Evaluation method for gastric cancer peritoneal metastasis state recognition and PCI score estimation

Abstract

The invention provides an evaluation method for gastric cancer peritoneal metastasis state recognition and PCI score estimation, and the method comprises the steps: receiving a preoperative abdominal enhancement CT image, and carrying out the automatic positioning preprocessing of the peritoneum through resampling, normalization, image enhancement and a region attention mechanism; image omics features are extracted from the region of interest and fused with the depth features of the multi-scale convolutional neural network, and a binary classification model based on a residual network / Transform is constructed to output transition state probability and confidence; pCI scores of all the areas are quantitatively evaluated synchronously through the peritoneal thirteen subareas, and finally a transfer prediction result, a PCI spatial distribution map, a visual heat map and a model interpretation report are integrated to form structured diagnosis output. According to the method, transition state identification and PCI score quantification can be realized, and the problems of single function, weak generalization, low interpretability and the like of a traditional model are solved.

Description

Technical Field

[0001] The present application relates to the technical field of medical image analysis, and in particular to an evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating PCI scores. Background Art

[0002] Gastric cancer is a digestive system malignancy with high morbidity and mortality worldwide. Peritoneal metastasis (PM) is one of the most common and devastating forms of distant metastasis, accounting for approximately 35% to 50% of metastatic types in patients with advanced gastric cancer. Peritoneal metastasis often progresses insidiously, and traditional imaging modalities (such as CT and MRI) lack sensitivity in identifying early lesions, resulting in a high rate of preoperative misdiagnosis and severely impacting the accuracy of clinical staging and treatment decisions.

[0003] Currently, the final diagnosis of peritoneal metastasis in clinical practice relies primarily on intraoperative exploration and peritoneal lavage fluid cytology. The Peritoneal Cancer Index (PCI), widely used to assess peritoneal metastasis distribution and tumor burden, requires intraoperative scoring based on the distribution and size of lesions within 13 peritoneal regions. However, the PCI score relies heavily on surgeon experience, is difficult to standardize, and cannot be quantitatively predicted preoperatively, limiting its value as a prospective guide in treatment planning. Summary of the Invention

[0004] In view of this, the embodiment of the present application provides an evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score, which can realize metastasis status identification and PCI score quantification, and solve the problems of traditional models with single function, weak generalization and low interpretability.

[0005] The technical solution of the embodiment of the present application is implemented as follows: In a first aspect, embodiments of the present application provide an evaluation method for identifying gastric cancer peritoneal metastasis status and estimating PCI scores, the method comprising: Receive preoperative abdominal enhanced CT imaging data of gastric cancer patients; Preprocessing the preoperative abdominal enhanced CT image data, wherein the preprocessing includes resampling, normalization, image enhancement, and automatic positioning of the peritoneal region; wherein the automatic positioning is based on a regional attention mechanism; Extracting radiomics features and deep learning features from the region of interest (ROI) of the peritoneal region, and fusing the radiomics features and the deep learning features to obtain fused features; wherein the radiomics features include shape, texture, grayscale distribution, and edge structure features, and the deep learning features include image semantic features extracted by a multi-scale convolutional neural network; Constructing a peritoneal metastasis status classification model based on the fusion features; wherein the peritoneal metastasis status classification model is constructed based on a binary classification neural network with a residual network or a Transformer architecture, and the peritoneal metastasis status model outputs a predicted probability and a confidence score of whether peritoneal metastasis exists; The tumor burden in the peritoneal compartment was quantitatively assessed to obtain the PCI score for each peritoneal compartment; The final evaluation result is displayed based on the output result of the peritoneal metastasis status model and the evaluation result of the quantitative evaluation, wherein the final evaluation result includes at least one of the peritoneal metastasis status prediction result, PCI regional scores and spatial distribution maps, visual heat maps, model explanation outputs, and doctor-patient interaction records.

[0006] In a second aspect, an embodiment of the present application further provides an evaluation system for identifying gastric cancer peritoneal metastasis status and estimating PCI scores, the system comprising: An image data input module is used to receive preoperative abdominal enhanced CT image data of gastric cancer patients; An image preprocessing module, configured to preprocess the preoperative abdominal enhanced CT image data, wherein the preprocessing includes resampling, normalization, image enhancement, and automatic positioning of the peritoneal region; wherein the automatic positioning is based on a regional attention mechanism; a feature extraction module, configured to extract radiomics features and deep learning features from the region of interest (ROI) of the peritoneal region, and fuse the radiomics features and the deep learning features to obtain fused features; wherein the radiomics features include shape, texture, grayscale distribution, and edge structure features, and the deep learning features include image semantic features extracted by a multi-scale convolutional neural network; a peritoneal metastasis status recognition module, configured to construct a peritoneal metastasis status classification model based on the fusion features; wherein the peritoneal metastasis status classification model is constructed based on a binary classification neural network with a residual network or a Transformer architecture, and the peritoneal metastasis status model outputs a predicted probability and a confidence score for the presence of peritoneal metastasis; The PCI score estimation module is used to quantitatively evaluate the tumor burden in the peritoneal partition and obtain the PCI score of each peritoneal partition; An auxiliary diagnosis interaction module is used to display the final evaluation results based on the output results of the peritoneal metastasis status model and the evaluation results of the quantitative evaluation, wherein the final evaluation results include at least one of the peritoneal metastasis status prediction results, PCI regional scores and spatial distribution maps, visual heat maps, model explanation outputs, and doctor-patient interaction records.

[0007] In a third aspect, an embodiment of the present application further provides an electronic device comprising: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to execute the evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score as described in any one of the first aspects.

[0008] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score as described in any one of the first aspects is executed.

[0009] The embodiments of the present application have the following beneficial effects: The embodiment of the present application realizes the simultaneous output of the dual tasks of "metastasis status recognition + tumor burden quantification (PCI score)", automatically recognizes the thirteen peritoneal zones driven by anatomical priors, enhances the perception of small lesions through the fusion of multimodal features (radiology / deep features / clinical indicators), optimizes feature integration through the SE attention mechanism, and combines model interpretability visualization (heat map / structured report) and lightweight containerized deployment architecture to solve the problems of traditional AI models with single functions, manual reliance on partitioning, weak generalization ability and low clinical trust, providing a high-precision, highly adaptable and feasible technical platform for precise diagnosis and treatment of tumors, with significant originality and clinical translation value. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 10 is a flow chart of steps S101-S106 provided in an embodiment of the present application; Figure 2 Schematic diagram of the structure of an evaluation system for gastric cancer peritoneal metastasis status identification and PCI score estimation provided in an embodiment of the present application; Figure 3 It is a schematic diagram of the composition structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of illustration and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps without logical context can be reversed or implemented simultaneously. In addition, those skilled in the art, under the guidance of the contents of this application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0013] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0014] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0015] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0016] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0018] Before implementing the embodiments of the present application, the applicant discovered that the prior art had the following problems: Currently, clinical practice generally relies on CT or MRI images, which are manually identified by imaging departments or surgeons based on typical manifestations such as peritoneal thickening, nodules, ascites, and blurred intestinal spaces. This method has the following drawbacks: Highly subjective: Different doctors have different judgment criteria, and there is a certain threshold of experience in reading films; Occult peritoneal metastasis is easily missed: low sensitivity for early metastases without obvious morphological changes; Unable to quantify peritoneal burden: only the presence or absence of the disease is determined, and no PCI quantitative score can be performed, which makes it difficult to reflect the extent of the disease. Poor repeatability: Manual evaluation relies on the doctor's experience and lacks objectivity and standardization.

[0019] Therefore, the clinical value of predicting peritoneal metastasis based solely on conventional imaging findings is severely limited.

[0020] Laparoscopic exploration combined with peritoneal lavage cytology is currently the gold standard for the diagnosis of peritoneal metastasis, and PCI (Peritoneal Cancer Index) assessment is performed manually during surgery. However, this method has the following significant limitations: Traumatic: Intraoperative evaluation cannot provide preoperative information, which is not conducive to advance decision-making and individualized treatment path planning; Large differences in standardization: PCI scoring relies on the surgeon's experience, especially in the identification of small nodules, which is highly subjective and leads to scoring differences; Irreproducibility: The accuracy of the scoring cannot be verified retrospectively after surgery, which affects the stability of scientific research data and the quality of clinical research; Low evaluation efficiency: The evaluation process requires observation and scoring of each area and nodule, which is not suitable for parallel processing of high-intensity or multi-center cases.

[0021] Therefore, there is an urgent need for non-invasive, preoperative, objective and standardized alternative PCI scoring methods.

[0022] In recent years, some research teams have begun to try to use artificial intelligence and imaging omics to predict peritoneal metastasis before surgery. For example: Extracting texture and density features from CT images based on manually annotated ROIs and combining them with traditional machine learning classifiers (such as SVM and Random Forest) to determine the risk of peritoneal metastasis. Alternatively, a shallow convolutional neural network can be used to identify lesions in the peritoneal area.

[0023] However, the above attempts still have the following technical defects: Incomplete methods: Most methods can only classify the status of peritoneal metastasis and fail to address the important clinical quantitative task of PCI scoring; High degree of manual involvement: Many methods still rely on manual delineation of the peritoneal area, making it difficult to automatically generalize and subject to significant inter-operator variability; Imprecise lesion localization: Ignoring the 13 peritoneal compartmental structures and lacking detailed tumor region modeling and scoring mechanisms; Lack of explainability: Unable to provide doctors with visual positioning and model reasoning process, affecting clinical trust; System fragmentation: Each study tends to focus on a single problem and lacks an integrated diagnostic system from image input, structural analysis to multi-task output.

[0024] In response to the limitations of the above-mentioned background technologies, the present embodiment system proposes a new intelligent auxiliary diagnosis and evaluation framework with "dual-task, multi-modal fusion, and fully automatic closed-loop", which has the following unique technical advantages: Integrating the dual functions of "state recognition + partition scoring" into an integrated system: breaking through the application bottleneck of existing AI models that "can only recognize but not quantify"; Automatic region recognition solution integrating anatomical priors and deep learning: Automatic segmentation of peritoneal structures improves spatial accuracy and model stability; Neural network-based partition load modeling and regression prediction: 13 peritoneal partitions are encoded as structural information, and structured PCI scores and heat map prompts are output to facilitate preoperative judgment; Equipped with model interpretation capabilities and graphical visualization interfaces: Improves physician acceptance and enhances the collaborative decision-making capabilities of doctors and engineers; Support system deployment and continuous training and update mechanism: applicable to different hospital data standards and image acquisition protocols, enhancing promotion adaptability.

[0025] Therefore, the embodiments of the present application not only fill the gaps in the existing technology in preoperative identification of peritoneal metastasis and prediction of PCI scores, but also establish a set of auxiliary diagnosis systems with automated, standardized and intelligent features, which have significant technological progress and practical clinical value in the application of AI in gastric cancer imaging.

[0026] See also Figure 1 , Figure 1 This is a flow chart of steps S101-S106 of the evaluation method for gastric cancer peritoneal metastasis status identification and PCI score estimation provided in the embodiment of the present application, which will be combined with Figure 1 Steps S101-S106 are shown for explanation.

[0027] In step S101 , preoperative abdominal enhanced CT image data of a gastric cancer patient is received.

[0028] Here, preoperative abdominal enhanced CT images of gastric cancer patients can be imported into the system in DICOM format, and image structure standardization, sequence verification, and patient identity information desensitization are completed. The system automatically generates a unique case identification code to support model training data management.

[0029] In step S102, the preoperative abdominal enhanced CT image data is preprocessed, and the preprocessing includes resampling, normalization, image enhancement, and automatic positioning of the peritoneal region; wherein the automatic positioning is based on a regional attention mechanism.

[0030] Here, the original image is resampled (unified to 1.0 mm³ resolution), normalized (based on HU value truncation and z-score standardization), and artifacts are removed (based on metal artifact suppression algorithm). Then, a peritoneal ROI extraction network built based on a deep segmentation network (such as U-Net) is called.

[0031] In step S103, radiomics features and deep learning features are extracted from the region of interest (ROI) of the peritoneal area, and the radiomics features and the deep learning features are fused to obtain fused features; wherein the radiomics features include shape, texture, grayscale distribution and edge structure features, and the deep learning features include image semantic features extracted by a multi-scale convolutional neural network.

[0032] In some embodiments, when extracting radiomics features and deep learning features from the region of interest (ROI) of the peritoneal region, the method further comprises: Modeling the spatial topological structure information within the peritoneal area based on the graph neural network structure; Enhance the ability to express the relationship between lesions based on modeling results.

[0033] Here, the preprocessed ROI image is subjected to feature extraction, which consists of two paths: Imageomics path: Extract traditional radiomics features (including grayscale histogram, GLCM texture, LoG edge, etc.), with a total feature dimension of no less than 1000; Deep learning path: Use multi-scale convolutional neural networks (such as ResNet-34 or EfficientNet-B0) to extract high-level semantic features, and reduce the output dimension through fully connected layers and concatenate them; Finally, the feature vectors of the two paths are weighted and fused through the attention mechanism to form a unified representation for use by subsequent models.

[0034] In step S104, a peritoneal metastasis status classification model is constructed based on the fusion features; wherein the peritoneal metastasis status classification model is constructed based on a binary classification neural network of a residual network or a Transformer architecture, and the peritoneal metastasis status model outputs a predicted probability and a confidence score of whether peritoneal metastasis exists.

[0035] In some embodiments, the peritoneal metastasis status classification model performs multimodal fusion prediction by simultaneously inputting imaging features and clinical features; wherein the clinical features include CEA, CA125, and tumor differentiation degree.

[0036] Here, the fused features are used as input, and a binary classification neural network model outputs the peritoneal metastasis status (metastasis / non-metastasis) and its probability. This model supports multimodal input (imaging and clinical features), where clinical variables include tumor type, differentiation level, serum CA125, CEA, and N stage. The network architecture uses a dual-channel shared weight fusion structure, with the fusion layer consisting of a multilayer perceptron (MLP) and a sigmoid output.

[0037] In step S105 , the tumor burden in the peritoneal partition is quantitatively evaluated to obtain the PCI score of each peritoneal partition.

[0038] In some embodiments, the quantitative assessment of tumor burden in the peritoneal partition to obtain the PCI score of each peritoneal partition comprises: The 13 peritoneal regions are segmented using the nnU-Net multi-label semantic segmentation network. Detect the number of lesions and the maximum nodule diameter in each partition; The PCI score of each peritoneal region is output based on the regression network and automatically accumulated to obtain the total PCI score; the PCI score of each peritoneal region has a score interval of [0, 3].

[0039] Here, we first use a 13-partition semantic segmentation network based on the nnU-Net architecture; After partitioning, the lesion detection model (YOLOv5 or ResUNet) is called to automatically identify the maximum diameter and number of nodules in the area; Scores are automatically assigned to each region based on the judgment rules (0: no lesion, 1: maximum diameter <0.5 cm, 2: 0.5–5 cm, 3: >5 cm or fused tumor), and the total PCI score is automatically accumulated. Finally, the prediction output synchronously displays the scores of each partition, a visualization heat map of suspected lesions, and a confidence interval.

[0040] In step S106, the final evaluation result is displayed based on the output result of the peritoneal metastasis status model and the evaluation result of the quantitative evaluation, wherein the final evaluation result includes at least one of the peritoneal metastasis status prediction result, PCI regional scores and spatial distribution maps, visualization heat maps, model explanation output, and doctor-patient interaction records.

[0041] In some embodiments, the PCI scoring results for each peritoneal region are accompanied by a model confidence interval or uncertainty index, and provide information explaining the source of error.

[0042] In some embodiments, the final assessment result displays the predicted results and risk levels of each partition in the form of a 3D abdominal cavity reconstruction view to assist in preoperative lesion localization.

[0043] In some embodiments, the model interpretation output is implemented based on Grad-CAM, Layer-wise RelevancePropagation, or SHAP methods, and provides local image evidence for peritoneal metastasis prediction.

[0044] Here, the embodiment of the present application provides a system front-end interface, which displays the model output content through an interactive Web GUI, including: Peritoneal metastasis prediction results (positive / negative + probability value); PCI score radar chart and regional score list; Peritoneal 3D regional structure map and nodule positioning heat map; Doctor revision interface, case review interface and structured report export interface.

[0045] The training process of the embodiment of the present application is as follows: Data source: Abdominal enhanced CT images and intraoperative exploration records (including PCI scores) of ≥500 gastric cancer patients were collected; Annotation method: Peritoneal metastasis status and 13-zone PCI score were annotated by ≥2 attending abdominal surgeons; Data processing: Use a 3:1:1 method to construct training, validation, and test sets; Network training: Use the Adam optimizer with an initial learning rate of 1e-4 and an early stopping strategy to prevent overfitting. Performance evaluation: ROC-AUC, F1-score, MAE, and Pearson correlation coefficient were used to evaluate the classification and regression models respectively.

[0046] The above embodiment can be applied in the actual deployment process of the hospital. Doctors can load the patient's preoperative CT image through the system. The system will automatically complete the peritoneal area positioning, peritoneal metastasis risk assessment and PCI score estimation within 5 minutes, and output a graphic and text structured auxiliary report for preoperative MDT discussion, surgical plan formulation and patient prognosis communication, significantly improving the efficiency of peritoneal lesion identification and prediction consistency.

[0047] In order to verify the effectiveness and adaptability of the system of the present invention, the following provides a specific example to illustrate the system in combination with actual clinical data and the system implementation process: Example 1: Identification and auxiliary judgment of peritoneal metastasis status Patient ID: Case_GC_0231 Gender: Male; Age: 64 Diagnosis: Gastric antral adenocarcinoma, clinical T4aN2M0 Preoperative examination: abdominal enhanced CT scan, 5mm slice thickness, exported to the system via PACS Processing flow: 1. The system automatically receives the patient's CT images and completes image preprocessing, including resampling to 1.0mm isovoxels, metal artifact suppression, and automatic segmentation of the peritoneal region.

[0048] 2. The feature extraction module extracts a total of 1286-dimensional traditional radiomics features and 512-dimensional deep image features extracted by ResNet-34; at the same time, it loads the patient's CEA and CA125 clinical indicators.

[0049] 3. The peritoneal metastasis status recognition module performs inference based on the fusion features and predicts a positive peritoneal metastasis value with an output probability of 89.7% and a confidence interval of [83.2%, 93.5%].

[0050] Intraoperative verification results: Abdominal exploration revealed scattered tumor nodules in the anterior peritoneum and the root of the mesentery, with positive lavage cytology, and the metastatic status was consistent with the model prediction.

[0051] Example 2: PCI score prediction and regional score output Patient ID: Case_GC_0452 Gender: Female; Age: 58 Diagnosis: Poorly differentiated adenocarcinoma of the gastric body, clinical T4bN3M0, planned preoperative evaluation for laparoscopic exploration Processing flow: 1. After loading the CT image, the system automatically segmented 13 peritoneal regions using a multi-label network based on nnU-Net, with an average Dice coefficient of 0.85.

[0052] 2. Within each region, the model detects the maximum nodule diameter and number and automatically assigns a value based on the PCI scoring rules: Area 2 (right upper quadrant): 2 points, maximum diameter 1.2cm; Area 5 (left mid-abdomen): 3 points, confluent tumor mass > 5 cm; The remaining areas: ranging from 0 to 1 points, with lesions in a total of 10 areas.

[0053] System output: The total PCI prediction score is 17 points, and a radar chart, a zoning color score chart, and a hot zone positioning visualization are generated. It is recommended that areas 2, 5, and 9 be focused on during surgery.

[0054] Actual measured value during the operation: PCI score was 18 points, which deviated from the system prediction by only 1 point, with an absolute error rate of 5.5%.

[0055] Example 3: System deployment and user feedback Deployed hospital: Department of Gastrointestinal Surgery, The Fourth Hospital of Hebei Medical University Deployment mode: LAN PACS linkage deployment + local Dockerized backend server Model running time: Single case processing time is about 4.2 minutes (including uploading, prediction, and display) User feedback: Ten abdominal surgeons participated in the usability evaluation. 80% of them said the system was "significantly helpful" for preoperative path planning and anticipated PCI assessment, and 90% were satisfied with the interface and report presentation.

[0056] In summary, the embodiments of the present application have the following beneficial effects: (1) Joint dual-task prediction architecture: For the first time, the simultaneous output of peritoneal metastasis identification and PCI scoring is achieved.

[0057] Existing research has mostly focused on the classification of peritoneal metastasis status, lacking a quantitative expression of tumor burden. This system, for the first time, introduces a joint task framework, building a peritoneal metastasis classification model and a PCI score regression model in parallel within the same model. By sharing an image feature encoder and branching the decoding output, it achieves dual determination of "metastasis" and "metastasis degree" within a single process: Overcome the problems of single function and insufficient auxiliary dimension of existing forecasting systems; Meet the synchronous information needs of "status + classification" for preoperative auxiliary diagnosis and treatment; Improve model parameter sharing and learning efficiency.

[0058] This strategy can be extended to metastasis identification and load assessment tasks of other tumors and has universal expansion value.

[0059] (2) An anatomically driven automatic identification mechanism for peritoneal zonation to build a structured scoring space foundation.

[0060] This application example presents a groundbreaking method for automatic image region segmentation based on the peritoneal thirteen-partition rule and anatomical priors. The system, combined with a deep semantic segmentation network (such as nnU-Net), performs fine-grained peritoneal segmentation, ensuring regional correspondence for subsequent PCI scoring. Define the spatial boundaries of 13 peritoneal regions and establish a spatial mapping structure for prediction output; Automatically assign scores (0-3 points) to each lesion zone and output the total PCI value; Avoid manual zoning errors and achieve standardized and reusable regional scoring.

[0061] This partitioning mechanism overcomes the problem that current AI models are unable to identify the spatial distribution structure of the peritoneum, providing support for precise disease management.

[0062] (3) Multimodal fusion and enhanced attention mechanism significantly improve the ability to perceive small lesions.

[0063] This system integrates traditional radiomics features (shape, grayscale, texture, etc.) + deep features (CNN / Transformer extraction) + clinical variables (such as CA125, CEA, etc.) to construct a three-channel multimodal input structure. It also introduces the SE attention mechanism and multi-scale residual network to adaptively adjust the importance of different modal features, enhancing the model's ability to integrate heterogeneous lesion characteristics: Effectively improve the ability to identify tiny, hidden peritoneal lesions; Improve the adaptability and robustness of the model under different hospital image acquisition conditions; Realize the coordinated expression of three layers of information: "imaging-clinical-structural".

[0064] This fusion strategy breaks through the limitations of existing models, such as strong dependence on specific modalities and poor generalization ability, and adapts to complex clinical reality scenarios.

[0065] (4) Enhanced model interpretability and structured visual output improve clinical usability and trust.

[0066] Different from traditional black-box AI prediction, the system of this invention integrates multiple model interpretation algorithms (such as Grad-CAM, SHAP, LRP) with a structured visualization interface, providing the following functions: Display significant area heatmaps of peritoneal metastasis classification output results; The PCI scores and predictions for each zone are visualized in the peritoneal zone structure diagram; Provide structured risk warning statements and auxiliary diagnosis recommendation reports.

[0067] This mechanism significantly enhances doctors' comprehensibility, traceability and acceptability of AI results, and is a key supporting technology for the transition from algorithms to real clinical applications.

[0068] (5) Support system deployment and online migration capabilities to adapt to multi-center imaging heterogeneity and data update requirements.

[0069] This system adopts the Docker container deployment architecture, featuring lightweight deployment, standard interface integration (such as DICOM, FHIR, and HL7), and a continuous model training mechanism. It has the following technical advantages: It can be deployed on different hospital servers, connected to PACS and EMR systems, and data can flow in automatically; Supports online incremental training mechanisms (such as EWC or fine-tuning) to perform micro-model updates based on new cases; Provides controllable version management and model update logs to ensure algorithm security and compliance.

[0070] This strategy overcomes the performance degradation problem caused by the heterogeneity of data sources in the actual implementation of AI models, and significantly enhances the system's promotion capabilities.

[0071] In summary: The embodiments of this application achieve technological breakthroughs in many aspects, including system architecture, algorithm modeling, partition standardization, information fusion, visual interaction, and deployment strategies. An evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating PCI scores is proposed. The method has high precision, strong adaptability, interpretability, and engineering feasibility, providing a key technical platform for intelligent diagnosis and treatment of tumors, and possesses significant originality, advancement, and clinical translation value.

[0072] Based on the same inventive concept, the embodiments of the present application also provide an evaluation device for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score, which corresponds to the evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score in the first embodiment. Since the principle of solving the problem by the device in the embodiments of the present application is similar to the above-mentioned evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0073] like Figure 2 As shown, Figure 2 2 is a schematic diagram of the structure of an evaluation system 200 for gastric cancer peritoneal metastasis status identification and PCI score estimation provided in an embodiment of the present application. The evaluation system 200 for gastric cancer peritoneal metastasis status identification and PCI score estimation includes: Image data input module 201, for receiving preoperative abdominal enhanced CT image data of a gastric cancer patient; An image preprocessing module 202 is configured to preprocess the preoperative abdominal enhanced CT image data, wherein the preprocessing includes resampling, normalization, image enhancement, and automatic positioning of the peritoneal region; wherein the automatic positioning is based on a regional attention mechanism; a feature extraction module 203 for extracting radiomics features and deep learning features from the region of interest (ROI) of the peritoneal region, and fusing the radiomics features and the deep learning features to obtain fused features; wherein the radiomics features include shape, texture, grayscale distribution, and edge structure features, and the deep learning features include image semantic features extracted by a multi-scale convolutional neural network; a peritoneal metastasis status recognition module 204 for constructing a peritoneal metastasis status classification model based on the fused features; wherein the peritoneal metastasis status classification model is constructed based on a binary classification neural network with a residual network or a Transformer architecture, and the peritoneal metastasis status model outputs a predicted probability and a confidence score of whether peritoneal metastasis exists; The PCI score estimation module 205 is used to quantitatively evaluate the tumor burden in the peritoneal partitions and obtain the PCI score of each peritoneal partition; The auxiliary diagnosis interaction module 206 is used to display the final evaluation results based on the output results of the peritoneal metastasis status model and the evaluation results of the quantitative evaluation, wherein the final evaluation results include at least one of the peritoneal metastasis status prediction results, PCI regional scores and spatial distribution maps, visual heat maps, model interpretation outputs, and doctor-patient interaction records.

[0074] Those skilled in the art should understand that Figure 2 The functions implemented by each unit in the evaluation system 200 for identifying gastric cancer peritoneal metastasis status and estimating PCI score can be understood by referring to the description of the evaluation method for identifying gastric cancer peritoneal metastasis status and estimating PCI score. Figure 2 The functions of the various units in the evaluation system 200 for identifying gastric cancer peritoneal metastasis status and estimating PCI scores can be implemented by a program running on a processor or by a specific logic circuit.

[0075] In a possible embodiment, when the feature extraction module 203 extracts radiomics features and deep learning features from the region of interest (ROI) of the peritoneal area, it further includes: Modeling the spatial topological structure information within the peritoneal area based on the graph neural network structure; Enhance the ability to express the relationship between lesions based on modeling results.

[0076] In one possible embodiment, the peritoneal metastasis status classification model performs multimodal fusion prediction by simultaneously inputting imaging features and clinical features; wherein the clinical features include CEA, CA125, and tumor differentiation degree.

[0077] In one possible embodiment, the quantitative assessment of the tumor burden in the peritoneal partition to obtain the PCI score of each peritoneal partition includes: The 13 peritoneal regions are segmented using the nnU-Net multi-label semantic segmentation network. Detect the number of lesions and the maximum nodule diameter in each partition; The PCI score of each peritoneal region is output based on the regression network and automatically accumulated to obtain the total PCI score; the PCI score of each peritoneal region has a score interval of [0, 3].

[0078] In one possible embodiment, the PCI scoring results for each peritoneal region are accompanied by a model confidence interval or uncertainty index, and error source explanation information is provided.

[0079] In a possible embodiment, the final evaluation result displays the prediction result and risk level of each partition in the form of a 3D abdominal cavity reconstruction view to assist in preoperative lesion localization.

[0080] In one possible embodiment, the model interpretation output is implemented based on Grad-CAM, Layer-wise Relevance Propagation or SHAP method, and provides local image evidence for peritoneal metastasis prediction.

[0081] The aforementioned assessment system for gastric cancer peritoneal metastasis status identification and PCI score estimation primarily targets peritoneal metastasis and PCI score prediction in gastric cancer. Its core technical architecture boasts excellent cross-disease migration and model adaptability, making it particularly suitable for solid tumors with a propensity for peritoneal implantation and requiring peritoneal regional burden assessment. It possesses the following three levels of universal foundation and adaptation logic: 1. Universal design of model structure.

[0082] 1. The anatomical region division mechanism is universal: The spatial modeling and regional mask generation logic for the system's 13 peritoneal regions are based on the human peritoneal anatomy and are independent of specific tumor types. Therefore, the regional modeling framework and scoring logic can be directly applied to other diseases that also use PCI as a metric for lesion burden assessment, such as ovarian cancer and colorectal cancer peritoneal metastasis.

[0083] 2. Modularization of model input structure: The system input supports enhanced CT images in DICOM format. The image preprocessing module can adjust parameters such as window width and organ enhancement timing to adapt to scanning schemes for different diseases. At the same time, the model backbone structure adopts an encoder-decoder decoupling design to facilitate retaining common parts and replacing specific branches.

[0084] 3. Task separation and branch customization: This system adopts a "shared trunk + dual-task branch" design, which can achieve different task configurations by replacing the output branches, such as: Alternative classification branches are used to identify peritoneal dissemination status in ovarian cancer; The replacement regression branch is used to estimate the extent of cancerous ascites effusion, etc.

[0085] 2. Analysis of typical suitable diseases.

[0086] 1. Ovarian Cancer.

[0087] Peritoneal metastasis is the earliest form of external spread of ovarian cancer. Preoperative diagnosis and PCI value are key to determining the timing of surgery. This system can be directly transferred and applied. Image preprocessing is adapted to the pelvic enhancement phase. The training data needs to include the annotation of pelvic peritoneal lesions. Clinical partners such as gynecologic oncology centers can use the system to perform non-invasive zonal burden assessments.

[0088] 2. Colorectal cancer peritoneal metastasis (CRPM).

[0089] CRPM also requires preoperative assessment of the extent of peritoneal implantation to determine whether to perform CRS+HIPEC; The peritoneal region detection logic and feature extraction backbone network of this system can be reused, and only targeted fine-tuning of the training data is required; The classification label definition is completely consistent with the PCI scoring logic.

[0090] 3. Gastrointestinal stromal tumor (GIST) and retroperitoneal soft tissue sarcoma.

[0091] Although peritoneal metastasis is rare, once it occurs, the implantation lesions are usually heavy and can be automatically identified by the system; If there are dense areas of peritoneal nodules, this system can be used to locate the target area and quantify the nodule size distribution.

[0092] 3. Adaptation method and path recommendations.

[0093] 1. Fine-tuning For diseases with large differences in lesion morphology (such as serous ovarian cancer vs. adenocarcinoma-type gastric cancer), the output layer parameters can be adjusted through small sample retraining, keeping the backbone structure unchanged, and achieving rapid migration.

[0094] 2. Label adjustment and task replacement: For diseases that do not use PCI scoring but also involve peritoneal area assessment, model branches can be used to output quantitative indicators such as tumor area and maximum diameter to replace standard PCI tasks and achieve functional flexibility.

[0095] 3. Anatomy prior template replacement: If the target disease lesions are concentrated in a specific area (e.g., ovarian cancer is mainly distributed in the pelvis), the partition weight can be locally enhanced, or replaced with a pelvic and abdominal partition template (e.g., regions 1\~6 are mapped to different paraovarian cavities) to improve regional recognition capabilities.

[0096] In summary, the assessment system for gastric cancer peritoneal metastasis status identification and PCI score estimation provided in this embodiment has good disease transferability and adaptability, providing a unified technical platform and assessment framework for peritoneal metastasis risk identification and burden stratification tasks for multiple solid tumors. In the future, by expanding the training sample library and optimizing the branch structure, it can be further developed into a "multi-disease integrated intelligent assisted assessment system" for peritoneal metastasis-related diseases.

[0097] like Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes: A processor 301, a storage medium 302 and a bus 303, wherein the storage medium 302 stores machine-readable instructions executable by the processor 301. When the electronic device 300 is running, the processor 301 communicates with the storage medium 302 via the bus 303, and the processor 301 executes the machine-readable instructions to perform the steps of the evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score as described in the embodiment of the present application.

[0098] In actual application, the various components in the electronic device 300 are coupled together via the bus 303. It is understood that the bus 303 is used to realize the connection and communication between these components. In addition to the data bus, the bus 303 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 3 Various buses are labeled as buses 303.

[0099] The electronic device has the following beneficial effects: (1) Joint dual-task prediction architecture: For the first time, the simultaneous output of peritoneal metastasis identification and PCI scoring is achieved.

[0100] Existing research has mostly focused on the classification of peritoneal metastasis status, lacking a quantitative expression of tumor burden. This system, for the first time, introduces a joint task framework, building a peritoneal metastasis classification model and a PCI score regression model in parallel within the same model. By sharing an image feature encoder and branching the decoding output, it achieves dual determination of "metastasis" and "metastasis degree" within a single process: Overcome the problems of single function and insufficient auxiliary dimension of existing forecasting systems; Meet the synchronous information needs of "status + classification" for preoperative auxiliary diagnosis and treatment; Improve model parameter sharing and learning efficiency.

[0101] This strategy can be extended to metastasis identification and load assessment tasks of other tumors and has universal expansion value.

[0102] (2) An anatomically driven automatic identification mechanism for peritoneal zonation to build a structured scoring space foundation.

[0103] This application example presents a groundbreaking method for automatic image region segmentation based on the thirteen-partition rule for the peritoneum and anatomical priors. The system incorporates a deep semantic segmentation network (such as nnU-Net) to perform fine-grained peritoneal segmentation, ensuring regional correspondence for subsequent PCI scoring. Define the spatial boundaries of 13 peritoneal regions and establish a spatial mapping structure for prediction output; Automatically assign scores (0-3 points) to each lesion zone and output the total PCI value; Avoid manual zoning errors and achieve standardized and reusable regional scoring.

[0104] This partitioning mechanism overcomes the problem that current AI models are unable to identify the spatial distribution structure of the peritoneum, providing support for precise disease management.

[0105] (3) Multimodal fusion and enhanced attention mechanism significantly improve the ability to perceive small lesions.

[0106] This system integrates traditional radiomics features (shape, grayscale, texture, etc.) + deep features (CNN / Transformer extraction) + clinical variables (such as CA125, CEA, etc.) to construct a three-channel multimodal input structure. It also introduces the SE attention mechanism and multi-scale residual network to adaptively adjust the importance of different modal features, enhancing the model's ability to integrate heterogeneous lesion characteristics: Effectively improve the ability to identify tiny, hidden peritoneal lesions; Improve the adaptability and robustness of the model under different hospital image acquisition conditions; Realize the coordinated expression of three layers of information: “imaging-clinical-structural”.

[0107] This fusion strategy breaks through the limitations of existing models, such as strong dependence on specific modalities and poor generalization ability, and adapts to complex clinical reality scenarios.

[0108] (4) Enhanced model interpretability and structured visual output improve clinical usability and trust.

[0109] Different from traditional black-box AI prediction, the system of this invention integrates multiple model interpretation algorithms (such as Grad-CAM, SHAP, LRP) with a structured visualization interface, providing the following functions: Display significant area heatmaps of peritoneal metastasis classification output results; The PCI scores and predictions for each zone are visualized in the peritoneal zone structure diagram; Provide structured risk warning statements and auxiliary diagnosis recommendation reports.

[0110] This mechanism significantly enhances doctors' comprehensibility, traceability and acceptability of AI results, and is a key supporting technology for the transition from algorithms to real clinical applications.

[0111] (5) Support system deployment and online migration capabilities to adapt to multi-center imaging heterogeneity and data update requirements.

[0112] This system adopts the Docker container deployment architecture, featuring lightweight deployment, standard interface integration (such as DICOM, FHIR, and HL7), and a continuous model training mechanism. It has the following technical advantages: It can be deployed on different hospital servers, connected to PACS and EMR systems, and data can flow in automatically; Supports online incremental training mechanisms (such as EWC or fine-tuning) to perform micro-model updates based on new cases; Provides controllable version management and model update logs to ensure algorithm security and compliance.

[0113] This strategy overcomes the performance degradation problem caused by the heterogeneity of data sources in the actual implementation of AI models, and significantly enhances the system's promotion capabilities.

[0114] In summary: The embodiments of this application achieve technological breakthroughs in many aspects, including system architecture, algorithm modeling, partition standardization, information fusion, visual interaction, and deployment strategies. An evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating PCI scores is proposed. The method has high precision, strong adaptability, interpretability, and engineering feasibility, providing a key technical platform for intelligent diagnosis and treatment of tumors, and possesses significant originality, advancement, and clinical translation value.

[0115] An embodiment of the present application also provides a computer-readable storage medium, which stores executable instructions. When the executable instructions are executed by at least one processor 301, the evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score described in the embodiment of the present application is implemented.

[0116] In some embodiments, the storage medium can be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface storage, an optical disc, or a compact disc read-only memory (CD-ROM); it can also be various devices including one or any combination of the above memories.

[0117] In some embodiments, executable instructions may be in the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0118] As an example, executable instructions may, but do not necessarily, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (for example, files storing one or more modules, subroutines, or code portions).

[0119] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0120] The computer-readable storage medium has the following advantages: (1) Joint dual-task prediction architecture: For the first time, the simultaneous output of peritoneal metastasis identification and PCI scoring is achieved.

[0121] Existing research has mostly focused on the classification of peritoneal metastasis status, lacking a quantitative expression of tumor burden. This system, for the first time, introduces a joint task framework, building a peritoneal metastasis classification model and a PCI score regression model in parallel within the same model. By sharing an image feature encoder and branching the decoding output, it achieves dual determination of "metastasis" and "metastasis degree" within a single process: Overcome the problems of single function and insufficient auxiliary dimension of existing forecasting systems; Meet the synchronous information needs of "status + classification" for preoperative auxiliary diagnosis and treatment; Improve model parameter sharing and learning efficiency.

[0122] This strategy can be extended to metastasis identification and load assessment tasks of other tumors and has universal expansion value.

[0123] (2) An anatomically driven automatic identification mechanism for peritoneal zonation to build a structured scoring space foundation.

[0124] This application example presents a groundbreaking method for automatic image region segmentation based on the thirteen-partition rule for the peritoneum and anatomical priors. The system incorporates a deep semantic segmentation network (such as nnU-Net) to perform fine-grained peritoneal segmentation, ensuring regional correspondence for subsequent PCI scoring. Define the spatial boundaries of 13 peritoneal regions and establish a spatial mapping structure for prediction output; Automatically assign scores (0-3 points) to each lesion zone and output the total PCI value; Avoid manual zoning errors and achieve standardized and reusable regional scoring.

[0125] This partitioning mechanism overcomes the problem that current AI models are unable to identify the spatial distribution structure of the peritoneum, providing support for precise disease management.

[0126] (3) Multimodal fusion and enhanced attention mechanism significantly improve the ability to perceive small lesions.

[0127] This system integrates traditional radiomics features (shape, grayscale, texture, etc.) + deep features (CNN / Transformer extraction) + clinical variables (such as CA125, CEA, etc.) to construct a three-channel multimodal input structure. It also introduces the SE attention mechanism and multi-scale residual network to adaptively adjust the importance of different modal features, enhancing the model's ability to integrate heterogeneous lesion characteristics: Effectively improve the ability to identify tiny, hidden peritoneal lesions; Improve the adaptability and robustness of the model under different hospital image acquisition conditions; Realize the coordinated expression of three layers of information: "imaging-clinical-structural".

[0128] This fusion strategy breaks through the limitations of existing models, such as strong dependence on specific modalities and poor generalization ability, and adapts to complex clinical reality scenarios.

[0129] (4) Enhanced model interpretability and structured visual output improve clinical usability and trust.

[0130] Different from traditional black-box AI prediction, the system of this invention integrates multiple model interpretation algorithms (such as Grad-CAM, SHAP, LRP) with a structured visualization interface, providing the following functions: Display significant area heatmaps of peritoneal metastasis classification output results; The PCI scores and predictions for each zone are visualized in the peritoneal zone structure diagram; Provide structured risk warning statements and auxiliary diagnosis recommendation reports.

[0131] This mechanism significantly enhances doctors' comprehensibility, traceability and acceptability of AI results, and is a key supporting technology for the transition from algorithms to real clinical applications.

[0132] (5) Support system deployment and online migration capabilities to adapt to multi-center imaging heterogeneity and data update requirements.

[0133] This system adopts the Docker container deployment architecture, featuring lightweight deployment, standard interface integration (such as DICOM, FHIR, and HL7), and a continuous model training mechanism. It has the following technical advantages: It can be deployed on different hospital servers, connected to PACS and EMR systems, and data can flow in automatically; Supports online incremental training mechanisms (such as EWC or fine-tuning) to perform micro-model updates based on new cases; Provides controllable version management and model update logs to ensure algorithm security and compliance.

[0134] This strategy overcomes the performance degradation problem caused by the heterogeneity of data sources in the actual implementation of AI models, and significantly enhances the system's promotion capabilities.

[0135] In summary: The embodiments of this application achieve technological breakthroughs in many aspects, including system architecture, algorithm modeling, partition standardization, information fusion, visual interaction, and deployment strategies. An evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating PCI scores is proposed. The method has high precision, strong adaptability, interpretability, and engineering feasibility, providing a key technical platform for intelligent diagnosis and treatment of tumors, and possesses significant originality, advancement, and clinical translation value.

[0136] In the several embodiments provided in this application, it should be understood that the disclosed methods and electronic devices can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0137] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0138] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0139] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, platform server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks, or optical disks.

[0140] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An evaluation method for identifying gastric cancer peritoneal metastasis status and estimating PCI score, characterized in that: The method comprises: Receive preoperative abdominal enhanced CT imaging data of gastric cancer patients; Preprocessing the preoperative abdominal enhanced CT image data, wherein the preprocessing includes resampling, normalization, image enhancement, and automatic positioning of the peritoneal region; wherein the automatic positioning is based on a regional attention mechanism; Extracting radiomics features and deep learning features from the region of interest (ROI) of the peritoneal region, and fusing the radiomics features and the deep learning features to obtain fused features; wherein the radiomics features include shape, texture, grayscale distribution, and edge structure features, and the deep learning features include image semantic features extracted by a multi-scale convolutional neural network; Constructing a peritoneal metastasis status classification model based on the fusion features; wherein the peritoneal metastasis status classification model is constructed based on a binary classification neural network with a residual network or a Transformer architecture, and the peritoneal metastasis status model outputs a predicted probability and a confidence score of whether peritoneal metastasis exists; The tumor burden in the peritoneal compartment was quantitatively assessed to obtain the PCI score for each peritoneal compartment; The final evaluation result is displayed based on the output result of the peritoneal metastasis status model and the evaluation result of the quantitative evaluation, wherein the final evaluation result includes at least one of the peritoneal metastasis status prediction result, PCI regional scores and spatial distribution maps, visual heat maps, model explanation outputs, and doctor-patient interaction records.

2. The method according to claim 1, characterized in that When extracting radiomics features and deep learning features from the region of interest (ROI) of the peritoneal area, the method further includes: Modeling the spatial topological structure information within the peritoneal area based on the graph neural network structure; Enhance the ability to express the relationship between lesions based on modeling results.

3. The method according to claim 1, characterized in that The peritoneal metastasis status classification model performs multimodal fusion prediction by simultaneously inputting imaging features and clinical features; wherein the clinical features include CEA, CA125, and tumor differentiation degree.

4. The method according to claim 1, wherein The tumor burden in the peritoneal partition is quantitatively assessed to obtain the PCI score for each peritoneal partition, including: The 13 peritoneal regions are segmented using the nnU-Net multi-label semantic segmentation network. Detect the number of lesions and the maximum nodule diameter in each partition; The PCI score of each peritoneal region is output based on the regression network and automatically accumulated to obtain the total PCI score; the PCI score of each peritoneal region has a score interval of [0, 3].

5. The method according to claim 4, characterized in that The PCI scoring results for each peritoneal region are accompanied by model confidence intervals or uncertainty indicators and provide information explaining the sources of error.

6. The method according to claim 1, characterized in that The final evaluation results display the predicted results and risk levels of each partition in the form of a 3D abdominal cavity reconstruction view to assist in preoperative lesion localization.

7. The method according to claim 1, characterized in that The model interpretation output is implemented based on Grad-CAM, Layer-wise Relevance Propagation or SHAP methods, and provides local image evidence for peritoneal metastasis prediction.

8. An evaluation system for identifying gastric cancer peritoneal metastasis status and estimating PCI scores, characterized by: The system comprises: An image data input module is used to receive preoperative abdominal enhanced CT image data of gastric cancer patients; An image preprocessing module, configured to preprocess the preoperative abdominal enhanced CT image data, wherein the preprocessing includes resampling, normalization, image enhancement, and automatic positioning of the peritoneal region; wherein the automatic positioning is based on a regional attention mechanism; a feature extraction module, configured to extract radiomics features and deep learning features from the region of interest (ROI) of the peritoneal region, and fuse the radiomics features and the deep learning features to obtain fused features; wherein the radiomics features include shape, texture, grayscale distribution, and edge structure features, and the deep learning features include image semantic features extracted by a multi-scale convolutional neural network; a peritoneal metastasis status recognition module, configured to construct a peritoneal metastasis status classification model based on the fusion features; wherein the peritoneal metastasis status classification model is constructed based on a binary classification neural network with a residual network or a Transformer architecture, and the peritoneal metastasis status model outputs a predicted probability and a confidence score for the presence of peritoneal metastasis; The PCI score estimation module is used to quantitatively evaluate the tumor burden in the peritoneal partition and obtain the PCI score of each peritoneal partition; An auxiliary diagnosis interaction module is used to display the final evaluation results based on the output results of the peritoneal metastasis status model and the evaluation results of the quantitative evaluation, wherein the final evaluation results include at least one of the peritoneal metastasis status prediction results, PCI regional scores and spatial distribution maps, visual heat maps, model explanation outputs, and doctor-patient interaction records.

9. The system according to claim 8, characterized in that The system further comprises: The system deployment and update module is used for local deployment and docking with the hospital PACS system interface, as well as dynamically updating model parameters based on the incremental training mechanism of new samples; the system deployment and update module is deployed based on the Docker container deployment architecture and integrates the HL7 or FHIR interface protocol to achieve data interoperability between hospital information systems.

10. An electronic device, characterized in that: include: A processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to perform the evaluation method for identifying the peritoneal metastasis status of gastric cancer and estimating the PCI score as described in any one of claims 1 to 7.

Images