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

Abstract

Embodiments of the present invention relate to a processing method and apparatus for M-stage classification prediction based on CT images. The method comprises: constructing first and second semantic segmentation models based on a U-Net model; constructing a first prediction model based on a class of machine learning classification models; and constructing a second prediction model based on a class of deep learning classification models; training the first and second semantic segmentation models and the first and second prediction models; after training the four models, receiving a CT image and a prediction mode; if the prediction mode is the first mode, performing a prediction based on the CT image, the first and second semantic segmentation models, and the first prediction model; if the prediction mode is the second mode, performing a prediction based on the CT image, the first and second semantic segmentation models, and the second prediction model; and outputting the resulting prediction type. This invention can improve the real-time performance and efficiency of predictions.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a processing method and device for performing M-phase classification prediction based on CT images. Background Art

[0002] The M staging rule is a gastric cancer staging rule that is based on whether the gastric tumor has metastasized to distant organs (such as the liver), distal lymph nodes (such as para-aortic lymph nodes, retropancreatic lymph nodes, mesenteric root lymph nodes, etc.) or peritoneum. Common M staging types of this rule include M0 and M1. M0 indicates that no distant metastasis has occurred, and M1 indicates that distant metastasis has occurred. Currently, medical experts can manually identify the M staging type based on electronic computed tomography (CT) images. However, it is obvious that this manual identification method is limited by the working time of experts and has obvious timeliness defects, especially the processing efficiency of batch CT image recognition will be very low. Summary of the Invention

[0003] The purpose of the present invention is to address the shortcomings of the existing technology and provide a processing method, device, electronic device and computer-readable storage medium for M-stage classification prediction based on CT images. The present invention constructs two semantic segmentation models with reference to the model structure of the U-Net model, wherein the first semantic segmentation model is used to semantically segment the semantic points of the gastric tumor lesion area on the CT image, and the second semantic segmentation model is used to semantically segment the semantic points of the distal key tissue lesion area on the CT image, and constructs a first prediction model that can perform M-stage classification prediction based on the input imaging omics features with reference to a class of machine learning classification models, and constructs a second prediction model that can perform M-stage classification prediction based on the input three-dimensional CT image with lesion point markings with reference to a class of deep learning classification models; and trains the above four types of models respectively; and after the training of the above four types of models is completed, different prediction model combinations are selected based on the received prediction mode for prediction: when the prediction mode is the first mode, the first and second semantic segmentation models and the first prediction model are used to perform M-stage classification prediction on the received CT image; when the prediction mode is the second mode, the first, second semantic segmentation models and the second prediction model are used to perform M-stage classification prediction on the received CT image. The present invention can achieve the purpose of improving prediction real-time performance and prediction efficiency.

[0004] To achieve the above-mentioned object, a first aspect of an embodiment of the present invention provides a processing method for M-stage classification prediction based on CT images, the method comprising:

[0005] Based on the U-Net model, a semantic segmentation model for semantic point segmentation of gastric tumor lesions on CT images is constructed as the corresponding first semantic segmentation model; and based on the U-Net model, a semantic segmentation model for semantic point segmentation of lesions on distal key tissues on the CT images is constructed as the corresponding second semantic segmentation model; and based on a class of machine learning classification models, a prediction model for M-stage classification prediction based on imaging omics features is constructed as the corresponding first prediction model; and based on a class of deep learning classification models, a prediction model for M-stage classification prediction based on three-dimensional CT images with lesion point markings is constructed as the corresponding second prediction model; wherein, the CT The image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissues; the distal key tissues include at least the liver, the para-aortic lymph node area, the retropancreatic lymph node area, the mesenteric root lymph node area and the peritoneum; the types of the 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 types of the deep learning classification models include at least a classification prediction model based on the ResNet model and a classification prediction model based on the Transformer model;

[0006] Training the first semantic segmentation model, the second semantic segmentation model, the first prediction model, and the second prediction model respectively;

[0007] After the training of the four types of models is completed, the first CT image and the first prediction mode are received; and the first prediction mode is identified; if the first prediction mode is the first mode, the corresponding first prediction type is obtained by prediction based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the first prediction model; if the first prediction mode is the second mode, the corresponding first prediction type is obtained by prediction based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the second prediction model; and the obtained first prediction type is output as the prediction result corresponding to the first CT image; wherein, the first CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissue; the first prediction mode includes the first mode and the second mode; the first prediction type includes the M0 type and the M1 type.

[0008] Preferably, the training of the first semantic segmentation model, the second semantic segmentation model, the first prediction model, and the second prediction model respectively includes:

[0009] A first subject set is formed by recruiting a plurality of subjects who do not have gastric cancer, a plurality of subjects who have gastric cancer but have not metastasized to any of the distal key tissues, and a plurality of subjects who have gastric cancer and have metastasized to one or more types of tissues in the distal key tissues; the first subject set includes a plurality of first subjects, each of the first subjects corresponds to a first M stage type; the first M stage type includes an M0 type and an M1 type; the first subject who does not have gastric cancer or has gastric cancer but has not metastasized to any of the distal key tissues corresponds to the M0 type, and the first subject who has gastric cancer and has metastasized to one or more types of tissues in the distal key tissues corresponds to the M1 type;

[0010] Acquire the most recent CT image of each of the first subjects as a corresponding first sample image; form a corresponding first original data record from each of the first sample images and the corresponding first M-phase type; and form a corresponding first original data set from all of the obtained first original data records; the first sample image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissue; the first sample image includes a plurality of first sample voxel points; each of the first sample voxel points corresponds to a first voxel point pixel value;

[0011] Constructing four types of model training data sets based on the first original data set, which are recorded as the corresponding first model data set, second model data set, third model data set and fourth model data set;

[0012] Training the first semantic segmentation model based on the first model dataset;

[0013] Training the second semantic segmentation model based on the second model dataset;

[0014] Training the first prediction model based on the third model dataset;

[0015] Training the second prediction model based on the fourth model dataset;

[0016] And after the first and second semantic segmentation models and the first and second prediction models have completed model training, it is confirmed that the four-category model training is completed.

[0017] Furthermore, the first model data set includes multiple first model data records; the first model data record includes a first training image and a first label semantic map; the three-dimensional size of the first label semantic map is consistent with the three-dimensional size of the first training image; the first label semantic map includes multiple first label voxel points; each first label voxel point corresponds to a binary first label semantic point, with a value of 0 or 1; when the first label semantic point is 0, it indicates that the corresponding first label voxel point is a background voxel point, and when it is 1, it indicates that the corresponding first label voxel point is a gastric tumor lesion area voxel point;

[0018] The second model data set includes a plurality of second model data records; the second model data records include a second training image and a second label semantic map; the three-dimensional size of the second label semantic map is consistent with the three-dimensional size of the second training image; the second label semantic map includes a plurality of second label voxel points; each second label voxel point corresponds to a binary second label semantic point, with a value of 0 or 1; when the second label semantic point is 0, it indicates that the corresponding first label voxel point is a background voxel point, and when it is 1, it indicates that the corresponding second label voxel point is a distal key tissue lesion voxel point;

[0019] The third model data set includes multiple third model data records; the third model data record includes a first training imaging omics feature set and a first label type vector; the first training imaging omics feature set includes a first gastric tumor feature set and a first distal key tissue feature set; the first distal key tissue feature set includes multiple first distal key tissue feature sets, each of the first distal key tissue feature sets corresponds to one distal key tissue; the first gastric tumor feature set and all the first distal key tissue feature sets are composed of a set of first-order feature sets, shape feature sets and texture feature sets of imaging omics features; the first label type vector includes two first label type probabilities, each of the first label type probability corresponds to an M stage type; the M stage type includes M0 type and M1 type;

[0020] The fourth model data set includes multiple fourth model data records; the fourth model data record includes a first training feature map and a second label type vector; the first training feature map includes multiple first feature voxel points; each of the first feature voxel points corresponds to a second voxel point feature vector; the second voxel point feature vector includes at least a second voxel point pixel value and a second voxel point semantics; the second voxel point semantics is a three-valued semantic value, which takes a value of 0, 1 or 2; when the second voxel point semantics is 0, it indicates that the corresponding first feature voxel point is a background voxel point; when it is 1, it indicates that the corresponding first feature voxel point is a gastric tumor lesion area voxel point; when it is 2, it indicates that the corresponding first feature voxel point is a distal critical tissue lesion voxel point; the second label type vector includes two second label type probabilities, each of the second label type probability corresponds to one of the M stage types.

[0021] Furthermore, the four types of model training data sets constructed based on the first original data set are recorded as the corresponding first model data set, second model data set, third model data set and fourth model data set, specifically including:

[0022] using each of the first raw data records of the first raw data set as a corresponding current raw data record; and using the first sample image and the corresponding first M-phase type of the current raw data record as a corresponding current sample image and current sample type;

[0023] and using the current sample image as the corresponding first training image and the second training image; and constructing the corresponding first and second label semantic maps based on the three-dimensional sizes of the first and second training images; and initializing the first and second label semantics of all the first and second label voxel points of the first and second label semantic maps to 0;

[0024] and marking all the first sample voxel points located in the gastric tumor lesion area on the current sample image as type I voxel points based on manual labeling or other machine labeling methods, and marking all the first sample voxel points located in all distal key tissue lesion areas on the current sample image as type II voxel points;

[0025] When the total number of the first type of voxel points is zero, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets is set to form a corresponding first gastric tumor feature set;

[0026] and when the total number of the first type of voxel points is greater than zero, resetting the first label semantics of the first label voxel points corresponding to each of the first type of voxel points on the first label semantic map to 1; and recording the three-dimensional structure composed of all the first type of voxel points on the current sample image as the corresponding first type of structure; and calculating the first-order features, shape features, and texture features of the imaging omics features of the first type of structure to obtain a corresponding first-order feature set, shape feature set, and texture feature set to form a corresponding first gastric tumor feature set;

[0027] When the total number of the two types of voxel points is zero, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets is set for each of the distal key tissues to form a corresponding first distal key tissue feature set; and all the obtained first distal key tissue feature sets are used to form a corresponding first distal key tissue feature set;

[0028] And when the total number of the second-category voxel points is greater than zero, the second label semantics of the second-category voxel points corresponding to each of the second-category voxel points on the second label semantic map are reset to 1; and based on the preset point group clustering algorithm, the second-category voxel points corresponding to each of the distal key tissues on the current sample image are clustered to obtain the corresponding first point group; and each of the first point groups is used as the corresponding current point group, and whether the current point group is empty is identified; if the current point group is empty, a set of all-zero first-order feature sets is set for the distal key tissue corresponding to the current point group, The shape feature set and the texture feature set constitute a corresponding first distal key tissue feature set; if the current point group is not empty, the three-dimensional structure formed by the current point group on the current sample image is recorded as the corresponding second-class structure, and the first-order features, shape features, and texture features of the imaging omics features of the second-class structure are calculated to obtain the corresponding first-order feature set, shape feature set, and texture feature set to constitute a corresponding first distal key tissue feature set; and all the obtained first distal key tissue feature sets constitute a corresponding first distal key tissue feature set;

[0029] and forming a corresponding first training radiomics feature set from the first gastric tumor feature set and the first distal key tissue feature set corresponding to the current sample image;

[0030] And each of the first sample voxel points of the current sample image is taken as the corresponding current sample voxel point; and the first and second label voxel points corresponding to the current sample voxel point on the first and second label semantic maps corresponding to the current sample image are recorded as the corresponding first and second matching voxel points; and the first voxel point pixel value of the current sample voxel point is taken as the corresponding second voxel point pixel value; and when the first and second label semantics corresponding to the first and second matching voxel points are both 0, a corresponding second voxel point semantic is set to 0; and the first label semantic corresponding to the first matching voxel point is 1, sets the semantics of a corresponding second voxel point to 1; and sets the semantics of a corresponding second voxel point to 2 when the second label semantics corresponding to the second matching voxel point is 1; and constructs a corresponding first feature voxel point based on the three-dimensional voxel coordinates of the current sample voxel point and records it as the corresponding current feature voxel point; and forms a second voxel point feature vector corresponding to the current feature voxel point by using the second voxel point pixel value and the second voxel point semantics corresponding to the current sample voxel point; and forms a corresponding first training feature map by using all the first feature voxel points corresponding to the current sample image;

[0031] and constructing the corresponding first and second label type vectors according to the current sample type; when the current sample type is type M0, the corresponding first and second label type probabilities corresponding to type M0 in the corresponding first and second label type vectors are 1, and the first and second label type probabilities corresponding to type M1 are 0; when the current sample type is type M1, the corresponding first and second label type probabilities corresponding to type M1 in the corresponding first and second label type vectors are 0, and the first and second label type probabilities corresponding to type M1 are 1;

[0032] The first training image and the first label semantic map corresponding to the current original data record form a corresponding first model data record; the second training image and the second label semantic map corresponding to the current original data record form a corresponding second model data record; the first training imaging feature set and the first label type vector corresponding to the current original data record form a corresponding third model data record; and the first training feature map and the second label type vector corresponding to the current original data record form a corresponding fourth model data record;

[0033] All the first model data records corresponding to the first original data set constitute the corresponding first model data set; all the second model data records corresponding to the first original data set constitute the corresponding second model data set; all the third model data records corresponding to the first original data set constitute the corresponding third model data set; and all the fourth model data records corresponding to the first original data set constitute the corresponding fourth model data set.

[0034] Furthermore, the training of the first semantic segmentation model based on the first model dataset specifically includes:

[0035] Step 51: taking the first first model data record of the first model data set as the corresponding current model data record;

[0036] Step 52: Input the first training image recorded in the current model data into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding first predicted semantic map;

[0037] Step 53: Input the first predicted semantic graph and the first label semantic graph of the current model data record into a preset first model loss function to calculate and obtain a corresponding first loss value;

[0038] Wherein, the first model loss function is implemented based on the L1 loss function or the L2 loss function;

[0039] Step 54: Identify whether the first loss value satisfies a preset first loss value range. If the first loss value satisfies the first loss value range, identify whether the current model data record is the last first model data record of the first model dataset. If so, proceed to step 55. If not, use the next first model data record of the first model dataset as the new current model data record and return to step 52 to continue training. If the first loss value does not satisfy the first loss value range, perform a round of model parameter optimization on the first semantic segmentation model based on a preset first model parameter optimizer in a direction to minimize the first model loss function. After this round of parameter optimization is completed, return to step 52 to continue training.

[0040] Wherein, the first model parameter optimizer includes at least an SDG optimizer;

[0041] Step 55: solidify the model parameters of the first semantic segmentation model and confirm that the model training of the first semantic segmentation model is completed.

[0042] Furthermore, the training of the second semantic segmentation model based on the second model dataset specifically includes:

[0043] Step 61: taking the first second model data record of the second model data set as the corresponding current model data record;

[0044] Step 62: Input the second training image recorded in the current model data into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding second predicted semantic map;

[0045] Step 63: Input the second predicted semantic graph and the second label semantic graph of the current model data record into a preset second model loss function to calculate and obtain a corresponding second loss value;

[0046] Wherein, the second model loss function is implemented based on the L1 loss function or the L2 loss function;

[0047] Step 64: Identify whether the second loss value satisfies a preset second loss value range. If the second loss value satisfies the second loss value range, identify whether the current model data record is the last second model data record of the second model dataset. If so, proceed to step 65. If not, use the next second model data record of the second model dataset as the new current model data record and return to step 62 to continue training. If the second loss value does not satisfy the second loss value range, perform a round of model parameter optimization on the second semantic segmentation model based on a preset second model parameter optimizer in a direction to minimize the second model loss function. At the end of this round of parameter optimization, return to step 62 to continue training.

[0048] Wherein, the second model parameter optimizer includes at least an SDG optimizer;

[0049] Step 65: solidify the model parameters of the second semantic segmentation model and confirm that the model training of the second semantic segmentation model is completed.

[0050] Furthermore, the training of the first prediction model based on the third model data set specifically includes:

[0051] Step 701: performing a full combination of hyperparameters of a first prediction model to obtain a plurality of corresponding first hyperparameter combinations; and performing a hyperparameter setting on the first prediction model based on each of the first hyperparameter combinations to obtain a corresponding first prediction model;

[0052] Step 702: Set the parameter K to a positive integer greater than or equal to 5; and divide the third model data set into K sub-data sets and record each sub-data set as the corresponding first sub-data set D i ; 1≤index i≤K;

[0053] Step 703: any of the first sub-datasets D i As a corresponding first validation dataset DB i , and the first sub-dataset D will be i The remaining K-1 first subsets D i Merge into a corresponding first test dataset DA i ; and the first test data sets DA obtained by each i and the corresponding first validation dataset DB i Form a corresponding first test validation set;

[0054] Step 704: taking the first of the first prediction models as the corresponding current prediction model;

[0055] Step 705: taking the first of the first test verification sets as the corresponding current test verification set;

[0056] Step 706: The first test data set DA of the current test validation set is i and the first validation dataset DB i As the corresponding current test dataset and current validation dataset;

[0057] Step 707: taking the first third model data record of the current test data set as the corresponding current model data record;

[0058] Step 708: Input the first training radiomics feature set recorded in the current model data into the current prediction model to perform M-stage classification prediction to obtain a corresponding first prediction vector;

[0059] The first prediction vector includes two first-type prediction probabilities, each of which corresponds to one of the M staging types;

[0060] Step 709: Substitute the first prediction vector and the first label type vector of the current model data record into a preset third model loss function; and perform a round of model parameter optimization on the model parameters of the current prediction model in a direction in which the third model loss function is minimized based on a preset third model parameter optimizer; and at the end of this round of parameter optimization, identify whether the current model data record is the last third model data record of the current test data set; if so, go to step 710; if not, extract the next third model data record of the current test data set as the new current model data record and return to step 708 for training;

[0061] Wherein, the third model loss function is implemented based on a binary cross entropy loss function; the third model parameter optimizer includes at least an SDG optimizer;

[0062] Step 710: Set three counters initially set to 0 as the corresponding TP counter, FP counter, and FN counter;

[0063] Step 711: traverse all the third model data records in the current verification data set; and during the traversal, use the currently traversed third model data record as the corresponding current verification record; and use the first training radiomics feature set and the first label type vector of the current verification record as the corresponding current radiomics feature set and current label type vector; and use the M stage type corresponding to the first label type probability with a probability value of 1 in the current label type vector as the corresponding current label type; and input the current radiomics feature set into the current prediction model to perform M stage classification prediction to obtain the corresponding second prediction vector; and use the M stage type corresponding to the second type prediction probability with the largest probability in the second prediction vector as the corresponding current prediction type; and compare the current prediction type with the current label type; if the current prediction type and the current label type are both M1 type, add 1 to the TP counter; if the current prediction type is M1 type and the current label type is M0 type, add 1 to the FP counter; if the current prediction type is M0 type and the current label type is M1 type, add 1 to the FN counter;

[0064] The second prediction vector includes two second-type prediction probabilities, each of which corresponds to one of the M staging types;

[0065] Step 712: After traversing all the third model data records in the current validation data set, the precision and recall are calculated based on the TP counter, the FP counter, and the FN counter to obtain the corresponding first precision and first recall; and the F1 score is calculated based on the first precision and the first recall to obtain the corresponding first score F1. i ;

[0066] Step 713: Identify whether the current test validation set is the Kth first test validation set; if so, go to step 714; if not, use the next first test validation set as the new current test validation set and return to step 706;

[0067] Step 714: Based on the obtained K first scores F1 i Calculate the mean square error to get the corresponding first error E,

[0068] Step 715: Identify whether the current prediction model is the last of the first-first prediction models; if so, go to step 716; if not, use the next first-first prediction model as the new current prediction model and return to step 705;

[0069] Step 716: taking the full model parameter set of the first prediction model corresponding to the first error E having the smallest error as the corresponding optimal full model parameter set;

[0070] Step 717: solidify the model parameters of the first prediction model based on the optimal full model parameter set and confirm that the model training of the first prediction model is completed.

[0071] Furthermore, the training of the second prediction model based on the fourth model data set specifically includes:

[0072] Step 81: taking the first fourth model data record of the fourth model data set as the corresponding current model data record;

[0073] Step 82: Input the first training feature map of the current model data record into the second prediction model to perform M-period classification prediction to obtain a corresponding third prediction vector;

[0074] The third prediction vector includes two third-type prediction probabilities, each of which corresponds to one of the M staging types;

[0075] Step 83: Input the third prediction vector and the second label type vector of the current model data record into a preset fourth model loss function to calculate a corresponding third loss value;

[0076] Wherein, the fourth model loss function is implemented based on the binary cross entropy loss function;

[0077] Step 84: Identify whether the third loss value satisfies a preset third loss value range; if the third loss value satisfies the third loss value range, identify whether the current model data record is the last fourth model data record of the fourth model data set; if so, go to step 85; if not, use the next fourth model data record of the fourth model data set as the new current model data record and return to step 82 to continue training; if the third loss value does not satisfy the third loss value range, perform a round of model parameter optimization on the second prediction model based on a preset fourth model parameter optimizer in a direction to minimize the fourth model loss function, and return to step 82 to continue training at the end of this round of parameter optimization;

[0078] Wherein, the fourth model parameter optimizer includes at least an SDG optimizer;

[0079] Step 85: solidify the model parameters of the second prediction model and confirm that the model training of the second prediction model is completed.

[0080] Preferably, the performing prediction according to the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the first prediction model to obtain the corresponding first prediction type specifically includes:

[0081] Step 91: Input the first CT image into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding first semantic map;

[0082] The first CT image includes a plurality of first voxel points; the three-dimensional size of the first semantic map is consistent with the three-dimensional size of the first CT image; the first semantic map includes a plurality of second voxel points, and the first and second voxel points correspond to each other one by one; each second voxel point corresponds to a binary gastric tumor lesion semantic point, with a value of 0 or 1; when the gastric tumor lesion semantic point is 0, it indicates that the corresponding second voxel point is a background voxel point, and when it is 1, it indicates that the corresponding second voxel point is a gastric tumor lesion voxel point;

[0083] Step 92: Count the number of the second voxel points in the gastric tumor lesion area with a semantic value of 1 on the first semantic map to obtain a corresponding first number; and determine whether the first number exceeds a preset first number threshold. If so, the process proceeds to step 93; if not, the process sets the corresponding first prediction type to type M0 and proceeds to step 99.

[0084] Step 93: Input the first CT image into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding second semantic map;

[0085] The three-dimensional size of the second semantic map is consistent with the three-dimensional size of the first CT image; the second semantic map includes a plurality of third voxel points, and the first and third voxel points correspond to each other one by one; each of the third voxel points corresponds to a binary distal key tissue lesion region semantic point, with a value of 0 or 1; when the distal key tissue lesion region semantic point is 0, it indicates that the corresponding third voxel point is a background voxel point, and when it is 1, it indicates that the corresponding third voxel point is a distal key tissue lesion voxel point;

[0086] Step 94, counting the number of the third voxel points with a semantic value of 1 in the distal key tissue lesion area on the second semantic map to obtain a corresponding second number; and identifying whether the second number exceeds a preset second number threshold. If so, the process proceeds to step 95; if not, the process sets the corresponding first prediction type to type M0 and proceeds to step 99;

[0087] Step 95: Record the first voxel points on the first CT image corresponding to the second voxel points with a semantic value of 1 in each of the gastric tumor lesion areas as corresponding first lesion points; and record the three-dimensional structure composed of all the first lesion points as the corresponding first structure; and calculate the first-order features, shape features, and texture features of the imaging omics features of the first structure to obtain corresponding first-order feature sets, shape feature sets, and texture feature sets to form a corresponding second gastric tumor feature set;

[0088] Step 96: Cluster the first voxel points corresponding to each of the distal key tissues on the first CT image based on a preset point group clustering algorithm to obtain a corresponding second point group; and use each of the second point groups as a corresponding current point group, and identify whether the current point group is empty; if the current point group is empty, set a set of all-zero first-order feature sets, shape feature sets, and texture feature sets for the distal key tissue corresponding to the current point group to form a corresponding second distal key tissue feature set; if the current point group is not empty, record the three-dimensional structure composed of the current point group on the first CT image as the corresponding second structure, and calculate the first-order features, shape features, and texture features of the imaging omics features of the second structure to obtain corresponding first-order feature sets, shape feature sets, and texture feature sets to form a corresponding second distal key tissue feature set; and form a corresponding second distal key tissue feature set from all the obtained second distal key tissue feature sets;

[0089] Step 97: Combining the obtained second gastric tumor feature set and the obtained second distal key tissue feature set into a corresponding first radiomics feature set; and inputting the first radiomics feature set into the first prediction model to perform M-stage classification prediction to obtain a corresponding fourth prediction vector;

[0090] The fourth prediction vector includes two fourth-type prediction probabilities, each of which corresponds to one of the M staging types;

[0091] Step 98: taking the M stage type corresponding to the fourth type prediction probability with the largest probability in the fourth prediction vector as the corresponding first prediction type;

[0092] Step 99: output the obtained first prediction type as the current prediction result.

[0093] Preferably, the performing prediction according to the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the second prediction model to obtain the corresponding first prediction type specifically includes:

[0094] Step 101: Input the first CT image into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding third semantic map;

[0095] The first CT image includes a plurality of first voxel points; each of the first voxel points corresponds to a first voxel point pixel value; the three-dimensional size of the third semantic map is consistent with the three-dimensional size of the first CT image; the third semantic map includes a plurality of fourth voxel points, and the first and fourth voxel points correspond to each other one by one; each of the fourth voxel points corresponds to a binary gastric tumor lesion semantic point, with a value of 0 or 1; when the gastric tumor lesion semantic point is 0, it indicates that the corresponding fourth voxel point is a background voxel point, and when it is 1, it indicates that the corresponding fourth voxel point is a gastric tumor lesion voxel point;

[0096] Step 102: Count the number of the fourth voxel points in the gastric tumor lesion area with a semantic value of 1 on the third semantic map to obtain a corresponding third number; and determine whether the third number exceeds a preset first number threshold. If so, proceed to step 103; if not, set the corresponding first prediction type to type M0 and proceed to step 108.

[0097] Step 103: Input the first CT image into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding fourth semantic map;

[0098] The three-dimensional size of the fourth semantic map is consistent with the three-dimensional size of the first CT image; the fourth semantic map includes a plurality of fifth voxel points, and the first and fifth voxel points correspond one to one; each of the fifth voxel points corresponds to a binary distal key tissue lesion area semantic, with a value of 0 or 1; when the distal key tissue lesion area semantic is 0, it indicates that the corresponding fifth voxel point is a background voxel point, and when it is 1, it indicates that the corresponding fifth voxel point is a distal key tissue lesion voxel point;

[0099] Step 104: Count the number of the fifth voxel points with a semantic value of 1 in the distal key tissue lesion area on the second semantic map to obtain a corresponding fourth number; and determine whether the fourth number exceeds a preset second number threshold. If so, proceed to step 105; if not, set the corresponding first prediction type to type M0 and proceed to step 108.

[0100] Step 105: each of the first voxel points of the first CT image is used as the corresponding current voxel point; and the fourth and fifth voxel points corresponding to the current voxel point on the third and fourth semantic maps are recorded as the corresponding third and fourth matching voxel points; and the first voxel point pixel value of the current voxel point is used as a corresponding feature voxel point pixel value; and when the semantics of the gastric tumor lesion area and the distal key tissue lesion area corresponding to the third and fourth matching voxel points are both 0, a corresponding feature voxel point semantic is set to 0; and when the semantics of the gastric tumor lesion area corresponding to the third matching voxel point is 1, setting the semantics of a corresponding feature voxel point to 1; and setting the semantics of a corresponding feature voxel point to 2 when the semantics of the distal key tissue lesion area corresponding to the fourth matching voxel point is 1; and constructing a corresponding second feature voxel point based on the three-dimensional voxel coordinates of the current voxel point and recording it as the corresponding current feature voxel point; and forming a feature voxel point feature vector corresponding to the current feature voxel point by using the pixel value of the feature voxel point corresponding to the current voxel point and the semantics of the feature voxel point; and forming a corresponding first feature map by using all the second feature voxel points corresponding to the first CT image;

[0101] Step 106: Input the first feature map into the second prediction model to perform M-period classification prediction to obtain a corresponding fifth prediction vector;

[0102] The fifth prediction vector includes two fifth-type prediction probabilities, each of which corresponds to one of the M staging types;

[0103] Step 107: taking the M stage type corresponding to the fifth type prediction probability with the largest probability in the fifth prediction vector as the corresponding first prediction type;

[0104] Step 108: Output the obtained first prediction type as the current prediction result.

[0105] A second aspect of the embodiments of the present invention provides a device for implementing the processing method for performing M-stage classification prediction based on CT images as described in the first aspect, the device comprising: a model building module, a model training module, and a model application module;

[0106] The model construction module is used to construct a semantic segmentation model for performing semantic point segmentation of the lesion area of ​​gastric tumors on CT images based on the U-Net model, which is recorded as the corresponding first semantic segmentation model; and to construct a semantic segmentation model for performing semantic point segmentation of the lesion area of ​​distal key tissues on the CT images based on the U-Net model, which is recorded as the corresponding second semantic segmentation model; and to construct a prediction model for performing M-stage classification prediction based on imaging omics features based on a class of machine learning classification models, which is recorded as the corresponding first prediction model; and to construct a prediction model for performing M-stage classification prediction based on a three-dimensional CT image with lesion point markings based on a class of deep learning classification models, which is recorded as the corresponding second prediction model; wherein , the CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissues; the distal key tissues include at least the liver, the para-aortic lymph node area, the retropancreatic lymph node area, the mesenteric root lymph node area and the peritoneum; the types of the 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 types of the deep learning classification models include at least a classification prediction model based on the ResNet model and a classification prediction model based on the Transformer model;

[0107] The model training module is used to train the first semantic segmentation model, the second semantic segmentation model, the first prediction model and the second prediction model respectively;

[0108] The model application module is used to receive a first CT image and a first prediction mode after the training of the four types of models is completed; and identify the first prediction mode; if the first prediction mode is the first mode, then predict the corresponding first prediction type based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the first prediction model; if the first prediction mode is the second mode, then predict the corresponding first prediction type based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the second prediction model; and output the obtained first prediction type as a prediction result corresponding to the first CT image; wherein, the first CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissue; the first prediction mode includes the first mode and the second mode; the first prediction type includes the M0 type and the M1 type.

[0109] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0110] The processor is configured to be coupled to the memory, read and execute instructions in the memory, so as to implement the method steps described in the first aspect above;

[0111] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0112] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a computer, the computer executes the instructions of the method described in the first aspect above.

[0113] The embodiments of the present invention provide a processing method, device, electronic device, and computer-readable storage medium for M-stage classification prediction based on CT images. As can be seen from the above content, the embodiments of the present invention construct two semantic segmentation models with reference to the model structure of the U-Net model, wherein the first semantic segmentation model is used to semantically segment the semantic points of the gastric tumor lesion area on the CT image, and the second semantic segmentation model is used to semantically segment the semantic points of the distal key tissue lesion area on the CT image, and construct a first prediction model that can perform M-stage classification prediction based on the input imaging omics features with reference to a class of machine learning classification models, and construct a second prediction model that can perform M-stage classification prediction based on the input three-dimensional CT image with lesion point markings with reference to a class of deep learning classification models; and train the above four types of models separately; and after the training of the above four types of models is completed, different prediction model combinations are selected based on the received prediction mode for prediction: when the prediction mode is the first mode, the first and second semantic segmentation models and the first prediction model are used to perform M-stage classification prediction on the received CT image; when the prediction mode is the second mode, the first, second semantic segmentation models and the second prediction model are used to perform M-stage classification prediction on the received CT image. The embodiments of the present invention improve both the real-time prediction and the prediction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] Figure 1 A schematic diagram of a processing method for M-stage classification prediction based on CT images provided in Example 1 of the present invention;

[0115] Figure 2 This is a module structure diagram of a processing device for performing M-stage classification prediction based on CT images provided in the second embodiment of the present invention;

[0116] Figure 3 This is a structural diagram of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0117] To make the objectives, technical solutions, and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the embodiments described herein are merely some, rather than all, of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0118] The first embodiment of the present invention provides a processing method for M-stage classification prediction based on CT images, such as Figure 1 A schematic diagram of a processing method for M-stage classification prediction based on CT images provided in the first embodiment of the present invention is shown. The method mainly includes the following steps:

[0119] Step 1: Based on the U-Net model, a semantic segmentation model is constructed to perform semantic point segmentation of the lesion area of ​​gastric tumors on CT images, which is recorded as the corresponding first semantic segmentation model; and based on the U-Net model, a semantic segmentation model is constructed to perform semantic point segmentation of the lesion area of ​​distal key tissues on CT images, which is recorded as the corresponding second semantic segmentation model; and based on a class of machine learning classification models, a prediction model is constructed to perform M-stage classification prediction based on imaging omics features, which is recorded as the corresponding first prediction model; and based on a class of deep learning classification models, a prediction model is constructed to perform M-stage classification prediction based on a three-dimensional CT image with lesion point markings, which is recorded as the corresponding second prediction model.

[0120] Here, the CT image of the embodiment of the present invention is a three-dimensional abdominal CT image with three-dimensional information of the stomach and distal key tissues; wherein, the distal key tissues include at least the liver, the para-aortic lymph node area, the retropancreatic lymph node area, the mesenteric root lymph node area and the peritoneum. The types of machine learning classification models of the embodiment of the present invention 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 types of deep learning classification models of the embodiment of the present invention include at least a classification prediction model based on the ResNet model and a classification prediction model based on the Transformer model.

[0121] Step 2: training the first semantic segmentation model, the second semantic segmentation model, the first prediction model, and the second prediction model respectively;

[0122] Specifically comprising: step 21, recruiting a plurality of subjects without gastric cancer, a plurality of subjects with gastric cancer but without metastasis to any distal key tissue, and a plurality of subjects with gastric cancer and metastasis to one or more types of distal key tissues to form a first subject set;

[0123] The first subject set includes multiple first subjects, each first subject corresponds to a first M stage type; the first M stage type includes M0 type and M1 type; the first M stage type corresponding to the first subject who does not have gastric cancer or has gastric cancer but has not metastasized to any distal key tissue is M0 type, and the first M stage type corresponding to the first subject who has gastric cancer and has metastasized to one or more types of distal key tissues is M1 type;

[0124] Step 22: Acquire the most recent CT image of each first subject as a corresponding first sample image; and form a corresponding first original data record from each first sample image and the corresponding first M-phase type; and form a corresponding first original data set from all the obtained first original data records;

[0125] The first sample image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and distal key tissues; the first sample image includes a plurality of first sample voxel points; each first sample voxel point corresponds to a first voxel point pixel value;

[0126] The first original data set obtained in the embodiment of the present invention includes a plurality of first original data records; the first original data record includes a first sample image and a corresponding first M-phase type; the first M-phase type includes an M0 type and an M1 type;

[0127] Step 23: construct four types of model training data sets based on the first original data set and record them as the corresponding first model data set, second model data set, third model data set and fourth model data set;

[0128] Specifically, the step 2301 includes: taking each first original data record of the first original data set as the corresponding current original data record; and taking the first sample image and the corresponding first M-phase type of the current original data record as the corresponding current sample image and current sample type;

[0129] Step 2302: The current sample image is used as the corresponding first training image and second training image; corresponding first and second label semantic maps are constructed based on the three-dimensional dimensions of the first and second training images; and the first and second label semantics of all first and second label voxel points in the first and second label semantic maps are initialized to 0;

[0130] Step 2303 , based on manual labeling or other machine labeling methods, all first sample voxel points located in the gastric tumor lesion area on the current sample image are marked as Class I voxel points, and all first sample voxel points located in all distal key tissue lesion areas on the current sample image are marked as Class II voxel points;

[0131] Step 2304: When the total number of voxel points of a type is zero, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets is set to form a corresponding first gastric tumor feature set;

[0132] In step 2305, when the total number of voxel points of the first class is greater than zero, the first label semantics of the first label voxel points corresponding to each voxel point of the first class on the first label semantic map are reset to 1; and the three-dimensional structure composed of all the voxel points of the first class on the current sample image is recorded as the corresponding structure of the first class; and the first-order features, shape features, and texture features of the imaging omics features of the structure of the first class are calculated to obtain the corresponding first-order feature set, shape feature set, and texture feature set to form a corresponding first gastric tumor feature set;

[0133] Step 2306: When the total number of the second-class voxel points is zero, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets is set for each distal key tissue to form a corresponding first distal key tissue feature set; and all the obtained first distal key tissue feature sets are used to form a corresponding first distal key tissue feature set;

[0134] Step 2307: When the total number of the second-class voxel points is greater than zero, the second label semantics of the second-class voxel points corresponding to each second-class voxel point on the second label semantic map are reset to 1; and based on the preset point group clustering algorithm, the second-class voxel points corresponding to each distal key tissue on the current sample image are clustered to obtain a corresponding first point group; and each first point group is used as the corresponding current point group, and whether the current point group is empty is identified; if the current point group is empty, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets are set for the distal key tissue corresponding to the current point group to form a corresponding first distal key tissue feature set; if the current point group is not empty, the three-dimensional structure formed by the current point group on the current sample image is recorded as the corresponding second-class structure, and the first-order features, shape features, and texture features of the imaging omics features of the second-class structure are calculated to obtain the corresponding first-order feature set, shape feature set, and texture feature set to form a corresponding first distal key tissue feature set; and all the obtained first distal key tissue feature sets are used to form a corresponding first distal key tissue feature set;

[0135] Here, the calculation methods for the first-order features, shape features, and texture features of the radiomics features mentioned in the embodiments of the present invention are three types of public radiomics feature calculation methods. The calculation details can be obtained by consulting public technical literature and are not further described here. The point group clustering algorithm in the embodiments of the present invention is a conventional clustering algorithm, such as the K-means clustering algorithm and the DBSCAN clustering algorithm, and the algorithm can be configured according to actual application requirements.

[0136] Step 2308 , forming a corresponding first training radiomics feature set from the first gastric tumor feature set and the first distal key tissue feature set corresponding to the current sample image;

[0137] Step 2309, each first sample voxel point of the current sample image is used as the corresponding current sample voxel point; the first and second label voxel points corresponding to the current sample voxel point on the first and second label semantic maps corresponding to the current sample image are recorded as the corresponding first and second matching voxel points; the first voxel point pixel value of the current sample voxel point is used as the corresponding second voxel point pixel value; and when the first and second label semantics corresponding to the first and second matching voxel points are both 0, the corresponding second voxel point semantics is set to 0; and when the first label semantics corresponding to the first matching voxel point is 1, the corresponding second voxel point semantics is set to 1; and when the second label semantics corresponding to the second matching voxel point is 1, the corresponding second voxel point semantics is set to 2; and based on the three-dimensional voxel coordinates of the current sample voxel point, a corresponding first feature voxel point is constructed and recorded as the corresponding current feature voxel point; and a second voxel point feature vector corresponding to the current feature voxel point is formed by the second voxel point pixel value and the second voxel point semantics corresponding to the current sample voxel point; and a corresponding first training feature map is formed by all the first feature voxel points corresponding to the current sample image;

[0138] Here, the first training feature map of the embodiment of the present invention is essentially a three-dimensional CT image with lesion point markings;

[0139] Step 2310 , construct corresponding first and second label type vectors according to the current sample type;

[0140] Among them, when the current sample type is type M0, the corresponding first and second label type probabilities corresponding to type M0 in the corresponding first and second label type vectors are 1, and the first and second label type probabilities corresponding to type M1 are 0; when the current sample type is type M1, the corresponding first and second label type probabilities corresponding to type M1 in the corresponding first and second label type vectors are 0, and the first and second label type probabilities corresponding to type M1 are 1;

[0141] Step 2311, a first model data record is formed by the first training image and the first label semantic map corresponding to the current raw data record; a second model data record is formed by the second training image and the second label semantic map corresponding to the current raw data record; a third model data record is formed by the first training radiomics feature set and the first label type vector corresponding to the current raw data record; and a fourth model data record is formed by the first training feature map and the second label type vector corresponding to the current raw data record.

[0142] Step 2312: All first model data records corresponding to the first original data set form a corresponding first model data set; all second model data records corresponding to the first original data set form a corresponding second model data set; all third model data records corresponding to the first original data set form a corresponding third model data set; and all fourth model data records corresponding to the first original data set form a corresponding fourth model data set.

[0143] Here, the first model data set obtained in the embodiment of the present invention includes multiple first model data records; the first model data record includes a first training image and a first label semantic map; the three-dimensional size of the first label semantic map is consistent with the three-dimensional size of the first training image; the first label semantic map includes multiple first label voxel points; each first label voxel point corresponds to a binary first label semantic value of 0 or 1; when the first label semantic value is 0, it indicates that the corresponding first label voxel point is a background voxel point, and when it is 1, it indicates that the corresponding first label voxel point is a gastric tumor lesion area voxel point;

[0144] The second model data set obtained in an embodiment of the present invention includes multiple second model data records; the second model data records include a second training image and a second label semantic map; the three-dimensional size of the second label semantic map is consistent with the three-dimensional size of the second training image; the second label semantic map includes multiple second label voxel points; each second label voxel point corresponds to a binary second label semantic value of 0 or 1; when the second label semantic value is 0, it indicates that the corresponding first label voxel point is a background voxel point, and when it is 1, it indicates that the corresponding second label voxel point is a distal key tissue lesion voxel point;

[0145] The third model data set obtained in the embodiment of the present invention includes multiple third model data records; the third model data record includes a first training imaging omics feature set and a first label type vector; the first training imaging omics feature set includes a first gastric tumor feature set and a first distal key tissue feature set; the first distal key tissue feature set includes multiple first distal key tissue feature sets, each first distal key tissue feature set corresponds to a distal key tissue; the first gastric tumor feature set and all first distal key tissue feature sets are composed of a set of first-order feature sets, shape feature sets and texture feature sets of imaging omics features; the first label type vector includes two first label type probabilities, each first label type probability corresponds to an M stage type; the M stage type includes M0 type and M1 type;

[0146] The fourth model data set obtained in the embodiment of the present invention includes multiple fourth model data records; the fourth model data record includes a first training feature map and a second label type vector; the first training feature map includes multiple first feature voxel points; each first feature voxel point corresponds to a second voxel point feature vector; the second voxel point feature vector includes at least a second voxel point pixel value and a second voxel point semantics; the second voxel point semantics is a three-valued semantic value, which takes a value of 0, 1 or 2; when the second voxel point semantics is 0, it indicates that the corresponding first feature voxel point is a background voxel point, when it is 1, it indicates that the corresponding first feature voxel point is a gastric tumor lesion area voxel point, and when it is 2, it indicates that the corresponding first feature voxel point is a distal key tissue lesion voxel point; the second label type vector includes two second label type probabilities, each second label type probability corresponds to an M stage type;

[0147] Step 24: training a first semantic segmentation model based on the first model dataset;

[0148] Specifically comprising: step 241, taking the first first model data record of the first model data set as the corresponding current model data record;

[0149] Step 242: Input the first training image recorded in the current model data into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding first predicted semantic map;

[0150] Step 243: Input the first predicted semantic graph and the first label semantic graph of the current model data record into a preset first model loss function to calculate and obtain a corresponding first loss value;

[0151] Wherein, the first model loss function is implemented based on the L1 loss function or the L2 loss function;

[0152] Step 244: Identify whether the first loss value satisfies a preset first loss value range. If the first loss value satisfies the first loss value range, identify whether the current model data record is the last first model data record of the first model data set. If so, proceed to step 245. If not, use the next first model data record of the first model data set as the new current model data record and return to step 242 to continue training. If the first loss value does not satisfy the first loss value range, perform a round of model parameter optimization on the first semantic segmentation model based on a preset first model parameter optimizer in a direction to minimize the first model loss function. After this round of parameter optimization is completed, return to step 242 to continue training.

[0153] Wherein, the first model parameter optimizer includes at least an SDG optimizer; the first loss value range is a preset loss value range;

[0154] Step 245 , solidifying the model parameters of the first semantic segmentation model and confirming that the model training of the first semantic segmentation model is completed;

[0155] Step 25: training a second semantic segmentation model based on the second model dataset;

[0156] Specifically comprising: step 251, taking the first second model data record of the second model data set as the corresponding current model data record;

[0157] Step 252: Input the second training image recorded in the current model data into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding second predicted semantic map;

[0158] Step 253: Input the second predicted semantic graph and the second label semantic graph of the current model data record into a preset second model loss function to calculate and obtain a corresponding second loss value;

[0159] The second model loss function is implemented based on the L1 loss function or the L2 loss function;

[0160] Step 254: Identify whether the second loss value satisfies a preset second loss value range. If the second loss value satisfies the second loss value range, identify whether the current model data record is the last second model data record of the second model dataset. If so, proceed to step 255. If not, use the next second model data record of the second model dataset as the new current model data record and return to step 252 to continue training. If the second loss value does not satisfy the second loss value range, perform a round of model parameter optimization on the second semantic segmentation model based on a preset second model parameter optimizer in a direction to minimize the second model loss function. After this round of parameter optimization is completed, return to step 252 to continue training.

[0161] The second model parameter optimizer includes at least an SDG optimizer; the second loss value range is a preset loss value range;

[0162] Step 255: solidify the model parameters of the second semantic segmentation model and confirm that the model training of the second semantic segmentation model is completed;

[0163] Step 26: training the first prediction model based on the third model data set;

[0164] Specifically, the method includes: step 2601, performing a full combination of hyperparameters of the first prediction model to obtain a plurality of corresponding first hyperparameter combinations; and performing a hyperparameter setting on the first prediction model based on each first hyperparameter combination to obtain a corresponding first prediction model;

[0165] Step 2602: Set the parameter K to a positive integer greater than or equal to 5; divide the third model dataset into K sub-datasets and record each sub-dataset as the corresponding first sub-dataset D i ;

[0166] Where 1≤index i≤K;

[0167] Step 2603: any first sub-dataset D i As a corresponding first validation dataset DB i , and except the current first sub-dataset D i The remaining K-1 first sub-datasets D i Merge into a corresponding first test dataset DA i ; and the first test data sets DA obtained i and the corresponding first validation dataset DB i Form a corresponding first test validation set;

[0168] Step 2604: Use the first prediction model as the corresponding current prediction model;

[0169] Step 2605: Use the first test validation set as the corresponding current test validation set;

[0170] Step 2606: The first test data set DA of the current test validation set is i and the first validation dataset DB i As the corresponding current test dataset and current validation dataset;

[0171] Step 2607: Use the first third model data record of the current test data set as the corresponding current model data record;

[0172] Step 2608 , inputting the first training radiomics feature set recorded in the current model data into the current prediction model to perform M-stage classification prediction to obtain a corresponding first prediction vector;

[0173] The first prediction vector includes two first-type prediction probabilities, each of which corresponds to an M stage type;

[0174] Step 2609: Substitute the first prediction vector and the first label type vector of the current model data record into a preset third model loss function; and perform a round of model parameter optimization on the model parameters of the current prediction model in a direction that minimizes the third model loss function based on a preset third model parameter optimizer; and at the end of this round of parameter optimization, identify whether the current model data record is the last third model data record of the current test data set; if so, go to step 2610; if not, extract the next third model data record of the current test data set as the new current model data record and return to step 2608 for training;

[0175] The third model loss function is implemented based on a binary cross entropy loss function; the third model parameter optimizer includes at least an SDG optimizer;

[0176] Step 2610: Set three counters initially set to 0 as the corresponding TP counter, FP counter, and FN counter;

[0177] Step 2611, traverse all third model data records in the current verification data set; and during the traversal, use the currently traversed third model data record as the corresponding current verification record; and use the first training radiomics feature set and the first label type vector of the current verification record as the corresponding current radiomics feature set and the current label type vector; and use the M stage type corresponding to the first label type probability with a probability value of 1 in the current label type vector as the corresponding current label type; and input the current radiomics feature set into the current prediction model to perform M stage classification prediction to obtain the corresponding second prediction vector; and use the M stage type corresponding to the second type prediction probability with the largest probability in the second prediction vector as the corresponding current prediction type; and compare the current prediction type with the current label type; if the current prediction type and the current label type are both M1 type, add 1 to the TP counter; if the current prediction type is M1 type and the current label type is M0 type, add 1 to the FP counter; if the current prediction type is M0 type and the current label type is M1 type, add 1 to the FN counter;

[0178] The second prediction vector includes two second-type prediction probabilities, each second-type prediction probability corresponds to an M stage type;

[0179] Step 2612: After traversing all the third model data records in the current validation data set, the precision and recall are calculated based on the TP counter, FP counter, and FN counter to obtain the corresponding first precision and first recall; and the F1 score is calculated based on the first precision and the first recall to obtain the corresponding first score F1. i ;

[0180] here,

[0181] First accuracy

[0182] First recall rate

[0183] First score

[0184] Step 2613: Identify whether the current test validation set is the Kth first test validation set; if so, go to step 2614; if not, use the next first test validation set as the new current test validation set and return to step 2606;

[0185] Step 2614: Based on the obtained K first scores F1 i Perform mean square error calculation to obtain the corresponding first error E;

[0186] here,

[0187] Step 2615: Identify whether the current prediction model is the last first-one prediction model; if so, go to step 2616; if not, use the next first-one prediction model as the new current prediction model and return to step 2605;

[0188] Step 2616: The full model parameter set of the first prediction model corresponding to the first error E with the smallest error is used as the corresponding optimal full model parameter set;

[0189] Step 2617: solidify the model parameters of the first prediction model based on the optimal full model parameter set and confirm that the model training of the first prediction model is completed;

[0190] Step 27: training the second prediction model based on the fourth model data set;

[0191] Specifically comprising: step 271, taking the first fourth model data record of the fourth model data set as the corresponding current model data record;

[0192] Step 272: Input the first training feature map of the current model data record into the second prediction model to perform M-period classification prediction to obtain a corresponding third prediction vector;

[0193] The third prediction vector includes two third-type prediction probabilities, each of which corresponds to an M stage type;

[0194] Step 273: Input the third prediction vector and the second label type vector of the current model data record into a preset fourth model loss function to calculate and obtain a corresponding third loss value;

[0195] Among them, the fourth model loss function is implemented based on the binary cross entropy loss function;

[0196] Step 274: Identify whether the third loss value satisfies a preset third loss value range. If the third loss value satisfies the third loss value range, identify whether the current model data record is the last fourth model data record of the fourth model data set. If so, proceed to step 275. If not, use the next fourth model data record of the fourth model data set as the new current model data record and return to step 272 to continue training. If the third loss value does not satisfy the third loss value range, perform a round of model parameter optimization on the second prediction model based on a preset fourth model parameter optimizer in a direction to minimize the fourth model loss function. When this round of parameter optimization is completed, return to step 272 to continue training.

[0197] Wherein, the fourth model parameter optimizer includes at least an SDG optimizer; the third loss value range is a preset loss value range;

[0198] Step 275: solidify the model parameters of the second prediction model and confirm that the model training of the second prediction model is completed;

[0199] In step 28, after the first and second semantic segmentation models and the first and second prediction models have completed model training, it is confirmed that the training of the four types of models is completed.

[0200] Step 3: After the training of the four types of models is completed, the first CT image and the first prediction mode are received; the first prediction mode is identified; if the first prediction mode is the first mode, a corresponding first prediction type is obtained by predicting based on the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the first prediction model; if the first prediction mode is the second mode, a corresponding first prediction type is obtained by predicting based on the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the second prediction model; and the obtained first prediction type is output as a prediction result corresponding to the first CT image;

[0201] The first CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and distal key tissues; the first CT image includes a plurality of first voxel points; each first voxel point corresponds to a first voxel point pixel value; the first prediction mode includes a first mode and a second mode; the first prediction type includes an M0 type and an M1 type;

[0202] Specifically comprising: step 31, after the four types of model training are completed, receiving a first CT image and a first prediction mode;

[0203] Step 32, identifying the first prediction mode;

[0204] Step 33: If the first prediction mode is the first mode, performing prediction based on the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the first prediction model to obtain a corresponding first prediction type;

[0205] Specifically, the step 331 includes: inputting the first CT image into a first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding first semantic map;

[0206] The three-dimensional size of the first semantic map is consistent with the three-dimensional size of the first CT image; the first semantic map includes a plurality of second voxel points, and the first and second voxel points correspond one to one; each second voxel point corresponds to a binary gastric tumor lesion semantic point, with a value of 0 or 1; when the gastric tumor lesion semantic point is 0, it indicates that the corresponding second voxel point is a background voxel point, and when it is 1, it indicates that the corresponding second voxel point is a gastric tumor lesion voxel point;

[0207] Step 332: Count the number of second voxel points with semantic value 1 in the gastric tumor lesion area on the first semantic map to obtain a corresponding first number; determine whether the first number exceeds a preset first number threshold; if so, proceed to step 333; if not, set the corresponding first prediction type to type M0 and proceed to step 339;

[0208] Here, the first quantity threshold is a preset integer value;

[0209] Step 333: Input the first CT image into a second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding second semantic map;

[0210] The three-dimensional size of the second semantic map is consistent with the three-dimensional size of the first CT image; the second semantic map includes multiple third voxel points, and the first and third voxel points correspond one to one; each third voxel point corresponds to a binary remote critical tissue lesion area semantic, with a value of 0 or 1; when the remote critical tissue lesion area semantic is 0, it indicates that the corresponding third voxel point is a background voxel point, and when it is 1, it indicates that the corresponding third voxel point is a remote critical tissue lesion voxel point;

[0211] In step 334, the number of third voxel points with a semantic value of 1 in the distal key tissue lesion area on the second semantic map is counted to obtain a corresponding second number; and whether the second number exceeds a preset second number threshold is identified. If so, the process proceeds to step 335; if not, the corresponding first prediction type is set to type M0 and the process proceeds to step 339.

[0212] Here, the second quantity threshold is a preset integer value;

[0213] Step 335: The first voxel point corresponding to the second voxel point with a semantic value of 1 in each gastric tumor lesion area on the first CT image is recorded as the corresponding first lesion point; and the three-dimensional structure composed of all the first lesion points is recorded as the corresponding first structure; and the first-order features, shape features, and texture features of the radiomics features of the first structure are calculated to obtain a corresponding first-order feature set, shape feature set, and texture feature set to form a corresponding second gastric tumor feature set;

[0214] Step 336: Cluster the first voxel points corresponding to each distal key tissue on the first CT image based on a preset point group clustering algorithm to obtain a corresponding second point group; and use each second point group as the corresponding current point group, and identify whether the current point group is empty; if the current point group is empty, set a set of all-zero first-order feature sets, shape feature sets, and texture feature sets for the distal key tissue corresponding to the current point group to form a corresponding second distal key tissue feature set; if the current point group is not empty, record the three-dimensional structure formed by the current point group on the first CT image as the corresponding second structure, and calculate the first-order features, shape features, and texture features of the imaging omics features of the second structure to obtain the corresponding first-order feature set, shape feature set, and texture feature set to form a corresponding second distal key tissue feature set; and form a corresponding second distal key tissue feature set from all the obtained second distal key tissue feature sets;

[0215] Step 337 , composing a corresponding first radiomics feature set from the obtained second gastric tumor feature set and the second distal key tissue feature set; and inputting the first radiomics feature set into the first prediction model for M-stage classification prediction to obtain a corresponding fourth prediction vector;

[0216] The fourth prediction vector includes two fourth-type prediction probabilities, each fourth-type prediction probability corresponds to an M stage type;

[0217] Step 338: taking the M stage type corresponding to the fourth type prediction probability with the largest probability in the fourth prediction vector as the corresponding first prediction type;

[0218] Step 339: output the obtained first prediction type as the current prediction result;

[0219] Step 34: If the first prediction mode is the second mode, a corresponding first prediction type is obtained by performing prediction based on the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the second prediction model;

[0220] Specifically, the step 341 includes: inputting the first CT image into a first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding third semantic map;

[0221] The three-dimensional size of the third semantic map is consistent with the three-dimensional size of the first CT image; the third semantic map includes multiple fourth voxel points, and the first and fourth voxel points correspond one to one; each fourth voxel point corresponds to a binary gastric tumor lesion semantic point, with a value of 0 or 1; when the gastric tumor lesion semantic point is 0, it indicates that the corresponding fourth voxel point is a background voxel point, and when it is 1, it indicates that the corresponding fourth voxel point is a gastric tumor lesion voxel point;

[0222] Step 342: Count the number of fourth voxel points with semantic value 1 in the gastric tumor lesion area on the third semantic map to obtain a corresponding third number; determine whether the third number exceeds a preset first number threshold. If so, proceed to step 343; if not, set the corresponding first prediction type to type M0 and proceed to step 348.

[0223] Step 343 , inputting the first CT image into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding fourth semantic map;

[0224] The three-dimensional size of the fourth semantic map is consistent with the three-dimensional size of the first CT image; the fourth semantic map includes multiple fifth voxel points, and the first and fifth voxel points correspond one to one; each fifth voxel point corresponds to a binary distal key tissue lesion area semantic point, with a value of 0 or 1; when the distal key tissue lesion area semantic point is 0, it indicates that the corresponding fifth voxel point is a background voxel point, and when it is 1, it indicates that the corresponding fifth voxel point is a distal key tissue lesion voxel point;

[0225] In step 344, the number of fifth voxel points in the distal key tissue lesion region with a semantic value of 1 on the second semantic map is counted to obtain a corresponding fourth number; and whether the fourth number exceeds a preset second number threshold is identified. If so, the process proceeds to step 345; if not, the corresponding first prediction type is set to type M0 and the process proceeds to step 348.

[0226] Step 345: The first voxel points of each first CT image are taken as the corresponding current voxel points; the fourth and fifth voxel points corresponding to the current voxel points on the third and fourth semantic maps are recorded as the corresponding third and fourth matching voxel points; the first voxel point pixel value of the current voxel point is taken as the corresponding feature voxel point pixel value; and when the semantics of the gastric tumor lesion area and the distal key tissue lesion area corresponding to the third and fourth matching voxel points are both 0, the semantics of a corresponding feature voxel point is set to 0; and when the semantics of the gastric tumor lesion area corresponding to the third matching voxel point is 1, set a corresponding feature voxel point semantics to 1; and set a corresponding feature voxel point semantics to 2 when the semantics of the distal key tissue lesion area corresponding to the fourth matching voxel point is 1; and construct a corresponding second feature voxel point based on the three-dimensional voxel coordinates of the current voxel point and record it as the corresponding current feature voxel point; and form a feature voxel point feature vector corresponding to the current feature voxel point by using the feature voxel point pixel value and the feature voxel point semantics corresponding to the current voxel point; and form a corresponding first feature map by using all the second feature voxel points corresponding to the first CT image;

[0227] Here, the first characteristic image of the embodiment of the present invention is essentially a three-dimensional CT image with lesion point markings;

[0228] Step 346: Input the first feature map into the second prediction model to perform M-period classification prediction to obtain a corresponding fifth prediction vector;

[0229] The fifth prediction vector includes two fifth-type prediction probabilities, each fifth-type prediction probability corresponds to an M stage type;

[0230] Step 347: taking the M stage type corresponding to the fifth type prediction probability with the largest probability in the fifth prediction vector as the corresponding first prediction type;

[0231] Step 348: Output the obtained first prediction type as the current prediction result;

[0232] In step 35, the obtained first prediction type is output as a prediction result corresponding to the first CT image.

[0233] Figure 2 This is a module structure diagram of a processing device for M-stage classification prediction based on CT images provided in the second embodiment of the present invention. The device is a terminal device or server that implements the aforementioned method embodiment, and can also be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 2 As shown, the device includes: a model construction module 201, a model training module 202 and a model application module 203.

[0234] The model construction module 201 is used to construct a semantic segmentation model for semantic point segmentation of gastric tumor lesions on CT images based on the U-Net model, which is recorded as the corresponding first semantic segmentation model; and to construct a semantic segmentation model for semantic point segmentation of lesion areas of distal key tissues on CT images based on the U-Net model, which is recorded as the corresponding second semantic segmentation model; and to construct a prediction model for M-stage classification prediction based on imaging omics features based on a class of machine learning classification models, which is recorded as the corresponding first prediction model; and to construct a prediction model for M-stage classification prediction based on a three-dimensional CT image with lesion point markings based on a class of deep learning classification models, which is recorded as the corresponding second prediction model. model; wherein the CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and distal key tissues; the distal key tissues include at least the liver, para-aortic lymph node area, retropancreatic lymph node area, mesenteric root lymph node area and peritoneum; 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 types of deep learning classification models include at least a classification prediction model based on the ResNet model and a classification prediction model based on the Transformer model.

[0235] The model training module 202 is used to train the first semantic segmentation model, the second semantic segmentation model, the first prediction model and the second prediction model respectively.

[0236] The model application module 203 is used to receive the first CT image and the first prediction mode after the training of the four types of models is completed; and identify the first prediction mode; if the first prediction mode is the first mode, then the corresponding first prediction type is predicted based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the first prediction model; if the first prediction mode is the second mode, then the corresponding first prediction type is predicted based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the second prediction model; and the obtained first prediction type is output as the prediction result corresponding to the first CT image; wherein the first CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and distal key tissues; the first prediction mode includes the first mode and the second mode; the first prediction type includes the M0 type and the M1 type.

[0237] An embodiment of the present invention provides a processing device for M-phase classification prediction based on CT images, which can execute the method steps in the above method embodiment. Its implementation principles and technical effects are similar and will not be repeated here.

[0238] It should be noted that the division of the modules of the above devices is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into a physical entity or physically separated. Furthermore, these modules can all be implemented in the form of software called by a processing element; or all be implemented in the form of hardware; or some modules can be implemented in the form of software called by a processing element, and some modules can be implemented in the form of hardware. For example, the model building module can be a separate processing element, or it can be integrated into a chip of the above device. In addition, it can be stored in the form of program code in the memory of the above device, and called by a processing element of the above device to perform the functions of the above-mentioned module. The implementation of other modules is similar. In addition, these modules can all or partly be integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed by hardware integrated logic circuits in the processor element or instructions in the form of software.

[0239] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented by scheduling program code through a processing element, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0240] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the above method embodiments are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) means. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, hard disk, tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0241] Figure 3 This is a schematic diagram of the structure of an electronic device provided in the third embodiment of the present invention. The electronic device can be a terminal device or server that implements the method of the aforementioned embodiment, or it can be a terminal device or server that implements the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 3 As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303's transceiver actions. Various instructions may be stored in the memory 302 for completing various processing functions and implementing the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment 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 above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0242] exist Figure 3The system bus 305 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory (Non-Volatile Memory), such as at least one disk storage.

[0243] The above-mentioned processors 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.

[0244] It should be noted that an embodiment of the present invention further provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the methods and processing procedures provided in the above embodiments.

[0245] An embodiment of the present invention further provides a chip for executing instructions, which is used to execute the processing steps described in the above method embodiment.

[0246] The embodiments of the present invention provide a processing method, device, electronic device, and computer-readable storage medium for M-stage classification prediction based on CT images. As can be seen from the above content, the embodiments of the present invention construct two semantic segmentation models with reference to the model structure of the U-Net model, wherein the first semantic segmentation model is used to semantically segment the semantic points of the gastric tumor lesion area on the CT image, and the second semantic segmentation model is used to semantically segment the semantic points of the distal key tissue lesion area on the CT image, and construct a first prediction model that can perform M-stage classification prediction based on the input imaging omics features with reference to a class of machine learning classification models, and construct a second prediction model that can perform M-stage classification prediction based on the input three-dimensional CT image with lesion point markings with reference to a class of deep learning classification models; and train the above four types of models separately; and after the training of the above four types of models is completed, different prediction model combinations are selected based on the received prediction mode for prediction: when the prediction mode is the first mode, the first and second semantic segmentation models and the first prediction model are used to perform M-stage classification prediction on the received CT image; when the prediction mode is the second mode, the first, second semantic segmentation models and the second prediction model are used to perform M-stage classification prediction on the received CT image. The embodiments of the present invention improve both the real-time prediction and the prediction efficiency.

[0247] Professionals should also be further aware that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0248] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0249] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A processing method for M-stage classification prediction based on CT images, characterized in that: The method comprises: A first semantic segmentation model for performing semantic point segmentation of gastric tumor lesions on CT images is constructed based on a U-Net model; a second semantic segmentation model for performing semantic point segmentation of lesions on distal key tissues on the CT images is constructed based on the U-Net model; a first prediction model for performing M-stage classification prediction based on imaging omics features is constructed based on a machine learning classification model; and a second prediction model for performing M-stage classification prediction based on a three-dimensional CT image with lesion point markings is constructed based on a deep learning classification model; wherein the CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissues; Training the first semantic segmentation model, the second semantic segmentation model, the first prediction model, and the second prediction model respectively; After the four types of model training are completed, a first CT image and a first prediction mode are received, and the first prediction mode is identified; If the first prediction mode is the second mode, the corresponding first prediction type is obtained by prediction based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the second prediction model, including: inputting the first CT image into the first and second semantic segmentation models for segmentation to obtain third and fourth semantic maps; and taking each first voxel point of the first CT image as the current voxel point, and recording the voxel point corresponding to the current voxel point on the third and fourth semantic maps as the third and fourth matching voxel points, and taking the voxel point pixel value of the current voxel point as the corresponding feature voxel point pixel value, and setting the corresponding feature voxel point semantics to 0 when the gastric tumor lesion area semantics and the distal key tissue lesion area semantics corresponding to the third and fourth matching voxel points are both 0, and setting the corresponding feature voxel point semantics to 1 when the gastric tumor lesion area semantics corresponding to the third matching voxel point is 1. The semantic value of the characteristic voxel point is 1, and when the semantic value of the distal key tissue lesion area corresponding to the fourth matching voxel point is 1, the semantic value of the corresponding characteristic voxel point is set to 2, and a corresponding second characteristic voxel point is constructed based on the three-dimensional voxel coordinates of the current voxel point and recorded as the current characteristic voxel point, and a characteristic voxel point feature vector corresponding to the current characteristic voxel point is composed of the characteristic voxel point pixel value corresponding to the current voxel point and the characteristic voxel point semantics, and a first characteristic map is composed of all the second characteristic voxel points corresponding to the first CT image; and the first characteristic map is input into the second prediction model for M stage classification prediction to obtain a corresponding fifth prediction vector, the fifth prediction vector including two fifth type prediction probabilities; and the M stage type corresponding to the fifth type prediction probability with the largest probability in the fifth prediction vector is used as the corresponding first prediction type; The obtained first prediction type is output as a prediction result corresponding to the first CT image; the first prediction type includes M0 type and M1 type.

2. The method for M-stage classification prediction based on CT images according to claim 1, characterized in that: The training of the first semantic segmentation model, the second semantic segmentation model, the first prediction model, and the second prediction model respectively specifically includes: A first subject set is formed by recruiting a plurality of subjects who do not have gastric cancer, a plurality of subjects who have gastric cancer but have not metastasized to any of the distal key tissues, and a plurality of subjects who have gastric cancer and have metastasized to one or more types of tissues in the distal key tissues; the first subject set includes a plurality of first subjects, each of the first subjects corresponds to a first M stage type; the first M stage type includes an M0 type and an M1 type; the first subject who does not have gastric cancer or has gastric cancer but has not metastasized to any of the distal key tissues corresponds to the M0 type, and the first subject who has gastric cancer and has metastasized to one or more types of tissues in the distal key tissues corresponds to the M1 type; Acquire the most recent CT image of each of the first subjects as a corresponding first sample image; form a corresponding first original data record from each of the first sample images and the corresponding first M-phase type; and form a corresponding first original data set from all of the obtained first original data records; the first sample image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissue; the first sample image includes a plurality of first sample voxel points; each of the first sample voxel points corresponds to a first voxel point pixel value; Constructing four types of model training data sets based on the first original data set, which are recorded as the corresponding first model data set, second model data set, third model data set and fourth model data set; Training the first semantic segmentation model based on the first model dataset; Training the second semantic segmentation model based on the second model dataset; Training the first prediction model based on the third model dataset; Training the second prediction model based on the fourth model dataset; And after the first and second semantic segmentation models and the first and second prediction models have completed model training, it is confirmed that the four-category model training is completed.

3. The method for M-stage classification prediction based on CT images according to claim 2, characterized in that: The first model data set includes multiple first model data records; the first model data record includes a first training image and a first label semantic map; the three-dimensional size of the first label semantic map is consistent with the three-dimensional size of the first training image; the first label semantic map includes multiple first label voxel points; each first label voxel point corresponds to a binary first label semantic value of 0 or 1; when the first label semantic value is 0, it indicates that the corresponding first label voxel point is a background voxel point, and when it is 1, it indicates that the corresponding first label voxel point is a gastric tumor lesion area voxel point; The second model data set includes a plurality of second model data records; the second model data record includes a second training image and a second label semantic map; the three-dimensional size of the second label semantic map is consistent with the three-dimensional size of the second training image; The second label semantic map includes a plurality of second label voxel points; each second label voxel point corresponds to a binary second label semantic point, which takes a value of 0 or 1; when the second label semantic point is 0, it indicates that the corresponding first label voxel point is a background voxel point, and when it is 1, it indicates that the corresponding second label voxel point is a distal key tissue lesion voxel point; The third model data set includes multiple third model data records; the third model data record includes a first training imaging omics feature set and a first label type vector; the first training imaging omics feature set includes a first gastric tumor feature set and a first distal key tissue feature set; the first distal key tissue feature set includes multiple first distal key tissue feature sets, each of the first distal key tissue feature sets corresponds to one distal key tissue; the first gastric tumor feature set and all the first distal key tissue feature sets are composed of a set of first-order feature sets, shape feature sets and texture feature sets of imaging omics features; the first label type vector includes two first label type probabilities, each of the first label type probability corresponds to an M stage type; the M stage type includes M0 type and M1 type; The fourth model data set includes a plurality of fourth model data records; the fourth model data record includes a first training feature map and a second label type vector; the first training feature map includes a plurality of first feature voxel points; each of the first feature voxel points corresponds to a second voxel point feature vector; the second voxel point feature vector includes at least a second voxel point pixel value and a second voxel point semantics; The second voxel point semantics is a three-valued semantic value, which takes a value of 0, 1 or 2; when the second voxel point semantics is 0, it indicates that the corresponding first feature voxel point is a background voxel point; when it is 1, it indicates that the corresponding first feature voxel point is a gastric tumor lesion area voxel point; when it is 2, it indicates that the corresponding first feature voxel point is a distal key tissue lesion voxel point; the second label type vector includes two second label type probabilities, each of which corresponds to one of the M stage types.

4. The method for M-stage classification prediction based on CT images according to claim 3, characterized in that: The four types of model training data sets constructed based on the first original data set are recorded as the corresponding first model data set, second model data set, third model data set and fourth model data set, specifically including: using each of the first raw data records of the first raw data set as a corresponding current raw data record; and using the first sample image and the corresponding first M-phase type of the current raw data record as a corresponding current sample image and current sample type; and using the current sample image as the corresponding first training image and the second training image; and constructing the corresponding first and second label semantic maps based on the three-dimensional sizes of the first and second training images; and initializing the first and second label semantics of all the first and second label voxel points of the first and second label semantic maps to 0; and marking all the first sample voxel points located in the gastric tumor lesion area on the current sample image as type I voxel points based on manual labeling or other machine labeling methods, and marking all the first sample voxel points located in all distal key tissue lesion areas on the current sample image as type II voxel points; When the total number of the first type of voxel points is zero, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets is set to form a corresponding first gastric tumor feature set; and when the total number of the first type of voxel points is greater than zero, resetting the first label semantics of the first label voxel points corresponding to each of the first type of voxel points on the first label semantic map to 1; and recording the three-dimensional structure composed of all the first type of voxel points on the current sample image as the corresponding first type of structure; and calculating the first-order features, shape features, and texture features of the imaging omics features of the first type of structure to obtain a corresponding first-order feature set, shape feature set, and texture feature set to form a corresponding first gastric tumor feature set; When the total number of the two types of voxel points is zero, a set of all-zero first-order feature sets, shape feature sets, and texture feature sets is set for each of the distal key tissues to form a corresponding first distal key tissue feature set; and all the obtained first distal key tissue feature sets are used to form a corresponding first distal key tissue feature set; And when the total number of the second-category voxel points is greater than zero, the second label semantics of the second-category voxel points corresponding to each of the second-category voxel points on the second label semantic map are reset to 1; and based on the preset point group clustering algorithm, the second-category voxel points corresponding to each of the distal key tissues on the current sample image are clustered to obtain the corresponding first point group; and each of the first point groups is used as the corresponding current point group, and whether the current point group is empty is identified; if the current point group is empty, a set of all-zero first-order feature sets is set for the distal key tissue corresponding to the current point group, The shape feature set and the texture feature set constitute a corresponding first distal key tissue feature set; if the current point group is not empty, the three-dimensional structure formed by the current point group on the current sample image is recorded as the corresponding second-class structure, and the first-order features, shape features, and texture features of the imaging omics features of the second-class structure are calculated to obtain the corresponding first-order feature set, shape feature set, and texture feature set to constitute a corresponding first distal key tissue feature set; and all the obtained first distal key tissue feature sets constitute a corresponding first distal key tissue feature set; and forming a corresponding first training radiomics feature set from the first gastric tumor feature set and the first distal key tissue feature set corresponding to the current sample image; And each of the first sample voxel points of the current sample image is taken as the corresponding current sample voxel point; and the first and second label voxel points corresponding to the current sample voxel point on the first and second label semantic maps corresponding to the current sample image are recorded as the corresponding first and second matching voxel points; and the first voxel point pixel value of the current sample voxel point is taken as the corresponding second voxel point pixel value; and when the first and second label semantics corresponding to the first and second matching voxel points are both 0, a corresponding second voxel point semantic is set to 0; and the first label semantic corresponding to the first matching voxel point is 1, sets the semantics of a corresponding second voxel point to 1; and sets the semantics of a corresponding second voxel point to 2 when the second label semantics corresponding to the second matching voxel point is 1; and constructs a corresponding first feature voxel point based on the three-dimensional voxel coordinates of the current sample voxel point and records it as the corresponding current feature voxel point; and forms a second voxel point feature vector corresponding to the current feature voxel point by using the second voxel point pixel value and the second voxel point semantics corresponding to the current sample voxel point; and forms a corresponding first training feature map by using all the first feature voxel points corresponding to the current sample image; and constructing the corresponding first and second label type vectors according to the current sample type; when the current sample type is type M0, the corresponding first and second label type probabilities corresponding to type M0 in the corresponding first and second label type vectors are 1, and the first and second label type probabilities corresponding to type M1 are 0; when the current sample type is type M1, the corresponding first and second label type probabilities corresponding to type M1 in the corresponding first and second label type vectors are 0, and the first and second label type probabilities corresponding to type M1 are 1; The first training image and the first label semantic map corresponding to the current original data record form a corresponding first model data record; the second training image and the second label semantic map corresponding to the current original data record form a corresponding second model data record; the first training imaging feature set and the first label type vector corresponding to the current original data record form a corresponding third model data record; and the first training feature map and the second label type vector corresponding to the current original data record form a corresponding fourth model data record; All the first model data records corresponding to the first original data set constitute the corresponding first model data set; all the second model data records corresponding to the first original data set constitute the corresponding second model data set; all the third model data records corresponding to the first original data set constitute the corresponding third model data set; and all the fourth model data records corresponding to the first original data set constitute the corresponding fourth model data set.

5. The method for M-stage classification prediction based on CT images according to claim 3, characterized in that: The training of the first semantic segmentation model based on the first model dataset specifically includes: Step 51: taking the first first model data record of the first model data set as the corresponding current model data record; Step 52: Input the first training image recorded in the current model data into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding first predicted semantic map; Step 53: Input the first predicted semantic graph and the first label semantic graph of the current model data record into a preset first model loss function to calculate and obtain a corresponding first loss value; Wherein, the first model loss function is implemented based on the L1 loss function or the L2 loss function; Step 54: Identify whether the first loss value satisfies a preset first loss value range. If the first loss value satisfies the first loss value range, identify whether the current model data record is the last first model data record of the first model dataset. If so, proceed to step 55. If not, use the next first model data record of the first model dataset as the new current model data record and return to step 52 to continue training. If the first loss value does not satisfy the first loss value range, perform a round of model parameter optimization on the first semantic segmentation model based on a preset first model parameter optimizer in a direction to minimize the first model loss function. After this round of parameter optimization is completed, return to step 52 to continue training. Wherein, the first model parameter optimizer includes at least an SDG optimizer; Step 55: solidify the model parameters of the first semantic segmentation model and confirm that the model training of the first semantic segmentation model is completed.

6. The method for M-stage classification prediction based on CT images according to claim 3, characterized in that: The training of the second semantic segmentation model based on the second model dataset specifically includes: Step 61: taking the first second model data record of the second model data set as the corresponding current model data record; Step 62: Input the second training image recorded in the current model data into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding second predicted semantic map; Step 63: Input the second predicted semantic graph and the second label semantic graph of the current model data record into a preset second model loss function to calculate and obtain a corresponding second loss value; Wherein, the second model loss function is implemented based on the L1 loss function or the L2 loss function; Step 64: Identify whether the second loss value satisfies a preset second loss value range. If the second loss value satisfies the second loss value range, identify whether the current model data record is the last second model data record of the second model dataset. If so, proceed to step 65. If not, use the next second model data record of the second model dataset as the new current model data record and return to step 62 to continue training. If the second loss value does not satisfy the second loss value range, perform a round of model parameter optimization on the second semantic segmentation model based on a preset second model parameter optimizer in a direction to minimize the second model loss function. At the end of this round of parameter optimization, return to step 62 to continue training. Wherein, the second model parameter optimizer includes at least an SDG optimizer; Step 65: solidify the model parameters of the second semantic segmentation model and confirm that the model training of the second semantic segmentation model is completed.

7. The method for M-stage classification prediction based on CT images according to claim 3, characterized in that: The training of the first prediction model based on the third model data set specifically includes: Step 701: performing a full combination of hyperparameters of a first prediction model to obtain a plurality of corresponding first hyperparameter combinations; and performing a hyperparameter setting on the first prediction model based on each of the first hyperparameter combinations to obtain a corresponding first prediction model; Step 702: Set the parameter K to a positive integer greater than or equal to 5; and divide the third model data set into K sub-data sets and record each sub-data set as the corresponding first sub-data set D i ; 1≤index i≤K; Step 703: any of the first sub-datasets D i As a corresponding first validation dataset DB i , and the first sub-dataset D will be i The remaining K-1 first subsets D i Merge into a corresponding first test dataset DA i ; and the first test data sets DA obtained by each i and the corresponding first validation dataset DB i Form a corresponding first test validation set; Step 704: taking the first of the first prediction models as the corresponding current prediction model; Step 705: taking the first of the first test verification sets as the corresponding current test verification set; Step 706: The first test data set DA of the current test validation set is i and the first validation dataset DB i As the corresponding current test dataset and current validation dataset; Step 707: taking the first third model data record of the current test data set as the corresponding current model data record; Step 708: Input the first training radiomics feature set recorded in the current model data into the current prediction model to perform M-stage classification prediction to obtain a corresponding first prediction vector; The first prediction vector includes two first-type prediction probabilities, each of which corresponds to one of the M staging types; Step 709: Substitute the first prediction vector and the first label type vector of the current model data record into a preset third model loss function; and perform a round of model parameter optimization on the model parameters of the current prediction model in a direction in which the third model loss function is minimized based on a preset third model parameter optimizer; and at the end of this round of parameter optimization, identify whether the current model data record is the last third model data record of the current test data set; if so, go to step 710; if not, extract the next third model data record of the current test data set as the new current model data record and return to step 708 for training; Wherein, the third model loss function is implemented based on a binary cross entropy loss function; the third model parameter optimizer includes at least an SDG optimizer; Step 710: Set three counters initially set to 0 as the corresponding TP counter, FP counter, and FN counter; Step 711: traverse all the third model data records in the current verification data set; and during the traversal, use the currently traversed third model data record as the corresponding current verification record; and use the first training radiomics feature set and the first label type vector of the current verification record as the corresponding current radiomics feature set and current label type vector; and use the M stage type corresponding to the first label type probability with a probability value of 1 in the current label type vector as the corresponding current label type; and input the current radiomics feature set into the current prediction model to perform M stage classification prediction to obtain the corresponding second prediction vector; and use the M stage type corresponding to the second type prediction probability with the largest probability in the second prediction vector as the corresponding current prediction type; and compare the current prediction type with the current label type; if the current prediction type and the current label type are both M1 type, add 1 to the TP counter; if the current prediction type is M1 type and the current label type is M0 type, add 1 to the FP counter; if the current prediction type is M0 type and the current label type is M1 type, add 1 to the FN counter; The second prediction vector includes two second-type prediction probabilities, each of which corresponds to one of the M staging types; Step 712: After traversing all the third model data records in the current validation data set, the precision and recall are calculated based on the TP counter, the FP counter, and the FN counter to obtain the corresponding first precision and first recall; and the F1 score is calculated based on the first precision and the first recall to obtain the corresponding first score F1. i ; Step 713: Identify whether the current test validation set is the Kth first test validation set; if so, go to step 714; if not, use the next first test validation set as the new current test validation set and return to step 706; Step 714: Based on the obtained K first scores F1 i Calculate the mean square error to get the corresponding first error E, ; Step 715: Identify whether the current prediction model is the last of the first-first prediction models; if so, go to step 716; if not, use the next first-first prediction model as the new current prediction model and return to step 705; Step 716: taking the full model parameter set of the first prediction model corresponding to the first error E having the smallest error as the corresponding optimal full model parameter set; Step 717: solidify the model parameters of the first prediction model based on the optimal full model parameter set and confirm that the model training of the first prediction model is completed.

8. The method for classifying and predicting M phase based on CT images according to claim 3, characterized in that: The training of the second prediction model based on the fourth model data set specifically includes: Step 81: taking the first fourth model data record of the fourth model data set as the corresponding current model data record; Step 82: Input the first training feature map of the current model data record into the second prediction model to perform M-period classification prediction to obtain a corresponding third prediction vector; The third prediction vector includes two third-type prediction probabilities, each of which corresponds to one of the M staging types; Step 83: Input the third prediction vector and the second label type vector of the current model data record into a preset fourth model loss function to calculate a corresponding third loss value; Wherein, the fourth model loss function is implemented based on the binary cross entropy loss function; Step 84: Identify whether the third loss value satisfies a preset third loss value range; if the third loss value satisfies the third loss value range, identify whether the current model data record is the last fourth model data record of the fourth model data set; if so, go to step 85; if not, use the next fourth model data record of the fourth model data set as the new current model data record and return to step 82 to continue training; if the third loss value does not satisfy the third loss value range, perform a round of model parameter optimization on the second prediction model based on a preset fourth model parameter optimizer in a direction to minimize the fourth model loss function, and return to step 82 to continue training at the end of this round of parameter optimization; Wherein, the fourth model parameter optimizer includes at least an SDG optimizer; Step 85: solidify the model parameters of the second prediction model and confirm that the model training of the second prediction model is completed.

9. The method for M-stage classification prediction based on CT images according to claim 1, characterized in that: If the first prediction mode is the first mode, performing prediction according to the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the first prediction model to obtain a corresponding first prediction type specifically includes: Step 91: Input the first CT image into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding first semantic map; The first CT image includes a plurality of first voxel points; the three-dimensional size of the first semantic map is consistent with the three-dimensional size of the first CT image; the first semantic map includes a plurality of second voxel points, and the first and second voxel points correspond to each other one by one; each second voxel point corresponds to a binary gastric tumor lesion semantic point, with a value of 0 or 1; when the gastric tumor lesion semantic point is 0, it indicates that the corresponding second voxel point is a background voxel point, and when it is 1, it indicates that the corresponding second voxel point is a gastric tumor lesion voxel point; Step 92: Count the number of the second voxel points in the gastric tumor lesion area with a semantic value of 1 on the first semantic map to obtain a corresponding first number; and determine whether the first number exceeds a preset first number threshold. If so, the process proceeds to step 93; if not, the process sets the corresponding first prediction type to type M0 and proceeds to step 99. Step 93: Input the first CT image into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding second semantic map; The three-dimensional size of the second semantic map is consistent with the three-dimensional size of the first CT image; the second semantic map includes a plurality of third voxel points, and the first and third voxel points correspond to each other one by one; each of the third voxel points corresponds to a binary distal key tissue lesion region semantic point, with a value of 0 or 1; when the distal key tissue lesion region semantic point is 0, it indicates that the corresponding third voxel point is a background voxel point, and when it is 1, it indicates that the corresponding third voxel point is a distal key tissue lesion voxel point; Step 94, counting the number of the third voxel points with a semantic value of 1 in the distal key tissue lesion area on the second semantic map to obtain a corresponding second number; and identifying whether the second number exceeds a preset second number threshold. If so, the process proceeds to step 95; if not, the process sets the corresponding first prediction type to type M0 and proceeds to step 99; Step 95: Record the first voxel points on the first CT image corresponding to the second voxel points with a semantic value of 1 in each of the gastric tumor lesion areas as corresponding first lesion points; and record the three-dimensional structure composed of all the first lesion points as the corresponding first structure; and calculate the first-order features, shape features, and texture features of the imaging omics features of the first structure to obtain corresponding first-order feature sets, shape feature sets, and texture feature sets to form a corresponding second gastric tumor feature set; Step 96: Cluster the first voxel points corresponding to each of the distal key tissues on the first CT image based on a preset point group clustering algorithm to obtain a corresponding second point group; and use each of the second point groups as a corresponding current point group, and identify whether the current point group is empty; if the current point group is empty, set a set of all-zero first-order feature sets, shape feature sets, and texture feature sets for the distal key tissue corresponding to the current point group to form a corresponding second distal key tissue feature set; if the current point group is not empty, record the three-dimensional structure composed of the current point group on the first CT image as the corresponding second structure, and calculate the first-order features, shape features, and texture features of the imaging omics features of the second structure to obtain corresponding first-order feature sets, shape feature sets, and texture feature sets to form a corresponding second distal key tissue feature set; and form a corresponding second distal key tissue feature set from all the obtained second distal key tissue feature sets; Step 97: Combining the obtained second gastric tumor feature set and the obtained second distal key tissue feature set into a corresponding first radiomics feature set; and inputting the first radiomics feature set into the first prediction model to perform M-stage classification prediction to obtain a corresponding fourth prediction vector; The fourth prediction vector includes two fourth-type prediction probabilities, each of which corresponds to one of the M staging types; Step 98: taking the M stage type corresponding to the fourth type prediction probability with the largest probability in the fourth prediction vector as the corresponding first prediction type; Step 99: output the obtained first prediction type as the current prediction result.

10. The method for M-stage classification prediction based on CT images according to claim 1, characterized in that: The performing prediction according to the first CT image, the first semantic segmentation model, the second semantic segmentation model, and the second prediction model to obtain the corresponding first prediction type specifically includes: Step 101: Input the first CT image into the first semantic segmentation model to perform semantic point segmentation processing on the gastric tumor lesion area to obtain a corresponding third semantic map; The first CT image includes a plurality of first voxel points; each of the first voxel points corresponds to a first voxel point pixel value; the three-dimensional size of the third semantic map is consistent with the three-dimensional size of the first CT image; the third semantic map includes a plurality of fourth voxel points, and the first and fourth voxel points correspond to each other one by one; each of the fourth voxel points corresponds to a binary gastric tumor lesion semantic point, with a value of 0 or 1; when the gastric tumor lesion semantic point is 0, it indicates that the corresponding fourth voxel point is a background voxel point, and when it is 1, it indicates that the corresponding fourth voxel point is a gastric tumor lesion voxel point; Step 102: Count the number of the fourth voxel points in the gastric tumor lesion area with a semantic value of 1 on the third semantic map to obtain a corresponding third number; and determine whether the third number exceeds a preset first number threshold. If so, proceed to step 103; if not, set the corresponding first prediction type to type M0 and proceed to step 108. Step 103: Input the first CT image into the second semantic segmentation model to perform semantic point segmentation processing on the distal key tissue lesion area to obtain a corresponding fourth semantic map; The three-dimensional size of the fourth semantic map is consistent with the three-dimensional size of the first CT image; the fourth semantic map includes a plurality of fifth voxel points, and the first and fifth voxel points correspond one to one; each of the fifth voxel points corresponds to a binary distal key tissue lesion area semantic, with a value of 0 or 1; when the distal key tissue lesion area semantic is 0, it indicates that the corresponding fifth voxel point is a background voxel point, and when it is 1, it indicates that the corresponding fifth voxel point is a distal key tissue lesion voxel point; Step 104: Count the number of the fifth voxel points with a semantic value of 1 in the distal key tissue lesion area on the second semantic map to obtain a corresponding fourth number; and determine whether the fourth number exceeds a preset second number threshold. If so, proceed to step 105; if not, set the corresponding first prediction type to type M0 and proceed to step 108. Step 105: each of the first voxel points of the first CT image is used as the corresponding current voxel point; and the fourth and fifth voxel points corresponding to the current voxel point on the third and fourth semantic maps are recorded as the corresponding third and fourth matching voxel points; and the first voxel point pixel value of the current voxel point is used as a corresponding feature voxel point pixel value; and when the semantics of the gastric tumor lesion area and the distal key tissue lesion area corresponding to the third and fourth matching voxel points are both 0, a corresponding feature voxel point semantic is set to 0; and when the semantics of the gastric tumor lesion area corresponding to the third matching voxel point is 1, setting the semantics of a corresponding feature voxel point to 1; and setting the semantics of a corresponding feature voxel point to 2 when the semantics of the distal key tissue lesion area corresponding to the fourth matching voxel point is 1; and constructing a corresponding second feature voxel point based on the three-dimensional voxel coordinates of the current voxel point and recording it as the corresponding current feature voxel point; and forming a feature voxel point feature vector corresponding to the current feature voxel point by using the pixel value of the feature voxel point corresponding to the current voxel point and the semantics of the feature voxel point; and forming a corresponding first feature map by using all the second feature voxel points corresponding to the first CT image; Step 106: Input the first feature map into the second prediction model to perform M-period classification prediction to obtain a corresponding fifth prediction vector; The fifth prediction vector includes two fifth-type prediction probabilities, each of which corresponds to one of the M staging types; Step 107: taking the M stage type corresponding to the fifth type prediction probability with the largest probability in the fifth prediction vector as the corresponding first prediction type; Step 108: Output the obtained first prediction type as the current prediction result.

11. A device for executing the processing method for M-phase classification prediction based on CT images according to any one of claims 1 to 10, characterized in that: The device includes: a model construction module, a model training module and a model application module; The model construction module is used to construct a first semantic segmentation model for performing semantic point segmentation of lesion areas of gastric tumors on CT images based on a U-Net model; and to construct a second semantic segmentation model for performing semantic point segmentation of lesion areas of distal key tissues on the CT images based on the U-Net model; and to construct a first prediction model for performing M-stage classification prediction based on imaging omics features based on a machine learning classification model; and to construct a second prediction model for performing M-stage classification prediction based on a three-dimensional CT image with lesion point markings based on a deep learning classification model; wherein the CT image is a three-dimensional abdominal CT image with three-dimensional information of the stomach and the distal key tissues; The model training module is used to train the first semantic segmentation model, the second semantic segmentation model, the first prediction model and the second prediction model respectively; The model application module is used to receive a first CT image and a first prediction mode after the training of the four types of models is completed, and identify the first prediction mode; if the first prediction mode is the first mode, then the corresponding first prediction type is predicted based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the first prediction model; if the first prediction mode is the second mode, then the corresponding first prediction type is predicted based on the first CT image, the first semantic segmentation model, the second semantic segmentation model and the second prediction model, including: inputting the first CT image into the first and second semantic segmentation models for segmentation to obtain third and fourth semantic maps; and taking each first voxel point of the first CT image as the current voxel point, and recording the voxel point corresponding to the current voxel point on the third and fourth semantic maps as the third and fourth matching voxel points, and taking the voxel point pixel value of the current voxel point as the corresponding feature voxel point pixel value, and setting the corresponding feature voxel when the semantics of the gastric tumor lesion area and the semantics of the distal key tissue lesion area corresponding to the third and fourth matching voxel points are both 0. The point semantics is 0, and when the semantics of the gastric tumor lesion area corresponding to the third matching voxel point is 1, the corresponding feature voxel point semantics is set to 1, and when the semantics of the distal key tissue lesion area corresponding to the fourth matching voxel point is 1, the corresponding feature voxel point semantics is set to 2, and based on the three-dimensional voxel coordinates of the current voxel point, a corresponding second feature voxel point is constructed and recorded as the current feature voxel point, and the feature voxel point pixel value corresponding to the current voxel point and the feature voxel point semantics are used to form a feature voxel point feature vector corresponding to the current feature voxel point, and all the second feature voxel points corresponding to the first CT image are used to form a first feature map; and the first feature map is input into the second prediction model for M stage classification prediction to obtain a corresponding fifth prediction vector, the fifth prediction vector including two fifth type prediction probabilities; the M stage type corresponding to the fifth type prediction probability with the largest probability in the fifth prediction vector is used as the corresponding first prediction type; and the obtained first prediction type is output as the prediction result corresponding to the first CT image; the first prediction type includes M0 type and M1 type.

12. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is configured to be coupled to the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 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, and when the computer instructions are executed by a computer, the computer is caused to execute the method according to any one of claims 1 to 10.

Images