N-stage classification prediction processing method and apparatus based on CT image

Abstract

The embodiments of the present invention relate to an N-stage classification prediction processing method and apparatus based on a CT image. The method comprises: receiving a CT image and a prediction mode; on the basis of a first semantic segmentation model, performing semantic segmentation on the CT image regarding gastric tumor lesion region points and perigastric lymph node region points; when the prediction mode is a first mode, on the basis of a semantic segmentation map and the CT image, extracting radiomics features of a gastric tumor lesion region and all perigastric lymph node regions, so as to obtain a lesion feature set and a lymphatic feature set, and performing N-stage classification prediction on the basis of a first prediction model and the lesion feature set and the lymphatic feature set; when the prediction mode is a second mode, on the basis of the semantic segmentation map and the CT image, synthesizing an image by the gastric tumor lesion region and the perigastric lymph node regions, and performing N-stage classification prediction on the basis of a second prediction model and the synthesized image; and outputting a corresponding predicted type. By means of the present invention, the real-time performance and efficiency of prediction can be improved.

Description

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

[0001] This application claims priority to Chinese Patent Application No. 202410515491.7, filed on April 26, 2024, entitled "A Processing Method and Apparatus for N-Stage Classification Prediction Based on CT Images". Technical Field

[0002] This invention relates to the field of data processing technology, and in particular to a processing method and apparatus for N-stage classification prediction based on CT images. Background Technology

[0003] The N staging system is a gastric cancer staging system that classifies gastric tumors based on their metastasis to perigastric lymph nodes. Common N staging types include N0, N1, N2, N3a, and N3b. Currently, medical experts can manually identify N staging types based on computed tomography (CT) images. However, this manual identification method is clearly limited by the time constraints of expert work, resulting in significant time-sensitivity issues, especially when processing large batches of CT images. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a processing method, apparatus, electronic device, and computer-readable storage medium for N-stage classification prediction based on CT images. This invention first performs semantic segmentation on received CT images using a semantic segmentation model to obtain corresponding semantic segmentation maps for the gastric tumor lesion region and perigastric lymph node region. Then, it provides two optional prediction modes: 1) In the first mode, feature extraction is performed on the radiomics features of the gastric tumor lesion region and all perigastric lymph node regions based on the semantic segmentation map and CT images. A first prediction model based on a machine learning classification model then performs N-stage classification prediction based on the extracted first lesion feature set and first lymph node feature set to obtain the corresponding prediction type (N0 type, N1 type, N2 type, N3a type, and N3b type); 2) In the second mode, a composite map of the gastric tumor lesion and perigastric lymph node regions is synthesized based on the semantic segmentation map and CT images. A second prediction model based on a deep learning classification model then performs N-stage classification prediction based on the composite map to obtain the prediction type (N0 type, N1 type, N2 type, N3a type, and N3b type). This invention improves the real-time performance and efficiency of prediction.

[0005] To achieve the above objectives, a first aspect of the present invention provides a method for N-stage classification prediction based on CT images, the method comprising:

[0006] Receive a first CT image and a first prediction pattern; the first prediction pattern includes a first pattern and a second pattern;

[0007] Based on the first semantic segmentation model, the first CT image is semantically segmented into the gastric tumor lesion area and the perigastric lymph node area to obtain the corresponding first semantic segmentation map.

[0008] When the first prediction mode is the first mode, feature extraction processing is performed on the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set and first lymph node feature set; and based on the first prediction model, N-stage classification prediction processing is performed on the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type; the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type;

[0009] When the first prediction mode is the second mode, the first composite image is obtained by synthesizing the gastric tumor lesion and perigastric lymph node region map based on the first semantic segmentation map and the first CT image; and the first prediction type is obtained by performing N-stage classification prediction based on the first composite image using the second prediction model.

[0010] The obtained first prediction type is output as the prediction result corresponding to the first CT image.

[0011] Preferably, the first CT image is a three-dimensional abdominal CT image with complete three-dimensional information of the stomach and all perigastric lymph node regions; the shape of the first CT image is D0×H0×W0×C0, where D0, H0, and W0 are the depth, height, and width of the three-dimensional dimensions of the first CT image, respectively, and C0 is the feature channel dimension of the first CT image, C0 = 1; the first CT image is composed of D0×H0×W0 first voxel points; each first voxel point corresponds to a first voxel point feature vector of length C0; the first voxel point feature vector includes the first voxel point pixel value;

[0012] The shape of the first semantic segmentation map is D1×H1×W1×C1, where 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 D1×H1×W1 second voxel points; each of the second voxel points corresponds to a second voxel point feature vector with a length of C1; the second voxel point feature vector includes the first lesion point semantics, the first lymph node point semantics, and the first background point semantics; the first lesion point semantics, the first lymph node point semantics, and the first background point semantics are all binary semantics, taking values of 0 or 1, being true when the value is 1 and false when the value is 0; among the first lesion point semantics, the first lymph node point semantics, and the first background point semantics of each second voxel point feature vector, only one is 1;

[0013] The shape of the first composite map is D2×H2×W2×C2, where D2, H2, and W2 are the depth, height, and width of the three-dimensional size of the first composite map respectively, C2 is the feature channel dimension of the first composite map, D2 < D0, H2 < H0, W2 < W0, and C2 = C0 + C1; the first composite map is composed of D2×H2×W2 third voxel points; each of the third voxel points corresponds to a third voxel point feature vector with a length of C2; the third voxel point feature vector includes the second voxel point pixel value, the second lesion point semantics, the second lymph node point semantics, and the second background point semantics; the second lesion point semantics, the second lymph node point semantics, and the second background point semantics are all binary semantics, taking values of 0 or 1, being true when the value is 1 and false when the value is 0; among the second lesion point semantics, the second lymph node point semantics, and the second background point semantics of each third voxel point feature vector, only one is 1;

[0014] The first lesion feature set includes a first-order lesion area feature set, a lesion area shape feature set, and a lesion area texture feature set; the first-order lesion area feature set includes the median gray value of the lesion area, the average gray value of the lesion area, the minimum gray value of the lesion area, the maximum gray value of the lesion area, the standard deviation of the gray value of the lesion area, the skewness of the gray value of the lesion area, and the peak value of the gray value distribution of the lesion area; the lesion area shape feature set includes the surface area of the lesion area, the volume of the lesion area, the volume ratio of the lesion area, the sphericity of the lesion area, the compactness type of the lesion area, and the diameter of the lesion area; the lesion area texture feature set includes the gray-level co-occurrence matrix of the lesion area, the gray-level run-length matrix of the lesion area, the gray-level size-zone matrix of the lesion area, the neighborhood gray-level difference matrix of the lesion area, and the gray-level correlation matrix of the lesion area;

[0015] The first lymph node feature set includes multiple first lymph node region feature sets; each first lymph node region feature set corresponds to a perigastric lymph node region; the first lymph node region feature set includes 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; the first lymph node region 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 region first-order feature set includes the median gray value of the lymph node region, the average gray value of the lymph node region, and so on. The first lymph node region features include: minimum gray value, maximum gray value, standard deviation, skewness, and peak gray value distribution of the lymph node region; the first lymph node region shape feature set includes: lymph node surface area, lymph node volume, lymph node volume ratio, lymph node sphericity, lymph node compactness type, and lymph node diameter; the first lymph node region texture feature set includes: lymph node region gray-level co-occurrence matrix, lymph node region gray-level run length matrix, lymph node region gray-level size region matrix, lymph node region neighborhood gray-level difference matrix, and lymph node region gray-level correlation matrix.

[0016] Preferably, 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 implemented in parallel; the input terminals of the first and second U-Net segmentation networks are both connected to the model input terminal of the first semantic segmentation model, and the output terminals are both connected to the input terminal of the first fusion module; the first and second U-Net segmentation networks are both implemented based on the model structure of the U-Net model.

[0017] The first prediction model is implemented based on a machine learning classification model; the types of machine learning classification models include at least a classification prediction model based on the RF model, a classification prediction model based on the SVM model, a classification prediction model based on the GBDT model, and a classification prediction model based on the GBDT+LR model.

[0018] The second prediction model is implemented based on a deep learning classification model, and the types of deep learning classification models include at least classification prediction models implemented based on ResNet models and classification prediction models implemented based on Transformer models.

[0019] Preferably, the step of performing semantic segmentation processing on the first CT image based on the first semantic segmentation model to obtain the corresponding first semantic segmentation map includes:

[0020] The first CT image is input into the first U-Net segmentation network; the first U-Net segmentation network performs semantic segmentation of the voxel points of the gastric tumor lesion region on the first CT image to obtain the corresponding first segmentation semantic map; the shape of the first segmentation semantic map is D. 11 ×H 11 ×W 11 ×C 11 D 11 H 11 W 11 C represents the depth, height, and width of the three-dimensional dimensions of the first segmented semantic graph, respectively. 11 Let D be the feature channel dimension of the first segmentation semantic graph. 11 =D0, H 11 =H0, W 11 =W0,C 11 =2; the first segmentation semantic graph is composed of D 11 ×H 11 ×W 11 It consists of four fourth voxel points; each of the fourth voxel points corresponds to a length of C. 11 The fourth voxel feature vector includes the semantics of the third lesion point and the semantics of the third background point. The semantics of the third lesion point and the semantics of the third background point are both binary semantics, with a value of 0 or 1. A value of 1 indicates true and a value of 0 indicates false. In each fourth voxel feature vector, only one of the semantics of the third lesion point and the semantics of the third background point is 1.

[0021] The first CT image is input into the second U-Net segmentation network; the second U-Net segmentation network performs lymph node semantic segmentation on all voxel points of the perigastric lymph node region on the first CT image to obtain the corresponding first binary segmentation semantic map; the shape of the first binary segmentation semantic map is D. 12 ×H 12 ×W 12 ×C 12 D 12 H 12 W 12 C represents the depth, height, and width of the three-dimensional dimensions of the first two segmented semantic graphs, respectively. 12 D represents the feature channel dimension of the first and second segmentation semantic graphs. 12 =D0, H 12 =H0, W 12 =W0,C 12 =2; the first binary segmentation semantic graph is composed of D 12 ×H 12 ×W 12It consists of five fifth voxel points; each fifth voxel point corresponds to a length of C. 12 The fifth voxel feature vector includes the semantics of the third lymph node and the semantics of the fourth background node; the semantics of the third lymph node and the semantics of the fourth background node are both binary semantics, taking values ​​of 0 or 1, where a value of 1 is true and a value of 0 is false; for each fifth voxel feature vector, only one of the semantics of the third lymph node and the semantics of the fourth background node is 1.

[0022] The first semantic segmentation graph and the first second semantic segmentation graph are input into the first fusion module. The first fusion module initializes a first semantic segmentation graph with all zero semantics. All second voxel points on the first semantic segmentation graph are traversed. During the traversal, the currently traversed second voxel point is taken as the corresponding current voxel point, and the fourth voxel point and the fifth voxel point corresponding to the current voxel point on the first and first second semantic segmentation graphs are recorded as the corresponding first and second corresponding voxel points. When the semantics of the third lesion point of the first corresponding voxel point is 1, the semantics of the first lesion point of the current voxel point is reset to 1. When the semantics of the third lymph point of the second corresponding voxel point is 1, the semantics of the first lymph point of the current voxel point is reset to 1. When the semantics of the third and fourth background points of the first and second corresponding voxel points are both 1, the semantics of the first background point of the current voxel point is reset to 1. At the end of the traversal, the reset first semantic segmentation graph is output as the processing result of the first semantic segmentation model.

[0023] Preferably, the step of extracting features from the radiomics features of the gastric tumor lesion region and the radiomics features of all perigastric lymph node regions based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set and first lymph node feature set specifically includes:

[0024] Based on the first semantic segmentation map and the first CT image, feature extraction processing is performed on the radiomics features of the gastric tumor lesion region to obtain the corresponding first lesion feature set;

[0025] Based on the first semantic segmentation map and the first CT image, feature extraction processing is performed on the radiomics features of all perigastric lymph node regions to obtain the corresponding first lymph node feature set.

[0026] Furthermore, the step of extracting features from the radiomics features of the gastric tumor lesion region based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set specifically includes:

[0027] The second voxel points with a semantic value of 1 for each of the first lesion points on the first semantic segmentation map are recorded as the corresponding first lesion voxel points; and the first voxel points on the first CT image corresponding to each of the first lesion voxel points are recorded as the corresponding second lesion voxel points.

[0028] The maximum depth, minimum depth, maximum height, minimum height, maximum width, and minimum width values ​​of all second lesion voxels are statistically analyzed to obtain the corresponding first maximum depth, first minimum depth, first maximum height, first minimum height, first maximum width, and first minimum width values. Then, a three-dimensional CT sub-image from the first CT image whose depth range is between the first maximum depth and the first minimum depth, whose height range is between the first maximum height and the first minimum height, and whose width range is between the first maximum width and the first minimum width is extracted as the corresponding first lesion area sub-image.

[0029] The first lesion area sub-image is then converted to grayscale to obtain the corresponding grayscale image of the first lesion area.

[0030] The corresponding first-order feature set of the lesion area is obtained by calculating the median, average, minimum, maximum, standard deviation, skewness, and peak value of the grayscale image of the first lesion area.

[0031] The three-dimensional structure composed of all the second lesion body pixels on the grayscale image of the first lesion area is recorded as the 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 corresponding lesion area surface area, lesion area volume, lesion area volume ratio, lesion area sphericity, lesion area compactness type and lesion area diameter, which form the corresponding lesion area shape feature set;

[0032] The gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of the first lesion area gray-level image are calculated to obtain the corresponding gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of the lesion area, which together form the corresponding texture feature set of the lesion area.

[0033] The first lesion feature set is composed of the obtained 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.

[0034] Furthermore, the step of extracting features from the radiomics features of all perigastric lymph node regions based on the first semantic segmentation map and the first CT image to obtain the corresponding first lymph node feature set specifically includes:

[0035] The second voxel points with a semantic value of 1 for each of the first lymph nodes on the first semantic segmentation map are recorded as the corresponding first lymph node voxel points; and the first voxel points on the first CT image corresponding to each of the first lymph node voxel points are recorded as the corresponding second lymph node voxel points.

[0036] The system performs point clustering on all second lymphosome pixels to obtain multiple first-class point cluster sets; and combines each first-class point cluster into a corresponding current point cluster set; then, it statistically analyzes the maximum depth value, minimum depth value, maximum height value, minimum height value, maximum width value, and minimum width value of all second lymphosome pixels in the current point cluster set to obtain corresponding second maximum depth value, second minimum depth value, second maximum height value, second minimum height value, second maximum width value, and second minimum width value; and extracts the three-dimensional CT sub-images from the first CT image whose depth range is between the second maximum depth value and the second minimum depth value, whose height range is between the second maximum height value and the second minimum height value, and whose width range is between the second maximum width value and the second minimum width value as the corresponding first perigastric lymph node region sub-image; the first-class point cluster set includes multiple second lymphosome pixels;

[0037] The distance to the first lymph node region is calculated by taking the straight-line distance from the center point of each of the first perigastric lymph node region sub-maps to the center point of the first lesion region sub-map.

[0038] The grayscale images of each of the first perigastric lymph node regions are converted to grayscale to obtain the corresponding first lymph node region grayscale images.

[0039] The corresponding first-order feature set of the first lymph node region is formed by calculating the median, average, minimum, maximum, standard deviation, skewness, and peak value of the grayscale image of each first lymph node region.

[0040] Each first lymph node region's grayscale image, composed of all second lymphocyte pixels, is denoted as a corresponding first lymph node structure. The surface area, volume, volume ratio, sphericity, compactness type, and diameter of each first lymph node 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, forming a corresponding first lymph node region shape feature set.

[0041] The corresponding gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of each first lymph node region gray-level image are calculated to form the corresponding first lymph node region texture feature set.

[0042] Each first lymph node region feature set is composed of the distance to the first lymph node region corresponding to each first perigastric lymph node region sub-map, the first-order feature set of the first lymph node region, the shape feature set of the first lymph node region, and the texture feature set of the first lymph node region; and all the obtained first lymph node region feature sets are combined to form the corresponding first lymph node feature set.

[0043] Preferably, the step of performing N-stage classification prediction based on the first prediction model according to the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type specifically includes:

[0044] The first lesion feature set and each of the first lymph node region feature sets in the first lymph node feature set are combined to form a corresponding first lesion-lymph node feature set; and all the first lesion-lymph node feature sets are sorted in order of the distance between the first lymph node region feature sets in each first lesion-lymph node feature set from near to far to form a corresponding first feature set sequence.

[0045] The first feature set sequence is input into the first prediction model; the first prediction model identifies the metastatic features of gastric cancer tumors to all perigastric lymph node regions based on the radiomics features of the first feature set sequence, and classifies and predicts five N-stage types of gastric cancer tumors based on the metastatic feature identification results to obtain the corresponding first prediction vector; the first prediction vector includes five first prediction probabilities; each first prediction probability corresponds to one of the N-stage types; the N-stage types include N0 type, N1 type, N2 type, N3a type and N3b type;

[0046] The N-period type corresponding to the first prediction probability with the highest probability value in the first prediction vector is taken as the corresponding first prediction type.

[0047] Preferably, the step of synthesizing the gastric tumor lesion and perigastric lymph node region map based on the first semantic segmentation map and the first CT image to obtain the corresponding first composite map specifically includes:

[0048] The second voxel points whose semantic meaning of each first lesion point or the first lymph node point on the first semantic segmentation map is 1 are recorded as the corresponding first type voxel points; and the first voxel points on the first CT image corresponding to each first type voxel point are recorded as the corresponding second type voxel points.

[0049] And statistical analysis is performed on the maximum depth value, minimum depth value, maximum height value, minimum height value, maximum width value, and minimum width value of all second-type voxel points to obtain the 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;

[0050] The first semantic segmentation image and the first CT image are cropped out as the corresponding first segmentation sub-image and first CT sub-image, respectively, 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; the three-dimensional dimensions of the first segmentation sub-image and the first CT sub-image are consistent.

[0051] The depth, height, and width of the three-dimensional dimensions of the first segmented sub-image or the first CT sub-image are used as the corresponding depth D2, height H2, and width W2; the sum of the feature channel dimension C0 of the first CT sub-image and the feature channel dimension C1 of the first segmented sub-image is used as the corresponding feature channel dimension C2; and the first segmented sub-image and the first CT sub-image are stitched together according to the voxel point feature vector stitching method to obtain a first composite image with the shape D2×H2×W2×C2.

[0052] Preferably, the step of performing N-period classification prediction processing based on the first composite graph using the second prediction model to obtain the corresponding first prediction type specifically includes:

[0053] The first composite image is input into the second prediction model; the second prediction model identifies the metastasis features of gastric cancer tumors to all perigastric lymph node regions based on the image features of the first composite image, and classifies and predicts five N-stage types of gastric cancer tumors based on the metastasis feature identification results to obtain corresponding second prediction vectors; the second prediction vector includes five second prediction probabilities; each second prediction probability corresponds to one of the N-stage types; the N-stage types include N0 type, N1 type, N2 type, N3a type, and N3b type;

[0054] The N-period type corresponding to the second prediction probability with the highest probability value in the second prediction vector is taken as the corresponding first prediction type.

[0055] A second aspect of the present invention provides an apparatus for implementing the processing method for N-stage classification prediction based on CT images as described in the first aspect above. The apparatus 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.

[0056] 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.

[0057] The semantic segmentation module is used to perform semantic segmentation processing on the first CT image of the gastric tumor lesion area and the perigastric lymph node area based on the first semantic segmentation model to obtain the corresponding first semantic segmentation map.

[0058] The first prediction model processing module is used to extract features from the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image when the first prediction mode is the first mode, to obtain the corresponding first lesion feature set and first lymph node feature set; and to perform N-stage classification prediction based on the first prediction model according to the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type; the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type;

[0059] The second prediction model processing module is used to perform gastric tumor lesion and perigastric lymph node region map synthesis processing based on the first semantic segmentation map and the first CT image to obtain the corresponding first composite map when the first prediction mode is the second mode; and to perform N-stage classification prediction processing based on the first composite map using the second prediction model to obtain the corresponding first prediction type.

[0060] The prediction output module is used to output the obtained first prediction type as the prediction result corresponding to the first CT image.

[0061] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0062] The processor is used to couple with the memory, read and execute instructions in the memory to implement the steps of the method described in the first aspect above;

[0063] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0064] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions described in the first aspect.

[0065] This invention provides a processing method, apparatus, electronic device, and computer-readable storage medium for N-stage classification prediction based on CT images. As described above, the embodiments of the present invention first perform semantic segmentation processing on the received CT images to obtain corresponding semantic segmentation maps by semantic segmentation model for gastric tumor lesion area points and perigastric lymph node area points; then, two optional prediction modes are provided for prediction: 1) In the first mode, feature extraction is performed on the radiomics features of the gastric tumor lesion area and all perigastric lymph node areas based on the semantic segmentation map and CT images, and then a first prediction model based on a machine learning classification model is used to perform N-stage classification prediction based on the extracted first lesion feature set and first lymph node feature set to obtain the corresponding prediction type (N0 type, N1 type, N2 type, N3a type and N3b type); 2) In the second mode, the gastric tumor lesion and perigastric lymph node area maps are synthesized based on the semantic segmentation map and CT images to obtain the corresponding composite map, and then a second prediction model based on a deep learning classification model is used to perform N-stage classification prediction based on the composite map to obtain the prediction type (N0 type, N1 type, N2 type, N3a type and N3b type). The embodiments of the present invention improve both the real-time performance and the efficiency of prediction. Attached Figure Description

[0066] Figure 1 is a schematic diagram of a processing method for N-stage classification prediction based on CT images provided in Embodiment 1 of the present invention;

[0067] Figure 2 is a module structure diagram of a processing device for N-stage classification prediction based on CT images provided in Embodiment 2 of the present invention;

[0068] Figure 3 is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0070] Embodiment 1 of the present invention provides a processing method for N-stage classification prediction based on CT images. Figure 1 is a schematic diagram of the processing method for N-stage classification prediction based on CT images provided in Embodiment 1 of the present invention. The method mainly includes the following steps:

[0071] Step 1: Receive the first CT image and the first prediction pattern;

[0072] The first CT image is a three-dimensional abdominal CT image containing three-dimensional information of the complete stomach and all perigastric lymph node regions. The shape of the first CT image is D0×H0×W0×C0, where D0, H0, and W0 are the depth, height, and width of the three-dimensional dimensions of the first CT image, respectively, and C0 is the feature channel dimension of the first CT image, C0 = 1. The first CT image is composed of D0×H0×W0 first voxel points. Each first voxel point corresponds to a first voxel point feature vector of length 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.

[0073] Step 2: Based on the first semantic segmentation model, perform semantic segmentation processing on the first CT image to obtain the corresponding first semantic segmentation map of the gastric tumor lesion area and the perigastric lymph node area.

[0074] 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 operate in parallel. The inputs of both the first and second U-Net segmentation networks are connected to the input of the first semantic segmentation model, and their outputs are connected to the input of the first fusion module. Both the first and second U-Net segmentation networks are implemented based on the U-Net model structure. Here, the U-Net model is a visual model commonly used for semantic segmentation tasks.

[0075] The shape of the first semantic segmentation map is D1×H1×W1×C1, where D1, H1, and W1 are the depth, height, and width of the three-dimensional dimensions of the first semantic segmentation map, respectively, and 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 D1×H1×W1 second voxel points. Each second voxel point corresponds to a second voxel point feature vector of length C1. The second voxel point feature vector includes the semantics of the first lesion point, the semantics of the first lymph node point, and the semantics of the first background point. The semantics of the first lesion point, the semantics of the first lymph node point, and the semantics of the first background point are all binary semantics, taking values ​​of 0 or 1, where a value of 1 is true and a value of 0 is false. In each second voxel point feature vector, only one of the semantics of the first lesion point, the semantics of the first lymph node point, and the semantics of the first background point is 1.

[0076] The specific steps of current step 2 include:

[0077] Step 21: Input the first CT image into the first U-Net segmentation network; the first U-Net segmentation network performs semantic segmentation of the voxel points of the gastric tumor lesion area on the first CT image to obtain the corresponding first segmentation semantic map.

[0078] The shape of the first segmentation semantic graph is D. 11 ×H 11 ×W 11 ×C 11 D 11 H 11 W 11 C represents the depth, height, and width of the first segmentation semantic graph, respectively. 11 D represents the feature channel dimension of the first segmentation semantic graph. 11 =D0, H 11 =H0, W 11 =W0,C 11 =2; The first segmentation semantic graph is composed of D 11 ×H 11 ×W 11 It consists of four fourth voxels; each fourth voxel corresponds to a length of C. 11 The fourth voxel feature vector includes the semantics of the third lesion point and the semantics of the third background point. The semantics of the third lesion point and the semantics of the third background point are both binary semantics, taking values ​​of 0 or 1. When the value is 1, it is true, and when the value is 0, it is false. In each fourth voxel feature vector, there is only one of the semantics of the third lesion point and the semantics of the third background point that is 1.

[0079] Step 22, and input the first CT image into the second U-Net segmentation network; the second U-Net segmentation network performs lymph node semantic segmentation on all voxel points of the perigastric lymph node region on the first CT image to obtain the corresponding first binary segmentation semantic map;

[0080] The shape of the first two-segment semantic graph is D. 12 ×H 12 ×W 12 ×C 12 D 12 H 12 W 12 C represents the depth, height, and width of the first two-part semantic graph, respectively. 12 D represents the feature channel dimension of the first and second segmentation semantic graph. 12 =D0, H 12 =H0, W 12 =W0,C 12 =2; The first binary segmentation semantic graph is composed of D 12 ×H 12 ×W 12 It consists of five fifth voxels; each fifth voxel corresponds to a length of C. 12 The fifth voxel feature vector includes the semantics of the third lymph node and the semantics of the fourth background node. The semantics of the third lymph node and the semantics of the fourth background node are both binary semantics, taking values ​​of 0 or 1. A value of 1 indicates true and a value of 0 indicates false. In each fifth voxel feature vector, only one of the semantics of the third lymph node and the semantics of the fourth background node is 1.

[0081] Step 23: Input the first segmentation semantic map and the first second segmentation semantic map into the first fusion module; initialize a first semantic segmentation map with all zero semantics by the first fusion module; traverse all second voxel points on the first semantic segmentation map; during traversal, take the currently traversed second voxel point as the corresponding current voxel point, and record the fourth and fifth voxel points corresponding to the current voxel point on the first segmentation semantic map and the first second segmentation semantic map as the corresponding first and second corresponding voxel points; when the semantics of the third lesion point of the first corresponding voxel point is 1, reset the semantics of the first lesion point of the current voxel point to 1; when the semantics of the third lymph point of the second corresponding voxel point is 1, reset the semantics of the first lymph point of the current voxel point to 1; when the semantics of the third and fourth background points of the first and second corresponding voxel points are both 1, reset the semantics of the first background point of the current voxel point to 1; and at the end of the traversal, output the reset first semantic segmentation map as the processing result of the first semantic segmentation model.

[0082] Step 3: When the first prediction mode is the first mode, feature extraction processing is performed on the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set and first lymph node feature set; and based on the first prediction model, N-stage classification prediction processing is performed on the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type.

[0083] The first prediction type includes types N0, N1, N2, N3a, and N3b.

[0084] Specifically, it includes: Step 31, when the first prediction mode is the first mode, performing feature extraction processing on the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set and first lymph node feature set.

[0085] Specifically, it includes: step 311, extracting features from the radiomics features of the gastric tumor lesion region based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set;

[0086] The first lesion feature set includes 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. The first-order feature set of the lesion area includes the median gray value, average gray value, minimum gray value, maximum gray value, standard deviation, skewness, and peak value of the gray distribution of the lesion area. The shape feature set of the lesion area includes the surface area, volume, volume ratio, sphericity, compactness type, and diameter of the lesion area. The texture feature set of the lesion area includes the gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of the lesion area.

[0087] Specifically, it includes: step 3111, recording the second voxel point with semantic value 1 for each first lesion point on the first semantic segmentation map as the corresponding first lesion voxel point; and recording the first voxel point on the first CT image corresponding to each first lesion voxel point as the corresponding second lesion voxel point;

[0088] Step 3112 involves statistically analyzing the maximum depth, minimum depth, maximum height, minimum height, maximum width, and minimum width values ​​of all second lesion voxels to obtain the corresponding first maximum depth, first minimum depth, first maximum height, first minimum height, first maximum width, and first minimum width values; and then extracting the three-dimensional CT sub-images from the first CT image with depth ranging between the first maximum depth and first minimum depth, height ranging between the first maximum height and first minimum height, and width ranging between the first maximum width and first minimum width as the corresponding first lesion area sub-image.

[0089] Step 3113, and perform grayscale conversion on the first lesion area sub-image to obtain the corresponding grayscale image of the first lesion area;

[0090] Step 3114 involves calculating the median, average, minimum, maximum, standard deviation, skewness, and peak value of the grayscale image of the first lesion area to obtain the corresponding first-order feature set of the lesion area.

[0091] Here, the calculation method for calculating the first-order features (grayscale median, grayscale average, grayscale minimum, grayscale maximum, grayscale standard deviation, grayscale skewness, and grayscale distribution peak) of the three-dimensional grayscale image is a publicly available first-order feature calculation method for image omics features. The implementation details can be confirmed through relevant technical literature, and will not be repeated here.

[0092] Step 3115: The three-dimensional structure composed of all the pixel points of the second lesion on the grayscale image of the first lesion area is recorded as the 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 corresponding lesion area surface area, lesion area volume, lesion area volume ratio, lesion area sphericity, lesion area compactness type and lesion area diameter, which form the corresponding lesion area shape feature set;

[0093] Here, the calculation method for calculating the geometric features (surface area, volume, volume ratio, sphericity, compactness type, and diameter) of a specified three-dimensional structure in a three-dimensional grayscale image is a publicly available geometric feature calculation method for image omics features. The implementation details can be confirmed through relevant technical literature, and will not be repeated here.

[0094] Step 3116: Calculate the gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of the gray-level image of the first lesion area to obtain the corresponding lesion area gray-level co-occurrence matrix, lesion area gray-level run length matrix, lesion area gray-level size region matrix, lesion area neighborhood gray-level difference matrix, and lesion area gray-level correlation matrix, forming the corresponding lesion area texture feature set;

[0095] Here, the calculation method for calculating the texture features (gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix) in the 3D gray-level image is a publicly available method for calculating texture features of image omics features. The implementation details can be confirmed through relevant technical literature, and will not be repeated here.

[0096] Step 3117, and the obtained 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 are combined to form the corresponding first lesion feature set;

[0097] Step 312: Based on the first semantic segmentation map and the first CT image, feature extraction processing is performed on the radiomics features of all perigastric lymph node regions to obtain the corresponding first lymph node feature set;

[0098] The first lymph node feature set includes multiple first lymph node region feature sets; each first lymph node region feature set corresponds to a perigastric lymph node region; the first lymph node region feature set includes 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; the first lymph node region 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 region first-order feature set includes the median gray value of the lymph node region, the average gray value of the lymph node region, and the lymph node region texture feature set. The first lymph node region's gray-level minimum, maximum, standard deviation, skewness, and peak gray-level distribution are defined as follows: The first lymph node region's shape feature set includes lymph node surface area, volume, volume ratio, sphericity, compactness type, and diameter. The first lymph node region's texture feature set includes the lymph node region's gray-level co-occurrence matrix, run-length matrix, size region matrix, neighborhood gray-level difference matrix, and correlation matrix.

[0099] Specifically, it includes: step 3121, recording the second voxel points with semantic 1 for each first lymph node point on the first semantic segmentation map as the corresponding first lymph node voxel points; and recording the first voxel points on the first CT image corresponding to each first lymph node voxel point as the corresponding second lymph node voxel points;

[0100] Step 3122 involves clustering all second lymphocytes to obtain multiple first-class point cluster sets; combining each first-class point cluster into a current point cluster set; and statistically analyzing the maximum depth, minimum depth, maximum height, minimum height, maximum width, and minimum width values ​​of all second lymphocytes in the current point cluster set to obtain the corresponding second maximum depth, second minimum depth, second maximum height, second minimum height, second maximum width, and second minimum width values; and extracting the three-dimensional CT sub-image from the first CT image with a depth range between the second maximum depth and second minimum depth values, a height range between the second maximum height and second minimum height values, and a width range between the second maximum width and second minimum width values ​​as the corresponding first perigastric lymph node region sub-image.

[0101] The first type of point cluster includes multiple second lymphocyte points;

[0102] Step 3123, and calculate the corresponding first lymph node area distance by the straight-line distance from the center point of each first perigastric lymph node area sub-map to the center point of the first lesion area sub-map;

[0103] Step 3124, and perform grayscale conversion on each of the first perigastric lymph node sub-images to obtain the corresponding first lymph node area grayscale image;

[0104] Step 3125 involves calculating the median, average, minimum, maximum, standard deviation, skewness, and peak value of the grayscale image for each first lymph node region to obtain the corresponding first-order feature set for the lymph node region.

[0105] Step 3126: The three-dimensional structure composed of all the second lymphocyte pixels on the grayscale image of each first lymph node region is recorded as the corresponding first lymph node structure; and the surface area, volume, volume ratio, sphericity, compactness type and diameter of the first lymph node 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 form the corresponding first lymph node region shape feature set;

[0106] Step 3127: Calculate the gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of each first lymph node region's gray-level image to obtain the corresponding lymph node region's gray-level co-occurrence matrix, lymph node region's gray-level run length matrix, lymph node region's gray-level size region matrix, lymph node region's neighborhood gray-level difference matrix, and lymph node region's gray-level correlation matrix, thus forming the corresponding first lymph node region's texture feature set;

[0107] Step 3128, and a corresponding first lymph node region feature set is formed by the distance of the first lymph node region corresponding to each first perigastric lymph node region sub-map, the first-order feature set of the first lymph node region, the shape feature set of the first lymph node region, and the texture feature set of the first lymph node region; and a corresponding first lymph node feature set is formed by all the obtained first lymph node region feature sets;

[0108] Step 32, and based on the first prediction model, perform N-stage classification prediction processing according to the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type;

[0109] The first prediction model is based on a machine learning classification model; the types of machine learning classification models include at least classification prediction models based on RF models, classification prediction models based on SVM models, classification prediction models based on GBDT models, and classification prediction models based on GBDT+LR models.

[0110] Specifically, it includes: step 321, forming a corresponding first lesion-lymphatic feature set by combining the first lesion feature set with the first lymph node region feature sets of the first lymph node region feature set; and sorting all the first lesion-lymphatic feature sets in order of the distance between the first lymph node region feature sets in the first lesion-lymphatic feature set from near to far to form a corresponding first feature set sequence;

[0111] Step 322: Input the first feature set sequence into the first prediction model; the first prediction model identifies the metastatic features of gastric cancer tumors to all perigastric lymph node regions based on the radiomics features of the first feature set sequence, and classifies and predicts the five N-stage types of gastric cancer tumors based on the metastatic feature identification results to obtain the corresponding first prediction vector.

[0112] The first prediction vector includes five first prediction probabilities; each first prediction probability corresponds to an N-period type; the N-period types include N0 type, N1 type, N2 type, N3a type and N3b type;

[0113] Step 323: Take the N-period type corresponding to the first prediction probability with the largest probability value in the first prediction vector as the corresponding first prediction type.

[0114] Step 4, when the first prediction mode is the second mode, perform synthesis processing on the gastric tumor lesion and the perigastric lymph node region map based on the first semantic segmentation map and the first CT image to obtain the corresponding first synthesis map; and perform N-stage classification prediction processing on the first synthesis map based on the second prediction model to obtain the corresponding first prediction type;

[0115] Specifically, it includes: Step 41, when the first prediction mode is the second mode, perform synthesis processing on the gastric tumor lesion and the perigastric lymph node region map based on the first semantic segmentation map and the first CT image to obtain the corresponding first synthesis map;

[0116] Among them, the shape of the first synthesis map is D2×H2×W2×C2, where D2, H2, and W2 are the depth, height, and width of the three-dimensional size of the first synthesis map respectively, C2 is the feature channel dimension of the first synthesis map, D2 < D0, H2 < H0, W2 < W0, C2 = C0 + C1; the first synthesis map consists 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 the second voxel point pixel value, the second lesion point semantics, the second lymph node point semantics, and the second background point semantics; the second lesion point semantics, the second lymph node point semantics, and the second background point semantics are all binary semantics, taking values of 0 or 1, and taking the value of 1 as true and the value of 0 as false; among the second lesion point semantics, the second lymph node point semantics, and the second background point semantics of each third voxel point feature vector, only one is 1;

[0117] Specifically, it includes: Step 411, mark the second voxel points on the first semantic segmentation map where the first lesion point semantics or the first lymph node point semantics are 1 as the corresponding first type of voxel points; and mark the first voxel points on the first CT image corresponding to each first type of voxel points as the corresponding second type of voxel points;

[0118] Step 412, and count the maximum depth value, minimum depth value, maximum height value, minimum height value, maximum width value, and minimum width value among all the second type of voxel points to obtain the 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;

[0119] Step 413, and intercept the subgraphs on the first semantic segmentation map and the first CT image 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 as the corresponding first segmentation subgraph and the first CT subgraph;

[0120] Among them, the three-dimensional sizes of the first segmentation subgraph and the first CT subgraph are the same;

[0121] Step 414: The depth, height, and width of the three-dimensional dimensions of the first segmented sub-image or the first CT sub-image are taken as the corresponding depth D2, height H2, and width W2; the sum of the feature channel dimension C0 of the first CT sub-image and the feature channel dimension C1 of the first segmented sub-image is taken as the corresponding feature channel dimension C2; and the first segmented sub-image and the first CT sub-image are stitched together according to the voxel point feature vector stitching method to obtain a first composite image with the shape D2×H2×W2×C2.

[0122] Step 42, and based on the second prediction model, perform N-period classification prediction processing according to the first composite map to obtain the corresponding first prediction type;

[0123] The second prediction model is based on a deep learning classification model. The types of deep learning classification models include at least classification prediction models based on ResNet and classification prediction models based on Transformer.

[0124] Specifically, it includes: step 421, inputting the first composite image into the second prediction model; the second prediction model identifies the metastasis features of gastric cancer tumors to all perigastric lymph node regions based on the image features of the first composite image, and classifies and predicts the five N-stage types of gastric cancer tumors based on the metastasis feature identification results to obtain the corresponding second prediction vector;

[0125] The second prediction vector includes five second prediction probabilities; each second prediction probability corresponds to an N-period type; the N-period types include N0 type, N1 type, N2 type, N3a type and N3b type;

[0126] Step 422: Take the N-period type corresponding to the second prediction probability with the highest probability value in the second prediction vector as the corresponding first prediction type.

[0127] Step 5: Output the obtained first prediction type as the prediction result corresponding to the first CT image.

[0128] Here, in this embodiment of the invention, steps 1-5 above enable the prediction of gastric cancer N-stage classification based on two selectable prediction models and output the corresponding prediction results.

[0129] As can be seen from steps 1-5 above, this embodiment of the invention uses three types of models to complete the overall prediction process: a first semantic segmentation model, a first prediction model, and a second prediction model. It should be noted that the first semantic segmentation model, the first prediction model, and the second prediction model need to be trained separately before using these three types of models. The training steps for these three models will be described separately below.

[0130] 1) The processing steps for training the first semantic segmentation model are briefly described below:

[0131] Step A1: Collect a large number of three-dimensional abdominal CT images of patients diagnosed with various types of gastric cancer stage N as corresponding sample images to form the first image set;

[0132] The first image set includes multiple first sample images;

[0133] Step A2, and label the gastric tumor lesion areas on each first sample image based on manual annotation or other machine annotation methods to obtain the corresponding first labeled image; and according to the shape characteristics of the first segmentation semantic map output by the first U-Net segmentation network, construct a first label semantic map with the same three-dimensional size as the first labeled image and a feature channel dimension of 2 based on the gastric tumor lesion area labeling information of the first labeled image;

[0134] Step A3: Label all perigastric lymph node regions on each first sample image based on manual annotation or other machine annotation methods to obtain the corresponding second labeled image; and construct a second label semantic map with the same three-dimensional size as the second labeled image and a feature channel dimension of 2 based on the labeling information of all perigastric lymph node regions on the second labeled image, according to the shape characteristics of the first two-segment semantic map output by the second U-Net segmentation network.

[0135] Step A4, and the obtained first sample images and corresponding first label semantic maps form a corresponding first training data record; and the obtained first sample images and corresponding second label semantic maps form a corresponding second training data record; and all the obtained first training data records form a corresponding first training dataset; and all the obtained second training data records form a corresponding second training dataset;

[0136] Step A5, and train the first U-Net segmentation network based on the first training dataset;

[0137] Specifically, it includes: step A51, taking the first first training data record of the first training dataset as the corresponding current training data record;

[0138] Step A52: Input the first sample image of the current training data record into the first U-Net segmentation network; The first U-Net segmentation network performs semantic segmentation of the voxel points of the gastric tumor lesion area on the first sample image input this time to obtain the corresponding first predicted segmentation semantic map;

[0139] Step A53: Input the first predicted segmentation semantic map and the first label semantic map of the current training data record into the preset first model loss function to calculate the corresponding first loss value;

[0140] Here, the loss function of the first model is implemented based on the L1 loss function or the L2 loss function;

[0141] Step A54: Identify whether the first loss value meets the preset first loss value range; if the first loss value meets the first loss value range, identify whether the current training data record is the last first training data record of the first training dataset; if so, proceed to step A55; otherwise, extract the next first training data record of the first training dataset as the new current training data record and return to step A52 to continue training; if the first loss value does not meet the first loss value range, optimize the model parameters of the first U-Net segmentation network based on the preset first model parameter optimizer, and return to step A52 to continue training when the optimization ends.

[0142] Here, the first loss value range is a pre-set loss value range; the first model parameter optimizer includes at least the SDG optimizer;

[0143] Step A55: Solidify the model parameters of the first U-Net segmentation network and confirm that the training of the first U-Net segmentation network model is complete;

[0144] Step A6, and train the second U-Net segmentation network based on the second training dataset;

[0145] Specifically, this includes: step A61, taking the first second training data record of the second training dataset as the corresponding current training data record;

[0146] Step A62: Input the first sample image of the current training data record into the second U-Net segmentation network; The second U-Net segmentation network performs lymph node semantic segmentation on all voxel points of the perigastric lymph node region on the first sample image input this time to obtain the corresponding second predicted segmentation semantic map;

[0147] Step A63: Input the second predicted segmentation semantic map and the second label semantic map of the current training data record into the preset second model loss function to calculate the corresponding second loss value;

[0148] Here, the loss function of the second model is implemented based on the L1 loss function or the L2 loss function;

[0149] 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 in the second training dataset; if so, proceed to step A65; otherwise, extract the next second training data record in the second training dataset as the new current training data record and return to step A62 to continue training; if the second loss value does not meet the second loss value range, optimize the model parameters of the second U-Net segmentation network based on the preset second model parameter optimizer, and return to step A62 to continue training when the optimization ends.

[0150] Here, the second loss value range is a pre-set loss value range; the second model parameter optimizer includes at least the SDG optimizer;

[0151] Step A65: Solidify the model parameters of the second U-Net segmentation network and confirm that the training of the second U-Net segmentation network model is complete;

[0152] Step A7: After confirming that the training of the first and second U-Net segmentation networks has been completed, confirm that the training of the first semantic segmentation model has been completed.

[0153] 2) The steps for training the first prediction model are briefly described below:

[0154] Step B1: Each first sample image in the aforementioned first image set is taken as the corresponding current sample image; the first and second label semantic maps corresponding to the current sample images are taken as the corresponding current lesion area label semantic map and current perigastric lymph node area label semantic map; feature extraction is performed on the radiomics features of the gastric tumor lesion area based on the current lesion area label semantic map and the first sample images to obtain the corresponding first sample lesion feature set; feature extraction is then performed on the radiomics features of all perigastric lymph node areas based on the current perigastric lymph node area label semantic map and the first sample images to obtain the corresponding feature set. The first sample lymph node feature set is defined as follows: a first sample lesion-lymph node feature set is formed by combining the first sample lesion feature set and the lymph node region feature set of each first sample lymph node feature set; all first sample lesion-lymph node feature sets are sorted in order of distance from lymph node region from near to far to form a corresponding first sample feature set sequence; a corresponding first label vector is set based on the gastric cancer stage N type of the gastric cancer patient corresponding to the current sample image; and a corresponding third training data record is formed by the first sample feature set sequence and the first label vector corresponding to the current sample image.

[0155] The first sample lymph node feature set consists of multiple first sample lymph node region feature sets; the first label vector includes five first label probabilities; each first label probability corresponds to a stage N type; in the first label vector, the first label probability corresponding to the stage N type of gastric cancer in the gastric cancer patient corresponding to the current sample image is 1, and the other four first label probabilities are 0;

[0156] Step B2, and the corresponding third training dataset is composed of all the obtained third training data records;

[0157] Step B3 involves performing a full combination of the hyperparameters of the first prediction model to obtain multiple corresponding hyperparameter combinations; and setting the hyperparameters of the first prediction model based on each hyperparameter combination to obtain the corresponding first prediction model.

[0158] Step B4, and train and evaluate each first prediction model based on the third training dataset using k-fold cross-validation to obtain the corresponding first evaluation score;

[0159] Here, k is a positive integer greater than or equal to 3; model training and evaluation using k-fold cross-validation is a publicly available technique and will not be elaborated further here; it should be noted that the objective function or loss function used when training the model is based on the cross-entropy objective function or cross-entropy loss function for multi-class classification, and the evaluation method used when evaluating the model is the F1 score evaluation method based on the multi-class confusion matrix.

[0160] Step B5: Take the full parameter set of the first prediction model corresponding to the highest first evaluation score as the optimal full parameter set; and solidify the parameters of the first prediction model based on the optimal full parameter set, and confirm that the training of the first prediction model is completed.

[0161] 3) The steps for training the second prediction model are briefly described below:

[0162] Step C1: Take each first sample image in the aforementioned first image set as the 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 current perigastric lymph node area label semantic map; and perform gastric tumor lesion and perigastric lymph node area map synthesis processing based on the current lesion area label semantic map, the current perigastric lymph node area label semantic map and the current sample image to obtain the corresponding first sample composite map; and set a corresponding second label vector based on the gastric cancer stage N type of the gastric cancer patient corresponding to the current sample image; and form a corresponding fourth training data record by combining the first sample composite map and the second label vector corresponding to the current sample image.

[0163] The second label vector includes five second label probabilities; each second label probability corresponds to a stage N type; the second label probability corresponding to the stage N type of gastric cancer in the gastric cancer patient corresponding to the current sample image is 1, and the other four second label probabilities are 0.

[0164] Step C2, and the corresponding fourth training dataset is composed of all the obtained fourth training data records;

[0165] Step C3: Take the first fourth training data record of the fourth training dataset as the corresponding current training data record;

[0166] Step C4: Input the first sample composite image of the current training data record into the second prediction model; The second prediction model identifies the metastasis features of gastric cancer tumors to all perigastric lymph node regions based on the image features of the first sample composite image input this time, and classifies and predicts the five N-stage types of gastric cancer tumors based on the metastasis feature identification results to obtain the corresponding first training prediction vector;

[0167] The first training prediction vector includes five first training prediction probabilities; each first training prediction probability corresponds to an N-period type.

[0168] Step C5: Substitute the first training prediction vector and the second label vector of the current training data record into the preset third model loss function to calculate the corresponding third loss value;

[0169] Here, the loss function of the third model is implemented based on the cross-entropy loss function of multi-class classification;

[0170] Step C6: Identify whether the third loss value meets the preset range of the third loss value; if the third loss value meets the range of the third loss value, identify whether the current training data record is the last third training data record in the third training dataset. If yes, proceed to step C7; otherwise, extract the next third training data record in the third training dataset as the new current training data record and return to step C4 to continue training; if the third loss value does not meet the range of the third loss value, optimize the model parameters of the second prediction model based on the preset third model parameter optimizer, and return to step C4 to continue training when the optimization ends.

[0171] Here, the third loss value range is a pre-set loss value range; the third model parameter optimizer includes at least the SDG optimizer;

[0172] Step C7: Solidify the model parameters of the second prediction model and confirm that the training of the second prediction model is complete.

[0173] Figure 2 is a block diagram of a processing device for N-stage classification prediction based on CT images provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the aforementioned method embodiments, or it can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiments. For example, the device can be a device or chip system of the aforementioned 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.

[0174] The data receiving module 201 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.

[0175] The semantic segmentation module 202 is used to perform semantic segmentation processing on the first CT image of the gastric tumor lesion area and the perigastric lymph node area based on the first semantic segmentation model to obtain the corresponding first semantic segmentation map.

[0176] The first prediction model processing module 203 is used to extract features from the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image when the first prediction mode is the first mode, to obtain the corresponding first lesion feature set and first lymph node feature set; and to perform N-stage classification prediction based on the first prediction model according to the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type; the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type.

[0177] The second prediction model processing module 204 is used to perform gastric tumor lesion and perigastric lymph node region map synthesis processing based on the first semantic segmentation map and the first CT image to obtain the corresponding first composite map when the first prediction mode is the second mode; and to perform N-stage classification prediction processing based on the first composite map using the second prediction model to obtain the corresponding first prediction type.

[0178] The prediction output module 205 is used to output the obtained first prediction type as the prediction result corresponding to the first CT image.

[0179] The present invention provides a processing device for N-stage classification prediction based on CT images, which can execute the method steps in the above method embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0180] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented in software via processing element calls, while others are implemented in hardware. For example, the data receiving module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and called and executed by a processing element of the device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0181] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-on-a-Chip (SOC).

[0182] 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 disks (SSDs)).

[0183] Figure 3 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. As shown in Figure 3, 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 completing various processing functions and implementing the processing steps described in the methods of the aforementioned 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.

[0184] The system bus 305 mentioned in Figure 3 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0185] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), graphics processing units (GPUs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0186] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to perform the methods and processes provided in the above embodiments.

[0187] This invention also provides a chip for executing instructions, which is used to perform the processing steps described in the foregoing method embodiments.

[0188] This invention provides a processing method, apparatus, electronic device, and computer-readable storage medium for N-stage classification prediction based on CT images. As described above, the embodiments of the present invention first perform semantic segmentation processing on the received CT images to obtain corresponding semantic segmentation maps by semantic segmentation model for gastric tumor lesion area points and perigastric lymph node area points; then, two optional prediction modes are provided for prediction: 1) In the first mode, feature extraction is performed on the radiomics features of the gastric tumor lesion area and all perigastric lymph node areas based on the semantic segmentation map and CT images, and then a first prediction model based on a machine learning classification model is used to perform N-stage classification prediction based on the extracted first lesion feature set and first lymph node feature set to obtain the corresponding prediction type (N0 type, N1 type, N2 type, N3a type and N3b type); 2) In the second mode, the gastric tumor lesion and perigastric lymph node area maps are synthesized based on the semantic segmentation map and CT images to obtain the corresponding composite map, and then a second prediction model based on a deep learning classification model is used to perform N-stage classification prediction based on the composite map to obtain the prediction type (N0 type, N1 type, N2 type, N3a type and N3b type). The embodiments of the present invention improve both the real-time performance and the efficiency of prediction.

[0189] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction 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, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these 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 implementations should not be considered beyond the scope of this invention.

[0190] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0191] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A processing method for N-stage classification prediction based on CT images, characterized in that, The method includes: Receive a first CT image and a first prediction pattern; the first prediction pattern includes a first pattern and a second pattern; Based on the first semantic segmentation model, the first CT image is semantically segmented into the gastric tumor lesion area and the perigastric lymph node area to obtain the corresponding first semantic segmentation map. When the first prediction mode is the first mode, feature extraction processing is performed on the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set and first lymph node feature set; and based on the first prediction model, N-stage classification prediction processing is performed on the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type; the first prediction type includes N0 type, N1 type, N2 type, N3a type and N3b type; When the first prediction mode is the second mode, the first composite image is obtained by synthesizing the gastric tumor lesion and perigastric lymph node region map based on the first semantic segmentation map and the first CT image; and the first prediction type is obtained by performing N-stage classification prediction based on the first composite image using the second prediction model. The obtained first prediction type is output as the prediction result corresponding to the first CT image.

2. The processing method for N-stage classification prediction based on CT images according to claim 1, characterized in that, The first CT image is a three-dimensional abdominal CT image containing complete three-dimensional information of the stomach and all perigastric lymph node regions; the shape of the first CT image is D0×H0×W0×C0, where D0, H0, and W0 are the depth, height, and width of the three-dimensional dimensions of the first CT image, respectively, and C0 is the feature channel dimension of the first CT image, C0=1; the first CT image is composed of D0×H0×W0 first voxel points; each first voxel point corresponds to a first voxel point feature vector of length C0; the first voxel point feature vector includes the first voxel point pixel value. The first semantic segmentation map has a shape of D1×H1×W1×C1, where D1, H1, and W1 are the depth, height, and width of the three-dimensional dimensions of the first semantic segmentation map, respectively, and 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 D1×H1×W1 second voxel points. Each second voxel point corresponds to a second voxel point feature vector of length C1. The second voxel point feature vector includes the semantics of the first lesion point, the semantics of the first lymph node point, and the semantics of the first background point. The semantics of the first lesion point, the semantics of the first lymph node point, and the semantics of the first background point are all binary semantics, with values ​​of 0 or 1. A value of 1 indicates true, and a value of 0 indicates false. For each second voxel point feature vector, only one of the semantics of the first lesion point, the semantics of the first lymph node point, and the semantics of the first background point is 1. The shape of the first composite map is D2×H2×W2×C2, where D2, H2, and W2 are the depth, height, and width of the three-dimensional size of the first composite map respectively, C2 is the feature channel dimension of the first composite map, D2 < D0, H2 < H0, W2 < W0, and C2 = C0 + C1; the first composite map consists of D2×H2×W2 third voxel points; each of the third voxel points corresponds to a third voxel point feature vector with a length of C2; the third voxel point feature vector includes the second voxel point pixel value, the second lesion point semantics, the second lymph node point semantics, and the second background point semantics; the second lesion point semantics, the second lymph node point semantics, and the second background point semantics are all binary semantics, with a value of 0 or 1, where a value of 1 is true and a value of 0 is false; among the second lesion point semantics, the second lymph node point semantics, and the second background point semantics of each third voxel point feature vector, exactly one is 1; 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 the median gray value of the lesion area, the average gray value of the lesion area, the minimum gray value of the lesion area, the maximum gray value of the lesion area, the standard deviation of the gray value of the lesion area, the skewness of the gray value of the lesion area, and the peak value of the gray distribution of the lesion area; the lesion area shape feature set includes the surface area of the lesion area, the volume of the lesion area, the volume ratio of the lesion area, the sphericity of the lesion area, the compactness type of the lesion area, and the diameter of the lesion area; the lesion area texture feature set includes the gray-level co-occurrence matrix of the lesion area, the gray-level run length matrix of the lesion area, the gray-level size zone matrix of the lesion area, the neighborhood gray-level difference matrix of the lesion area, and the gray-level correlation matrix of the lesion area; The first lymph node feature set includes multiple 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 includes 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; the first lymph node region 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 region first-order feature set includes the median gray value of the lymph node region, the average gray value of the lymph node region, the minimum gray value of the lymph node region, the maximum gray value of the lymph node region, the standard deviation of the gray value of the lymph node region, the skewness of the gray value of the lymph node region, and the peak value of the gray distribution of the lymph node region; the first lymph node region shape feature set includes the surface area of the lymph node, the volume of the lymph node, the volume ratio of the lymph node, the sphericity of the lymph node, the compactness type of the lymph node, and the diameter of the lymph node; the first lymph node region texture feature set includes the gray-level co-occurrence matrix of the lymph node region, the gray-level run length matrix of the lymph node region, the gray-level size zone matrix of the lymph node region, the neighborhood gray-level difference matrix of the lymph node region, and the gray-level correlation matrix of the lymph node region.

3. The processing method for N-stage classification prediction based on CT images according to claim 2, wherein, 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 implemented in parallel; the input terminals of the first and second U-Net segmentation networks are both connected to the model input terminal of the first semantic segmentation model, and the output terminals are both connected to the input terminal of the first fusion module; the first and second U-Net segmentation networks are both implemented based on the model structure of the U-Net model. The first prediction model is implemented based on a machine learning classification model; the types of machine learning classification models include at least a classification prediction model based on the RF model, a classification prediction model based on the SVM model, a classification prediction model based on the GBDT model, and a classification prediction model based on the GBDT+LR model. The second prediction model is implemented based on a deep learning classification model, and the types of deep learning classification models include at least classification prediction models implemented based on ResNet models and classification prediction models implemented based on Transformer models.

4. The processing method for N-stage classification prediction based on CT images according to claim 3, characterized in that, The step of performing semantic segmentation processing on the first CT image based on the first semantic segmentation model to obtain the corresponding first semantic segmentation map specifically includes: The first CT image is input into the first U-Net segmentation network; the first U-Net segmentation network performs semantic segmentation of the voxel points of the gastric tumor lesion region on the first CT image to obtain the corresponding first segmentation semantic map; the shape of the first segmentation semantic map is D. 11 ×H 11 ×W 11 ×C 11 D 11 H 11 W 11 C represents the depth, height, and width of the three-dimensional dimensions of the first segmented semantic graph, respectively. 11 D represents the feature channel dimension of the first segmentation semantic graph. 11 =D0, H 11 =H0, W 11 =W0,C 11 =2; the first segmentation semantic graph is composed of D 11 ×H 11 ×W 11 It consists of four fourth voxel points; each of the fourth voxel points corresponds to a length of C. 11 The fourth voxel feature vector includes the semantics of the third lesion point and the semantics of the third background point; the semantics of the third lesion point and the semantics of the third background point are both binary semantics, taking values ​​of 0 or 1, where a value of 1 is true and a value of 0 is false; in each fourth voxel feature vector, only one of the semantics of the third lesion point and the semantics of the third background point is 1. The first CT image is input into the second U-Net segmentation network; the second U-Net segmentation network performs lymph node semantic segmentation on all voxel points of the perigastric lymph node region on the first CT image to obtain the corresponding first binary segmentation semantic map; the shape of the first binary segmentation semantic map is D. 12 ×H 12 ×W 12 ×C 12 D 12 H 12 W 12 C represents the depth, height, and width of the three-dimensional dimensions of the first two segmented semantic graphs, respectively. 12 D represents the feature channel dimension of the first and second segmentation semantic graphs. 12 =D0, H 12 =H0, W 12 =W0,C 12 =2; the first binary segmentation semantic graph is composed of D 12 ×H 12 ×W 12 It consists of five fifth voxel points; each fifth voxel point corresponds to a length of C. 12 The fifth voxel feature vector includes the semantics of the third lymph node and the semantics of the fourth background node; the semantics of the third lymph node and the semantics of the fourth background node are both binary semantics, taking values ​​of 0 or 1, where a value of 1 is true and a value of 0 is false; for each fifth voxel feature vector, only one of the semantics of the third lymph node and the semantics of the fourth background node is 1. The first semantic segmentation graph and the first second semantic segmentation graph are input into the first fusion module. The first fusion module initializes a first semantic segmentation graph with all zero semantics. All second voxel points on the first semantic segmentation graph are traversed. During the traversal, the currently traversed second voxel point is taken as the corresponding current voxel point, and the fourth voxel point and the fifth voxel point corresponding to the current voxel point on the first and first second semantic segmentation graphs are recorded as the corresponding first and second corresponding voxel points. When the semantics of the third lesion point of the first corresponding voxel point is 1, the semantics of the first lesion point of the current voxel point is reset to 1. When the semantics of the third lymph point of the second corresponding voxel point is 1, the semantics of the first lymph point of the current voxel point is reset to 1. When the semantics of the third and fourth background points of the first and second corresponding voxel points are both 1, the semantics of the first background point of the current voxel point is reset to 1. At the end of the traversal, the reset first semantic segmentation graph is output as the processing result of the first semantic segmentation model.

5. The processing method for N-stage classification prediction based on CT images according to claim 2, characterized in that, The step of extracting features from the radiomics features of the gastric tumor lesion region and the radiomics features of all perigastric lymph node regions based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set and first lymph node feature set specifically includes: Based on the first semantic segmentation map and the first CT image, feature extraction processing is performed on the radiomics features of the gastric tumor lesion region to obtain the corresponding first lesion feature set; Based on the first semantic segmentation map and the first CT image, feature extraction processing is performed on the radiomics features of all perigastric lymph node regions to obtain the corresponding first lymph node feature set.

6. The processing method for N-stage classification prediction based on CT images according to claim 5, characterized in that, The step of extracting features from the radiomics features of the gastric tumor lesion region based on the first semantic segmentation map and the first CT image to obtain the corresponding first lesion feature set specifically includes: The second voxel points with a semantic value of 1 for each of the first lesion points on the first semantic segmentation map are recorded as the corresponding first lesion voxel points; and the first voxel points on the first CT image corresponding to each of the first lesion voxel points are recorded as the corresponding second lesion voxel points. The maximum depth, minimum depth, maximum height, minimum height, maximum width, and minimum width values ​​of all second lesion voxels are statistically analyzed to obtain the corresponding first maximum depth, first minimum depth, first maximum height, first minimum height, first maximum width, and first minimum width values. Then, a three-dimensional CT sub-image from the first CT image whose depth range is between the first maximum depth and the first minimum depth, whose height range is between the first maximum height and the first minimum height, and whose width range is between the first maximum width and the first minimum width is extracted as the corresponding first lesion area sub-image. The first lesion area sub-image is then converted to grayscale to obtain the corresponding grayscale image of the first lesion area. The corresponding first-order feature set of the lesion area is obtained by calculating the median, average, minimum, maximum, standard deviation, skewness, and peak value of the grayscale image of the first lesion area. The three-dimensional structure composed of all the second lesion body pixels on the grayscale image of the first lesion area is recorded as the 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 corresponding lesion area surface area, lesion area volume, lesion area volume ratio, lesion area sphericity, lesion area compactness type and lesion area diameter, which form the corresponding lesion area shape feature set; The gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of the first lesion area gray-level image are calculated to obtain the corresponding gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of the lesion area, which together form the corresponding texture feature set of the lesion area. The first lesion feature set is composed of the obtained 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.

7. The processing method for N-stage classification prediction based on CT images according to claim 6, characterized in that, The step of extracting features from the radiomics features of all perigastric lymph node regions based on the first semantic segmentation map and the first CT image to obtain the corresponding first lymph node feature set specifically includes: The second voxel points with a semantic value of 1 for each of the first lymph nodes on the first semantic segmentation map are recorded as the corresponding first lymph node voxel points; and the first voxel points on the first CT image corresponding to each of the first lymph node voxel points are recorded as the corresponding second lymph node voxel points. The system performs point clustering on all second lymphosome pixels to obtain multiple first-class point cluster sets; and combines each first-class point cluster into a corresponding current point cluster set; then, it statistically analyzes the maximum depth value, minimum depth value, maximum height value, minimum height value, maximum width value, and minimum width value of all second lymphosome pixels in the current point cluster set to obtain corresponding second maximum depth value, second minimum depth value, second maximum height value, second minimum height value, second maximum width value, and second minimum width value; and extracts the three-dimensional CT sub-images from the first CT image whose depth range is between the second maximum depth value and the second minimum depth value, whose height range is between the second maximum height value and the second minimum height value, and whose width range is between the second maximum width value and the second minimum width value as the corresponding first perigastric lymph node region sub-image; the first-class point cluster set includes multiple second lymphosome pixels; The distance to the first lymph node region is calculated by taking the straight-line distance from the center point of each of the first perigastric lymph node region sub-maps to the center point of the first lesion region sub-map. The grayscale images of each of the first perigastric lymph node regions are converted to grayscale to obtain the corresponding first lymph node region grayscale images. The corresponding first-order feature set of the first lymph node region is formed by calculating the median, average, minimum, maximum, standard deviation, skewness, and peak value of the grayscale image of each first lymph node region. Each first lymph node region's grayscale image, composed of all second lymphocyte pixels, is denoted as a corresponding first lymph node structure. The surface area, volume, volume ratio, sphericity, compactness type, and diameter of each first lymph node 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, forming a corresponding first lymph node region shape feature set. The corresponding gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, neighborhood gray-level difference matrix, and gray-level correlation matrix of each first lymph node region gray-level image are calculated to form the corresponding first lymph node region texture feature set. Each first lymph node region feature set is composed of the distance to the first lymph node region corresponding to each first perigastric lymph node region sub-map, the first-order feature set of the first lymph node region, the shape feature set of the first lymph node region, and the texture feature set of the first lymph node region; and all the obtained first lymph node region feature sets are combined to form the corresponding first lymph node feature set.

8. The processing method for N-stage classification prediction based on CT images according to claim 2, characterized in that, The step of performing N-stage classification prediction based on the first prediction model according to the first lesion feature set and the first lymph node feature set to obtain the corresponding first prediction type specifically includes: The first lesion feature set and each of the first lymph node region feature sets in the first lymph node feature set are combined to form a corresponding first lesion-lymph node feature set; and all the first lesion-lymph node feature sets are sorted in order of the distance between the first lymph node region feature sets in each first lesion-lymph node feature set from near to far 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 metastatic features of gastric cancer tumors to all perigastric lymph node regions based on the radiomics features of the first feature set sequence, and classifies and predicts five N-stage types of gastric cancer tumors based on the metastatic feature identification results to obtain the corresponding first prediction vector; the first prediction vector includes five first prediction probabilities; each first prediction probability corresponds to one of the N-stage types; the N-stage types include N0 type, N1 type, N2 type, N3a type and N3b type; The N-period type corresponding to the first prediction probability with the highest probability value in the first prediction vector is taken as the corresponding first prediction type.

9. The processing method for N-stage classification prediction based on CT images according to claim 2, characterized in that, The step of synthesizing the gastric tumor lesion and perigastric lymph node region map based on the first semantic segmentation map and the first CT image to obtain the corresponding first composite map specifically includes: The second voxel points whose semantic meaning of each first lesion point or the first lymph node point on the first semantic segmentation map is 1 are recorded as the corresponding first type voxel points; and the first voxel points on the first CT image corresponding to each first type voxel point are recorded as the corresponding second type voxel points. And statistical analysis is performed on the maximum depth value, minimum depth value, maximum height value, minimum height value, maximum width value, and minimum width value of all second-type voxel points to obtain the 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; The first semantic segmentation image and the first CT image are cropped out as the corresponding first segmentation sub-image and first CT sub-image, respectively, 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; the three-dimensional dimensions of the first segmentation sub-image and the first CT sub-image are consistent. The depth, height, and width of the three-dimensional dimensions of the first segmented sub-image or the first CT sub-image are used as the corresponding depth D2, height H2, and width W2; the sum of the feature channel dimension C0 of the first CT sub-image and the feature channel dimension C1 of the first segmented sub-image is used as the corresponding feature channel dimension C2; and the first segmented sub-image and the first CT sub-image are stitched together according to the voxel point feature vector stitching method to obtain a first composite image with the shape D2×H2×W2×C2.

10. The processing method for N-stage classification prediction based on CT images according to claim 2, characterized in that, The step of performing N-period classification prediction based on the first composite graph using the second prediction model to obtain the corresponding first prediction type specifically includes: The first composite image is input into the second prediction model; the second prediction model identifies the metastasis features of gastric cancer tumors to all perigastric lymph node regions based on the image features of the first composite image, and classifies and predicts five N-stage types of gastric cancer tumors based on the metastasis feature identification results to obtain corresponding second prediction vectors; the second prediction vector includes five second prediction probabilities; each second prediction probability corresponds to one of the N-stage types; the N-stage types include N0 type, N1 type, N2 type, N3a type, and N3b type; The N-period type corresponding to the second prediction probability with the highest probability value in the second prediction vector is taken as the corresponding first prediction type.

11. An apparatus for performing the processing method for N-stage classification prediction based on CT images as described in any one of claims 1-10, 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 semantic segmentation processing on the first CT image of the gastric tumor lesion area and the perigastric lymph node area based on the first semantic segmentation model to obtain the corresponding first semantic segmentation map. The first prediction model processing module is used to extract features from the radiomics features of the gastric tumor lesion area and the radiomics features of all perigastric lymph node areas based on the first semantic segmentation map and the first CT image when the first prediction mode is the first mode, to obtain the corresponding first lesion feature set and first lymph node feature set; and to perform N-stage classification prediction based on the first prediction model according to the first lesion feature set and the first lymph node feature set to obtain the 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 perform gastric tumor lesion and perigastric lymph node region map synthesis processing based on the first semantic segmentation map and the first CT image to obtain the corresponding first composite map when the first prediction mode is the second mode; and to perform N-stage classification prediction processing based on the first composite map using the second prediction model to obtain the corresponding first prediction type. The prediction output module is used to output the obtained first prediction type as the prediction result corresponding to the first CT image.

12. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-10; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1-10.

Images