxel point, and the fourth voxel point and the fifth voxel point corresponding to the current voxel point on the first one segmentation semantic map and the first two segmentation semantic maps are recorded as first and second corresponding voxel points; when the third lesion point semantic of the first corresponding voxel point is 1, the first lesion point semantic of the current voxel point is reset to 1; when the third lymph node semantic of the second corresponding voxel point is 1, the first lymph node semantic of the current voxel point is reset to 1; when the third and fourth background point semantics of the first and second corresponding voxel points are both 1, the first background point semantic of the current voxel point is reset to 1; and when the traversal ends, the first semantic segmentation map after the reset is taken as the processing result of the first semantic segmentation model and is output. 4.The processing method for N-stage classification prediction based on CT images according to claim 2, wherein, The image feature extraction processing is performed on the image features of the gastric tumor lesion region and the image features of all the perigastric lymph node regions according to the first semantic segmentation map and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set, and the image feature extraction processing specifically includes: The image feature extraction processing is performed on the image features of the gastric tumor lesion region according to the first semantic segmentation map and the first CT image to obtain the first lesion feature set. The image feature extraction processing is performed on the image features of all the perigastric lymph node regions according to the first semantic segmentation map and the first CT image to obtain the first lymph feature set. 5.The processing method for N-stage classification prediction based on CT images according to claim 4, characterized in that, The image feature extraction processing is performed on the image features of the gastric tumor lesion region according to the first semantic segmentation map and the first CT image to obtain the first lesion feature set, and the image feature extraction processing specifically includes: The second voxel points on the first semantic segmentation map, each of which has a first lesion point semantic of 1, are recorded as corresponding first lesion voxel points; and the first voxel points on the first CT image, each of which corresponds to a first lesion voxel point, are recorded as corresponding second lesion voxel points; The maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value and the minimum width value in all the second lesion voxel points are counted to obtain a first maximum depth value, a first minimum depth value, a first maximum height value, a first minimum height value, a first maximum width value and a first minimum width value; and a three-dimensional CT subgraph with a depth range between the first maximum depth value and the first minimum depth value, a height range between the first maximum height value and the first minimum height value, and a width range between the first maximum width value and the first minimum width value is cut out from the first CT image as a corresponding first lesion region subgraph; And the first lesion area subgraph is converted into a gray scale graph to obtain a corresponding first lesion area gray scale graph; And the gray scale median value, the gray scale average value, the gray scale minimum value, the gray scale maximum value, the gray scale standard deviation, the gray scale skewness and the gray scale distribution peak value of the first lesion area gray scale graph are calculated to obtain the lesion area gray scale median value, the lesion area gray scale average value, the lesion area gray scale minimum value, the lesion area gray scale maximum value, the lesion area gray scale standard deviation, the lesion area gray scale skewness and the lesion area gray scale distribution peak value corresponding to the first lesion area gray scale graph. And the three-dimensional structure composed of all the second lesion voxel points on the first lesion area gray scale graph is recorded as a corresponding first tumor structure; and the surface area, volume, volume ratio, sphericity, compactness type and diameter of the first tumor structure are calculated to obtain the lesion area surface area, the lesion area volume, the lesion area volume ratio, the lesion area sphericity, the lesion area compactness type and the lesion area diameter, which constitute the shape feature set of the corresponding lesion area. And the gray scale co-occurrence matrix, gray scale run length matrix, gray scale size region matrix, neighborhood gray scale difference matrix and gray scale correlation matrix of the first lesion area gray scale graph are calculated to obtain the lesion area gray scale co-occurrence matrix, the lesion area gray scale run length matrix, the lesion area gray scale size region matrix, the lesion area neighborhood gray scale difference matrix and the lesion area gray scale correlation matrix, which constitute the texture feature set of the corresponding lesion area. And the first lesion feature set is composed of the obtained lesion area first-order feature set, the lesion area shape feature set and the lesion area texture feature set. 6.The processing method for N-stage classification prediction based on CT images according to claim 5, wherein, The image feature extraction processing of the image feature set of all perigastric lymph nodes is performed according to the first semantic segmentation graph and the first CT image to obtain a corresponding first lymph feature set, which specifically includes: The second voxel points with the first lymph point semantic value of 1 on the first semantic segmentation graph are recorded as corresponding first lymph voxel points; and the first voxel points corresponding to each first lymph voxel point on the first CT image are recorded as corresponding second lymph voxel points; The second voxel points with the first lymph point semantic value of 1 on the first semantic segmentation graph are recorded as corresponding first lymph voxel points; and the first voxel points corresponding to each first lymph voxel point on the first CT image are recorded as corresponding second lymph voxel points; And the point cluster of all the second lymphatic points is obtained to obtain a plurality of first point group sets; and each first point group set is taken as a corresponding current point group set; and the maximum depth value, the minimum depth value, the maximum height value, the minimum height value, the maximum width value and the minimum width value of all the second lymphatic points in the current point group set are counted to obtain a corresponding second maximum depth value, a second minimum depth value, a second maximum height value, a second minimum height value, a second maximum width value and a second minimum width value; and the three-dimensional CT subgraph of the first CT image with a depth range between the second maximum depth value and the second minimum depth value, a height range between the second maximum height value and the second minimum height value, and a width range between the second maximum width value and the second minimum width value is cut out as a corresponding first perigastric lymph node region subgraph; the first point group set includes a plurality of second lymphatic points; And the straight line distance from the center point of each first perigastric lymph node region subgraph to the center point of the first lesion region subgraph is calculated to obtain a corresponding first lymph node region distance; And the gray scale graph conversion of each first perigastric lymph node region subgraph is performed to obtain a corresponding first lymph node region gray scale graph; And the gray scale median value, the gray scale average value, the gray scale minimum value, the gray scale maximum value, the gray scale standard deviation, the gray scale skewness and the gray scale distribution peak value of each first lymph node region gray scale graph are calculated to obtain the lymph node region gray scale median value, the lymph node region gray scale average value, the lymph node region gray scale minimum value, the lymph node region gray scale maximum value, the lymph node region gray scale standard deviation, the lymph node region gray scale skewness and the lymph node region gray scale distribution peak value, which form a corresponding first lymph node region first-order feature set; And a three-dimensional structure composed of all the second lymphatic points on each first lymph node region gray scale graph is recorded as a corresponding first lymph structure; and the surface area, the volume, the volume ratio, the sphericity, the compactness type and the diameter of the first lymph structure are calculated to obtain the lymph node surface area, the lymph node volume, the lymph node volume ratio, the lymph node sphericity, the lymph node compactness type and the lymph node diameter, which form a corresponding first lymph node region shape feature set; And the gray scale co-occurrence matrix, the gray scale run length matrix, the gray scale size region matrix, the neighborhood gray scale difference matrix and the gray scale correlation matrix of each first lymph node region gray scale graph are calculated to obtain the lymph node region gray scale co-occurrence matrix, the lymph node region gray scale run length matrix, the lymph node region gray scale size region matrix, the lymph node region neighborhood gray scale difference matrix and the lymph node region gray scale correlation matrix, which form a corresponding first lymph node region texture feature set; And each of the first perigastric lymph node region subgraph corresponds to the first lymph node region distance, the first lymph node region first feature set, the first lymph node region shape feature set and the first lymph node region texture feature set to form a corresponding first lymph node region feature set; and all the first lymph node region feature sets obtained form a corresponding first lymph node feature set.

7. The processing method for N-stage classification prediction based on CT images according to claim 2, wherein, The first prediction model is based on the first lesion feature set and the first lymph node feature set to perform N stage classification prediction processing to obtain a corresponding first prediction type, specifically including: The first lesion feature set and each of the first lymph node region feature set of the first lymph node feature set form a corresponding first lesion-lymph node feature set; and all the first lesion-lymph node feature sets are sorted in order of the first lymph node region distance from near to far in each of the first lymph node region feature set in the first lesion-lymph node feature set to form a corresponding first feature set sequence; The first feature set sequence is input into the first prediction model; the first prediction model identifies the metastasis characteristics of gastric cancer tumor metastasis to all perigastric lymph node regions according to the imaging features of the first feature set sequence, and based on the metastasis characteristic identification result, five N stage types of gastric cancer tumor are classified and predicted to obtain a corresponding first prediction vector; the first prediction vector includes five first prediction probabilities; each first prediction probability corresponds to one N stage type; the N stage type includes N0 type, N1 type, N2 type, N3a type and N3b type; The N stage type corresponding to the first prediction probability with the maximum probability value in the first prediction vector is taken as the corresponding first prediction type. 8.The processing method for N-stage classification prediction based on CT images according to claim 2, wherein, The first semantic segmentation map and the first CT image are combined to obtain a corresponding first synthesis map, specifically including: The second voxel point with each first lesion point semantic or first lymph point semantic of 1 on the first semantic segmentation map is recorded as a corresponding first voxel point; and the first voxel point corresponding to each first voxel point on the first CT image is recorded as a corresponding second voxel point; And the maximum depth value, minimum depth value, maximum height value, minimum height value, maximum width value and minimum width value of all the second voxel points are counted to obtain a corresponding third maximum depth value, third minimum depth value, third maximum height value, third minimum height value, third maximum width value and third minimum width value; And the first semantic segmentation map and the first CT image are cut out as a corresponding first segmentation subgraph and first CT subgraph in the depth range between the third maximum depth value and the third minimum depth value, the height range between the third maximum height value and the third minimum height value, and the width range between the third maximum width value and the third minimum width value; the three-dimensional size of the first segmentation subgraph and the first CT subgraph is consistent; and the depth, height and width of the three-dimensional size of the first CT subgraph are taken as corresponding depth D2, height H2 and width W2; and the sum of the feature channel dimension C0 of the first CT subgraph and the feature channel dimension C1 of the first segmentation subgraph is taken as corresponding feature channel dimension C2; and the first segmentation subgraph and the first CT subgraph are spliced in a voxel point feature vector splicing manner to obtain the first synthesis graph with a shape of D2×H2×W2×C2. 9.The processing method for N-stage classification prediction based on CT images of claim 2, wherein, The N-stage classification prediction processing based on the second prediction model according to the first synthesis graph obtains the corresponding first prediction type, and specifically includes: The first synthesis graph is input into the second prediction model; the second prediction model identifies the metastasis characteristics of the gastric cancer tumor to all perigastric lymph node regions according to the image features of the first synthesis graph, and classifies and predicts the five N-stage types of the gastric cancer tumor based on the metastasis characteristic identification result to obtain a corresponding second prediction vector; the second prediction vector includes five second prediction probabilities; each second prediction probability corresponds to one N-stage type; the N-stage type includes N0 type, N1 type, N2 type, N3a type and N3b type; The N-stage type corresponding to the second prediction probability with the maximum probability value in the second prediction vector is taken as the corresponding first prediction type.

10. An apparatus for performing the processing method of predicting N-stage classification based on CT images according to any one of claims 1-9, characterized in that, The device includes a data receiving module, a semantic segmentation module, a first prediction model processing module, a second prediction model processing module and a prediction output module. The data receiving module is used to receive a first CT image and a first prediction mode; the first prediction mode includes a first mode and a second mode; The semantic segmentation module is used to perform gastric tumor lesion region point and perigastric lymph node region point semantic segmentation processing on the first CT image based on a first semantic segmentation model to obtain a corresponding first semantic segmentation graph; The first prediction model processing module is used to, when the first prediction mode is the first mode, perform feature extraction processing on the imageomics characteristics of the gastric tumor lesion region and the imageomics characteristics of all perigastric lymph node regions according to the first semantic segmentation graph and the first CT image to obtain a corresponding first lesion feature set and a first lymph feature set; and perform N-stage classification prediction processing based on a first prediction model according to the first lesion feature set and the first lymph feature set to obtain a corresponding first prediction type; the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type; The second prediction model processing module is used to, when the first prediction mode is the second mode, perform gastric tumor lesion and perigastric lymph node region graph synthesis processing according to the first semantic segmentation graph and the first CT image to obtain a corresponding first synthesis graph; and perform N-stage classification prediction processing based on a second prediction model according to the first synthesis graph to obtain a corresponding first prediction type; The prediction output module is used to output the obtained first prediction type as a prediction result corresponding to the first CT image.

11. An electronic device, comprising: It includes: a memory, a processor and a transceiver; The processor is coupled with the memory to read and execute instructions in the memory to implement the method of any one of claims 1-9. The transceiver is coupled with the processor to be controlled by the processor to perform message transceiving.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, when the computer instructions are executed by a computer, cause the computer to perform the method of any one of claims 1-9.

Images