Liver cancer immunotherapy prediction method and system based on spatial distribution characteristics of cd8+trm

Abstract

This invention discloses a method and system for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM cells, belonging to the field of medical artificial intelligence technology. This invention collects multi-source data from liver cancer patients, including radiomics, genomics, pathomics, and clinical data. It utilizes the ComBat algorithm and the nnU-Net framework for data standardization and image segmentation. Spatial distribution characteristics such as the density, aggregation, and invasion depth of CD8+TRM cells are extracted using digital pathology technology. A multimodal fusion algorithm based on an attention mechanism is employed to deeply integrate CD8+TRM features with image and genomic features, constructing an ensemble learning prediction model. This invention solves the problems of traditional prediction models relying on single indicators, lacking non-invasive methods, and being uninterpretable (like a black box). It achieves non-invasive and accurate prediction of immunotherapy efficacy using only routine imaging and clinical data, and can assess treatment response and provide early warning of toxic side effects in real time, assisting physicians in making individualized clinical decisions. It has extremely high clinical application value.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary fields of medical image processing, bioinformatics and artificial intelligence. Specifically, it relates to a method and system for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM. Background Technology

[0002] Hepatocellular carcinoma (HCC) is one of the most common malignant tumors worldwide, and immune checkpoint inhibitors (ICIs) have become a first-line treatment for advanced HCC. However, the objective response rate of current immunotherapy for HCC is only 20%-30%, and there is a lack of effective non-invasive predictive methods. Existing technologies have the following limitations: Prediction methods are limited: Traditional methods often rely on single biomarkers such as PD-L1 expression and TMB, ignoring the spatial complexity of the immune system; Limitations of invasive testing: Obtaining the spatial distribution characteristics of CD8+ TRM (tissue-resident memory T cells) usually relies on postoperative pathological sections or puncture biopsy, which is invasive and subject to sampling bias; Lack of multimodal fusion: Failure to effectively integrate imaging, genetic, pathological and clinical information leads to insufficient prediction accuracy; Poor model interpretability: Most existing AI models are "black boxes," making it difficult for doctors to trust their output results and hindering clinical translation.

[0003] Therefore, there is an urgent need for a non-invasive, highly interpretable immunotherapy prediction scheme for liver cancer that can integrate multimodal data. Summary of the Invention

[0004] To address the problems in related technologies, this invention proposes a method and system for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM, in order to overcome the aforementioned technical problems existing in the current related technologies.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention relates to a method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM, characterized by comprising the following steps: S1. Collect multi-source heterogeneous medical data of liver cancer patients, wherein the multi-source heterogeneous medical data includes at least radiomics data, genomics data, pathomics data and clinical data; S2. Preprocess and standardize the multi-source heterogeneous medical data to generate a standardized dataset; S3. Extract CD8+TRM spatial distribution features, image texture features, and molecular marker features from the standardized dataset; S4. Based on the multimodal fusion algorithm, the spatial distribution features of CD8+TRM, image texture features and molecular marker features are hierarchically fused to construct an immunotherapy response prediction model. S5. Input the standardized dataset into the immunotherapy response prediction model and output the efficacy prediction results and toxic side effect risk score of liver cancer immunotherapy.

[0006] Furthermore, S1 specifically includes: The patient's CT image data were collected at multiple time points, including multi-phase images of the plain scan phase, arterial phase, portal venous phase, and delayed phase. Genomic information was extracted from tumor tissue and peripheral blood samples from patients, with a focus on detecting CD69, CD103 biomarker genes and immune checkpoint genes. Tumor pathological sections were obtained and CD8+TRM cells were labeled by immunohistochemical staining. Pathological image data was generated by combining digital pathology technology. Collect patients' demographic information, laboratory indicators, and treatment plan information.

[0007] Further, S2 includes: The ComBat algorithm is used to correct image batch effects caused by different scanning devices; The nnU-Net deep learning framework was used to perform three-dimensional segmentation of tumor regions and liver parenchyma in CT images, and the segmentation was manually reviewed and corrected. Missing values ​​were filled using multiple interpolation, and the Z-score normalization method was used to normalize the data for each modality. Redundant features with a correlation coefficient greater than 0.9 were removed by screening using the Pearson correlation coefficient.

[0008] Further, S3 includes: Based on digital pathology analysis technology, CD8+TRM cells can be accurately identified and distinguished from other immune cells; Spatial localization information of CD8+TRM cells in tumor parenchyma, tumor stroma and liver tissue was extracted; Calculate the quantitative indicators of CD8+TRM cells, which include at least cell density per unit area, cell cluster aggregation coefficient, invasion depth, and spatial distance from tumor cells. LASSO regression and recursive feature elimination were used to screen out the key spatial distribution features that were most correlated with treatment response.

[0009] Further, S4 includes: First, single-modal feature extraction was performed on radiomics data, genomics data, and pathomics data, respectively. An attention mechanism is introduced to calculate the weight allocation of each modality feature in the prediction task, where the CD8+TRM spatial distribution feature is assigned the core weight. The weighted multimodal feature vectors are concatenated and fused to construct a fused feature matrix; An ensemble learning model is trained using the fused feature matrix. The ensemble learning model integrates logistic regression, random forest, XGBoost, and neural network algorithms.

[0010] Furthermore, it also includes model training and validation steps: Construct a multi-center retrospective dataset, which is divided into a training queue, an internal testing queue, and an external validation queue; Optimize model hyperparameters using 5-fold cross-validation; The model performance was evaluated using the area under the ROC curve (AUC) and the C-index. The AUC value of the external validation set was required to be no less than 0.75, and the C-index value of the overall survival prediction was required to be no less than 0.74.

[0011] Furthermore, it also includes a non-invasive mapping step: Establish a correlation mapping relationship between radiomics characteristics and the spatial distribution characteristics of CD8+TRM; By using a trained prediction model, the infiltration status and spatial distribution characteristics of CD8+TRM can be inferred by inputting only pre-treatment CT image data and routine clinical test data, achieving non-invasive prediction without the need for puncture biopsy.

[0012] Furthermore, after S5, it also includes: Receive follow-up data from patients during treatment and update the predictive model in real time; Based on the prediction results and clinical guidelines, individualized treatment strategy recommendations are generated, including recommendations for the selection of immunotherapy monotherapy or combination therapy, drug dosage adjustment, and combination of local therapy. By using Grad-CAM visualization and SHAP analysis, the importance ranking of features is output, explaining the basis for model predictions.

[0013] Furthermore, it also includes data visualization steps: The spatial distribution heatmap of CD8+TRM, the predicted survival curve, and the treatment risk assessment results are displayed graphically. The generated individualized treatment strategy recommendations will be output to the user's terminal in the form of a structured report.

[0014] A predictive system for liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM includes the following modules: The data acquisition module is used to collect radiomics data, genomics data, pathomics data, and clinical data from liver cancer patients. The data processing module is connected to the data acquisition module and is used to perform data preprocessing and standardization processes. The feature extraction module, connected to the data processing module, is used to extract CD8+TRM spatial distribution features, image texture features, and molecular marker features. The model building module, connected to the feature extraction module, is used to implement the multimodal fusion algorithm and build an immunotherapy response prediction model. The decision support module, connected to the model building module, is used to output treatment effect prediction results, toxic side effect risk scores, and individualized treatment strategy suggestions. The data storage and follow-up module is used to store patient data and iteratively update model parameters based on new follow-up data.

[0015] The present invention has the following beneficial effects: 1. In this invention, the spatial distribution characteristics of CD8+TRM are used as the core weights for multimodal fusion, establishing a causal chain from microscopic cell behavior to macroscopic clinical outcomes. Specifically, CD8+TRM cells, as key effector cells mediating local immune responses in tumors, directly reflect the body's immune recognition status through their high-density infiltration at the tumor margin. This invention quantifies this microscopic feature using digital pathology technology and establishes a correlation mapping with radiomics features. When the model outputs a high response probability, SHAP analysis clearly shows that CD8+TRM density contributes the most. This algorithm design based on immunological mechanisms provides a solid scientific explanation for the prediction results, significantly improving the interpretability of the model and the trust of clinicians.

[0016] 2. This invention employs a hierarchical fusion strategy and attention mechanism. This strategy utilizes an autoencoder to reduce the dimensionality of data at different scales and highlights the core position of CD8+TRM features through weight allocation. Its technical effect lies in achieving feature complementarity between microscopic molecules, mesoscopic cells, and macroscopic images. When faced with complex cases with atypical imaging manifestations but active immunity, the model can eliminate redundant noise through mutual verification of multi-source information, thereby obtaining more robust and accurate conclusions than a single-modality model. This significantly improves the AUC value and C-index of the external validation set, thereby enhancing the model's accuracy.

[0017] 3. In this invention, a feature inversion mechanism is constructed through training with a large amount of paired data in the early stage. The model learns the nonlinear mapping relationship between specific CT image texture features and CD8+TRM infiltration degree. Based on this, in the clinical inference stage, the system only needs to input conventional CT images and clinical data to infer the virtual CD8+TRM distribution state using the learned mapping relationship, thereby achieving true non-invasive prediction and reducing patient suffering and medical risks.

[0018] 4. This invention introduces a time-series data update and incremental learning mechanism. During the treatment process, the system receives new follow-up CT images and serological indicators, and dynamically corrects the predicted trajectory. For example, by comparing the virtual distribution changes of CD8+TRM at baseline and mid-treatment, the model can identify patients who truly benefit earlier and avoid erroneous discontinuation of medication due to premature ineffective assessment.

[0019] 5. This invention not only outputs a treatment effect prediction score, but also predicts the risk of toxic side effects through multimodal feature analysis, and generates specific individualized treatment strategy suggestions in combination with clinical guidelines. This ability to directly map the algorithm output to the clinical pathway solves the problem of blind clinical treatment decision-making, helps to reduce the proportion of ineffective treatment, lowers the incidence of toxic side effects, and ultimately improves the quality of life of liver cancer patients.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the liver cancer immunotherapy prediction method based on the spatial distribution characteristics of CD8+TRM of the present invention; Figure 2 This is a structural diagram of the liver cancer immunotherapy prediction system based on the spatial distribution characteristics of CD8+TRM of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0024] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "top", "middle", "inner", etc., which indicate orientation or positional relationship, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the invention.

[0025] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM, including the following steps: S1. Collect multi-source heterogeneous medical data from liver cancer patients. The system first constructs a multicenter retrospective dataset covering baseline data from at least 1000 patients diagnosed with hepatocellular carcinoma (HCC). Specific data collected includes: Radiomics data: Abdominal CT images of patients taken within one week prior to treatment were acquired, strictly including multi-phase images of the plain phase, arterial phase, portal venous phase, and delayed phase. Image slice thickness was ≤1mm to ensure the quality of 3D reconstruction.

[0026] Genomic data: Tumor tissue biopsy samples and cell-free DNA (cfDNA) were extracted from patients. Next-generation sequencing (NGS) was used to detect genes closely related to CD8+TRM infiltration and function, focusing on the expression levels of marker genes such as CD69 and CD103, as well as the mutation status of immune checkpoint genes such as PD-L1 and CTLA-4.

[0027] Pathogen data: Obtain tumor pathological sections from patients, label CD8+ TRM cells by multiplex immunofluorescence (mIF) or immunohistochemistry (IHC) staining, and generate high-resolution full-view digital pathological images (WSI) using a digital slide scanner.

[0028] Clinical data: Through the hospital information system (HIS) interface, we collect patients' demographic information (age, gender), laboratory indicators (AFP, PIVKA-II, PLR, etc.) and detailed clinical staging (BCLC staging).

[0029] S2. Preprocessing and standardizing multi-source heterogeneous medical data. To ensure data quality, strict standardized processes are implemented: Image correction: The ComBat algorithm is used to correct batch effects caused by equipment differences between different CT scanners (such as Siemens, GE, Philips).

[0030] Lesion segmentation: The nnU-Net deep learning framework was used to perform fully automated 3D segmentation of the tumor region (ROI) and liver parenchyma in CT images. Testing showed that the Dice similarity coefficient for this step reached over 0.99, and the results were manually reviewed and corrected by two senior radiologists.

[0031] Data cleaning: Genomic data were quality controlled using the GATK workflow; missing values ​​in clinical data were filled using multiple imputation; and all feature data were normalized using Z-score standardization.

[0032] Feature redundancy removal: Calculate the Pearson correlation coefficient between each feature and remove redundant features with a correlation coefficient greater than 0.9 to reduce model complexity.

[0033] S3. Extracting CD8+TRM spatial distribution features from the standardized dataset. Precise identification: Based on digital pathology analysis technology (using QuPath software), CD8+ TRM cells are accurately identified by combining morphological characteristics, and their differences from other immune cells such as CD56dim NK cells and memory B cells are distinguished.

[0034] Quantitative calculation: Spatial location information of CD8+TRM cells in the tumor core, tumor margin and adjacent liver tissue was extracted, and quantitative indicators were calculated, including cell density per unit area, cell cluster aggregation coefficient, invasion depth and Euclidean distance from tumor cells.

[0035] Feature screening: LASSO regression and recursive feature elimination (RFE) were used to screen out key spatial distribution features that were most correlated with the response to immunotherapy, such as the ratio of "tumor periphery CD8+TRM density to tumor core density".

[0036] S4. Constructing an immunotherapy response prediction model based on a multimodal fusion algorithm. Hierarchical Fusion: A hierarchical fusion strategy is adopted. First, single-modal features are extracted from radiomics, genomics, and pathomics data respectively. Then, a multi-head attention mechanism is introduced to calculate the weight allocation of each modality feature in the prediction task, where the CD8+TRM spatial distribution feature is assigned a core weight (e.g., 0.4).

[0037] Model Construction: The weighted multimodal feature vectors are concatenated and fused to construct a fused feature matrix. This matrix is ​​then used to train an ensemble learning model that integrates logistic regression, random forest, XGBoost, and one-dimensional convolutional neural network (1D-CNN) algorithms. The final prediction result is output through a soft voting mechanism.

[0038] Training and Validation: A multi-center retrospective dataset was constructed and divided into a training queue, an internal testing queue, and an external validation queue. Model hyperparameters were optimized using 5-fold cross-validation. Validation results show that the AUC value on the external validation set reached 0.82, and the C-index value for total survival prediction reached 0.78, significantly outperforming traditional models.

[0039] S5. Output prediction results and non-invasive mapping Non-invasive prediction: By establishing a correlation mapping relationship between radiomics features and the spatial distribution characteristics of CD8+TRM, the system can achieve non-invasive prediction. That is, by only inputting pre-treatment CT image data and routine clinical test data, the model can indirectly quantify the infiltration status and spatial distribution characteristics of CD8+TRM through feature inference, without the need for a second puncture biopsy.

[0040] Output results: The standardized dataset is input into the trained model, and the output results predict the efficacy of immunotherapy for liver cancer, including the objective response rate (ORR) probability, the progression-free survival (PFS) prediction curve, and the risk score of grade 3-4 immune-related adverse events (irAEs).

[0041] S6, Dynamic Updates and Decision Support Dynamic iteration: During treatment, the system receives the patient's follow-up CT data and serological indicators, updates the prediction model in real time, and dynamically adjusts the treatment response assessment.

[0042] Decision recommendations: Based on the predicted results and ASCO / ESMO clinical guidelines, individualized treatment strategy recommendations are generated. For example, for high-response, low-risk patients, "immunotherapy monotherapy maintenance therapy" is recommended; for low-response patients, "combination transarterial chemoembolization (TACE) or anti-angiogenic drugs" is recommended.

[0043] Interpretability analysis: Grad-CAM visualization technology is used to display hotspots in CT images that play a key role in prediction, and SHAP analysis is used to output the feature importance ranking, intuitively explaining why the model believes that the patient's CD8+TRM distribution predicts a good prognosis.

[0044] S7, Data Visualization Graphical Display: On the user interface, the system graphically displays the CD8+TRM spatial distribution heatmap, predicted survival curve, and treatment risk assessment radar chart.

[0045] Structured Reports: Automatically generates structured clinical reports, including predicted conclusions, confidence levels, a list of risk factors, and recommended individualized treatment plans for physician review and reference.

[0046] Through the above steps, this embodiment realizes a closed-loop process from data collection, feature extraction, model construction to clinical decision support, effectively solving the clinical pain point of lacking non-invasive and accurate prediction methods for liver cancer immunotherapy.

[0047] Example 2 like Figure 2 As shown, this embodiment provides a liver cancer immunotherapy prediction system based on the spatial distribution characteristics of CD8+TRM. The system includes: The data acquisition module is used to collect radiomics data, genomics data, pathomics data, and clinical data from liver cancer patients. The data processing module, connected to the data acquisition module, is used to perform data preprocessing and standardization processes. The feature extraction module, connected to the data processing module, is used to extract CD8+TRM spatial distribution features, image texture features, and molecular marker features. The model building module, connected to the feature extraction module, is used to implement the multimodal fusion algorithm and build an immunotherapy response prediction model. The decision support module, connected to the model building module, is used to output treatment effect prediction results, toxicity and side effect risk scores, and individualized treatment strategy suggestions. The data storage and follow-up module is used to store patient data and iteratively update model parameters based on new follow-up data.

[0048] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0049] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM, characterized in that, Includes the following steps: S1. Collect multi-source heterogeneous medical data of liver cancer patients, wherein the multi-source heterogeneous medical data includes at least radiomics data, genomics data, pathomics data and clinical data; S2. Preprocess and standardize the multi-source heterogeneous medical data to generate a standardized dataset; S3. Extract CD8+TRM spatial distribution features, image texture features, and molecular marker features from the standardized dataset; S4. Based on the multimodal fusion algorithm, the spatial distribution features of CD8+TRM, image texture features and molecular marker features are hierarchically fused to construct an immunotherapy response prediction model. S5. Input the standardized dataset into the immunotherapy response prediction model and output the efficacy prediction results and toxic side effect risk score of liver cancer immunotherapy.

2. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, S1 specifically includes: The patient's CT image data were collected at multiple time points, including multi-phase images of the plain scan phase, arterial phase, portal venous phase, and delayed phase. Genomic information was extracted from tumor tissue and peripheral blood samples from patients, with a focus on detecting CD69, CD103 biomarker genes and immune checkpoint genes. Tumor pathological sections were obtained and CD8+TRM cells were labeled by immunohistochemical staining. Pathological image data was generated by combining digital pathology technology. Collect patients' demographic information, laboratory indicators, and treatment plan information.

3. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, S2 includes: The ComBat algorithm is used to correct image batch effects caused by different scanning devices; The nnU-Net deep learning framework was used to perform three-dimensional segmentation of tumor regions and liver parenchyma in CT images, and the segmentation was manually reviewed and corrected. Missing values ​​were filled using multiple interpolation, and the Z-score normalization method was used to normalize the data for each modality. Redundant features with a correlation coefficient greater than 0.9 were removed by screening using the Pearson correlation coefficient.

4. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, S3 includes: Based on digital pathology analysis technology, CD8+TRM cells can be accurately identified and distinguished from other immune cells; Spatial localization information of CD8+TRM cells in tumor parenchyma, tumor stroma and liver tissue was extracted; Calculate the quantitative indicators of CD8+TRM cells, which include at least cell density per unit area, cell cluster aggregation coefficient, invasion depth, and spatial distance from tumor cells. LASSO regression and recursive feature elimination were used to screen out the key spatial distribution features that were most correlated with treatment response.

5. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, S4 includes: First, single-modal feature extraction was performed on the radiomics data, genomics data, and pathomics data, respectively. An attention mechanism is introduced to calculate the weight allocation of each modality feature in the prediction task, where the CD8+TRM spatial distribution feature is assigned the core weight. The weighted multimodal feature vectors are concatenated and fused to construct a fused feature matrix; An ensemble learning model is trained using the fused feature matrix. The ensemble learning model integrates logistic regression, random forest, XGBoost, and neural network algorithms.

6. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, It also includes model training and validation steps: Construct a multi-center retrospective dataset, which is divided into a training queue, an internal testing queue, and an external validation queue; Optimize model hyperparameters using 5-fold cross-validation; The model performance was evaluated using the area under the ROC curve (AUC) and the C-index. The AUC value of the external validation set was required to be no less than 0.75, and the C-index value of the overall survival prediction was required to be no less than 0.

74.

7. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, It also includes a non-invasive mapping step: Establish a correlation mapping relationship between radiomics characteristics and the spatial distribution characteristics of CD8+TRM; By using a trained prediction model, the infiltration status and spatial distribution characteristics of CD8+TRM can be inferred by inputting only pre-treatment CT image data and routine clinical test data, achieving non-invasive prediction without the need for puncture biopsy.

8. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, Following S5, the following also includes: Receive follow-up data from patients during treatment and update the predictive model in real time; Based on the prediction results and clinical guidelines, individualized treatment strategy recommendations are generated, including recommendations for the selection of immunotherapy monotherapy or combination therapy, drug dosage adjustment, and combination of local therapy. By using Grad-CAM visualization and SHAP analysis, the importance ranking of features is output, explaining the basis for model predictions.

9. The method for predicting liver cancer immunotherapy based on the spatial distribution characteristics of CD8+TRM according to claim 1, characterized in that, It also includes data visualization steps: The spatial distribution heatmap of CD8+TRM, the predicted survival curve, and the treatment risk assessment results are displayed graphically. The generated individualized treatment strategy recommendations will be output to the user's terminal in the form of a structured report.

10. A liver cancer immunotherapy prediction system based on CD8+TRM spatial distribution characteristics, used to implement the liver cancer immunotherapy prediction method based on CD8+TRM spatial distribution characteristics as described in any one of claims 1-9, characterized in that, Includes the following modules: The data acquisition module is used to collect radiomics data, genomics data, pathomics data, and clinical data from liver cancer patients. The data processing module is connected to the data acquisition module and is used to perform data preprocessing and standardization processes. The feature extraction module, connected to the data processing module, is used to extract CD8+TRM spatial distribution features, image texture features, and molecular marker features. The model building module, connected to the feature extraction module, is used to implement the multimodal fusion algorithm and build an immunotherapy response prediction model. The decision support module, connected to the model building module, is used to output treatment effect prediction results, toxic side effect risk scores, and individualized treatment strategy suggestions. The data storage and follow-up module is used to store patient data and iteratively update model parameters based on new follow-up data.

Images