Multi-mode prostate cancer biochemical recurrence risk layered prediction system based on artificial intelligence

Abstract

The invention provides a multi-mode prostate cancer biochemical recurrence risk layering prediction system based on artificial intelligence. Based on an Xgboost framework, a postoperative patient pathological panoramic pathological section scanning image is analyzed through end-to-end, multi-scale, multi-center and large-sample analysis, pathological information is utilized to the maximum extent, meanwhile, the prognosis risk of a patient can be evaluated more comprehensively in combination with clinical indexes such as CAPRA-S scores, and the method has obvious advantages compared with a traditional model. The method aims at better fitting the use scene of a hospital, the risk of prostate cancer recurrence of a patient is more efficiently and accurately predicted by fusing pathological section features and clinical features after a radical operation, and the risk of recurrence of the patient within 3 years and longer time after the radical operation can be accurately predicted. And an interpretable module is further combined to assist a doctor to interpret a result, so that precise layering and personalized follow-up visit of the BCR risk of the prostatic cancer patient are realized, the risk of excessive treatment and missed diagnosis is reduced, and the method has a good application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of predicting and analyzing the risk of biochemical recurrence of prostate cancer, and particularly relates to a multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence. Background Art

[0002] After radical prostatectomy for prostate cancer patients, 30-40% of the patients are observed to have an elevated prostate-specific antigen (PSA) level, indicating disease recurrence. Patients with a PSA level exceeding 0.2 ng / mL two or more times after surgery are considered to have experienced biochemical recurrence (BCR). BCR is an early signal of metastasis and cancer recurrence, often requiring interventions such as targeted radiotherapy and hormone blockade therapy. Currently, there is no good indicator for predicting BCR to guide the follow-up frequency after surgery. The traditional prognostic assessment paradigm for prostate cancer mainly relies on the pathological Gleason grading system, which is regarded as a reference standard for tumor BCR risk assessment to a certain extent. However, a large amount of clinical practice data has fully confirmed that even patients with a relatively low Gleason score may still experience biochemical recurrence within a very limited time period. The prediction mode relying solely on the Gleason score has difficulty meeting the accurate clinical risk stratification requirements.

[0003] In response to the above problems, based on a large amount of clinical practice data, prognostic assessment is generated by combining indicators such as preoperative PSA level, Gleason score staging, and clinical staging, and related prediction models such as CAPRA-S-Score are thus born. However, due to the large heterogeneity among individuals, the widespread multi-dimensional data missing in clinical data, and the systematic differences in data collection standards among different medical institutions, these factors further increase the cognitive and technical difficulties of accurate prediction. Using only clinical data to predict biochemical recurrence is still not the optimal method.

[0004] Artificial intelligence (AI) is a new technical science of theories, methods, and technologies that are good at simulating, extending, and expanding human intelligence. With the rapid development of artificial intelligence, its powerful functions play a crucial role in processing and integrating complex information resources with multiple parameters and dimensions in clinical practice. Deep learning is a type of machine learning algorithm that includes neural network models. Its application in the early diagnosis and prognostic assessment of cancer has emerged. Recently, for the inventions reported on other cancer types, deep learning models that analyze whole-slide pathology scan images (WSIs) of pathological sections can accurately predict disease types and patient prognoses, with good accuracy and stability. This digital pathology image analysis paradigm based on artificial intelligence provides a new technical path for the accurate prediction of the risk of biochemical recurrence of prostate cancer.

[0005] Although artificial intelligence has made certain progress in the field of digital pathology image analysis, there are still many key technical challenges to achieve true clinical translation. Pathological image processing needs to handle massive multi-dimensional data, and the algorithm complexity of feature extraction and selection is extremely high. The image quality and staining differences between different pathological sections pose numerous tests to the stability and generalization ability of the model. More critically, the prediction model for clinical applications must have a high degree of interpretability. Clinicians need to clearly and transparently understand the decision-making mechanism of the model and quantitatively evaluate the marginal contribution of each feature to the prediction result. This requires that artificial intelligence should not be a closed "black box" prediction tool, but an intelligent auxiliary decision-making system that can provide intuitive and understandable explanations. Therefore, how to effectively integrate clinical indicators and pathological image features to construct a multi-modal prediction model that is both accurate and highly interpretable has become the core paradigm of current precision medicine research for prostate cancer.

[0006] Currently, an artificial intelligence prediction system that can integrate clinical and pathological data and achieve accurate stratification of the biochemical recurrence risk of prostate cancer has not been constructed. Existing research is either limited to a single data modality or lacks sufficient clinical interpretability. Given the individual differences in the prognosis of prostate cancer patients and the decisive impact of biochemical recurrence on patient survival, there is an urgent need for a more intelligent, accurate, and clinically transformative risk prediction methodological paradigm. The present invention aims to break through the cognitive limitations of traditional prediction models and construct a multi-modal intelligent prediction system that can provide personalized prognosis assessment by innovatively integrating clinical big data, artificial intelligence technology, and digital pathology image analysis. Summary of the Invention

[0007] The present invention aims to fill this gap and develop an artificial intelligence-based multi-modal prostate cancer biochemical recurrence risk stratification prediction system. A prediction model based on AI methods is designed. By integrating and analyzing the clinical information and pathological information of prostate cancer patients, it realizes the stratification prediction of the biochemical recurrence risk of prostate cancer in patients, provides a reliable basis for predicting the biochemical recurrence risk of prostate cancer, and provides strong help and guidance for reducing the negative impact of postoperative diseases and improving the survival rate.

[0008] To achieve this purpose, the specific technical solutions of the present invention are as follows:

[0009] In the first aspect of the present invention, a multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence is provided. By integrating multi-modal data such as whole-slide pathology scan images of pathological sections and clinical indicators, it can better achieve the early prediction of postoperative biochemical recurrence in prostate cancer patients, and help clinicians formulate more personalized postoperative follow-up and adjuvant treatment plans. This system has the following technical features, including:

[0010] An input display module, which is used to input the original medical record report of the patient and the surgical pathology report containing the panoramic pathology slide scan image with pathological sections, and display the analysis results after the analysis is completed;

[0011] A clinical feature extraction and processing module, which automatically extracts the clinical feature information of the patient from the original medical record report and the surgical pathology report, automatically determines and fills in the missing values, and then scores the filled information using a risk stratification algorithm based on the CAPRA-S scoring system to form a standardized clinical feature vector;

[0012] A pathological feature extraction and processing module. After the image is preprocessed, a multi-instance algorithm based on the pseudo-packet strategy and cyclic cross-attention is used. First, the deep features of tumor images at different objective lens magnifications are extracted, and then the feature representation at the panoramic pathology slide scan image level is generated in a weakly supervised network. The microscopic cell abnormalities and macroscopic tissue change features are extracted to obtain the pathological feature representation, and the biochemical recurrence prediction score at the pathological level is fused and generated;

[0013] A fusion prediction module, which performs multi-modal feature fusion on the pathological feature representation and the standardized clinical feature vector information to construct an end-to-end prostate cancer biochemical recurrence risk prediction model; highlights the weights of key features through an attention mechanism, analyzes the pathological feature representation and the standardized clinical feature vector information, classifies the patient into three risk levels of high risk, medium risk, and low risk, and visualizes the key prediction factors;

[0014] An interpretability module, which provides in-depth interpretations from three aspects: pathological images, fusion models, and similar cases, quantitatively displays the contribution degrees of each clinical index and pathological feature to the prediction result, and provides the judgment basis for understanding the model from multiple dimensions of vision, quantitative analysis, and analogical reasoning;

[0015] A storage module, which is used to store the information processed by each module,

[0016] A control module, which synthesizes the manual judgment result and the risk prediction result, and transmits the final result to the input display module.

[0017] The preferred technical solutions of each module in the system are as follows:

[0018] (1) Input display module

[0019] It is used to input the original medical record report of the patient and the surgical pathology report containing the panoramic pathology slide scan image with pathological sections, and display the analysis results after the analysis is completed. In the present invention, the input display module is the data entry of the system, and an intelligent adaptive interface is designed. Its intelligent design aims to achieve efficient and accurate collection of patient clinical data. For example, this module supports clinicians to directly upload or paste the original medical record and surgical pathology reports.

[0020] (2)Clinical Feature Extraction and Processing Module

[0021] It includes a clinical feature extraction sub-module and a clinical data processing sub-module.

[0022] The clinical feature extraction sub-module is a large language model API integrated into the system (such as GPT-4o, Claude, etc.), which can intelligently parse text content and is used to efficiently collect and process multi-dimensional clinical data of patients, including key indicators that make up the CAPRA-S Score, such as patient age, preoperative PSA level, pathological Gleason score, clinical pathological stage, surgical margin status, seminal vesicle invasion, etc.

[0023] During the extraction process, the accuracy of the extraction results is ensured based on the built-in data verification mechanism. For fields with ambiguity or uncertainty, the system will fluorescently mark the clinical features in the bold table to prompt the clinician for review, so as to better achieve the intelligent and standardized integration of clinical and pathological data and lay a solid data foundation for subsequent artificial intelligence risk prediction.

[0024] The clinical data processing sub-module first fuses the patient characteristics with the training cohort, and uses the Multiple Imputation by Chained Equations (MICE) algorithm to accurately fill in the missing values of the patient's key clinical variables, and send a warning when the number of missing values is not less than the preset number (such as >4 missing values); then, the filled data is scored by a risk stratification algorithm based on the CAPRA-S scoring system, and the PSA level, Gleason score, clinical pathological stage, surgical margin status, and lymph node status are assigned as follows: the PSA level scoring range is 0-4 points (corresponding to <6, 6-10, 10-20, 20-30, ≥30 respectively), the Gleason score range is 0-3 points (corresponding to ISUP grades 1-5, where ISUP grades 4 and 5 are 3 points), the clinical pathological stage is 0-1 point (0, T1 / T2 stage, 1, T3 / T4 stage), the surgical margin status is 0-2 points (0, R0 status, 1, R1 status, 2, R2 status), and lymph node involvement is 0-1 point (0, N0 status, 1, N1 status). Finally, a continuous risk assessment of 0-7 points is achieved, and the patients are accurately stratified into three levels: low risk (0-2 points), medium risk (3-5 points), and high risk (6 points and above).

[0025] (3)Pathological Feature Extraction and Processing Module

[0026] It includes a feature vector processing sub-module, a cyclic cross-attention sub-module, and a multi-scale feature fusion sub-module.

[0027] The feature vector processing sub-module first preprocesses the panoramic pathological section scanning image. In the present invention, the panoramic pathological section scanning image is preferably the panoramic pathological section scanning image of HE-stained tissue. The feature vector processing sub-module first annotates the tumor region based on the QuPath software for the HE-stained pathological whole-section image of the tissue, and exports the annotation information in the GeoJSON format; then preprocesses to exclude the background region, and the specific method is as follows: by adjusting the mean and standard deviation of the image patch data, the data is converted into a distribution with zero mean and unit variance, realizing the standardization of the data and the normalization of the image intensity. Subsequently, the tumor region is cropped into non-overlapping image patches of 256×256 pixels at the magnification of 10×, 20×, and 40× objective lenses, and a self-supervised vision encoder UNI is used to automatically extract features therefrom. After the last convolutional layer of UNI, 1024-dimensional feature vectors of each image patch are extracted respectively, generating a feature representation at the panoramic pathological section scanning image level. Each panoramic pathological section scanning image will generate a set of low-dimensional feature matrices at three magnification scales, with a size of {Ns×1024}, where N represents the number of image patches at this scale, and s∈{10×, 20×, 40×}.

[0028] The pseudo-packet strategy is used to solve the overfitting problem on the one hand, and at the same time predict the features of the BCR that may be sparse but crucial, such as cell populations with high metastatic potential. The feature vector processing sub-module refines the features based on the pseudo-packet strategy, while capturing the local key region information, retains the overall structural features, and improves the model's recognition and discrimination ability for heterogeneous lesions in the image. In the subsequent four-dimensional reconstruction of features and convolutional processing of different sizes, first, the feature representation at the panoramic pathological section scanning image level input, or the feature packet representation of the pseudo-packet and the parent packet, is reconstructed into a four-dimensional tensor X reshaped : , and then convolutional kernels of different sizes (7×7, 5×5, 3×3) are used to process the feature data to further extract features with spatial perception and context information, providing a more comprehensive and richer feature expression. By retaining the spatial correlation in the pathological image, the modeling of tissue morphological features and the extraction of spatial features are realized.

[0029] The cyclic cross-attention sub-module complements convolutional kernels of different scales, and gradually generates a new feature map with dense and rich context information based on cross-attention iteration, enabling the model to simultaneously identify microscopic cell abnormalities and macroscopic tissue changes; such as microscopic cell abnormalities including different cell sizes, nuclear atypia, increased chromatin, multiple mitotic figures, and abnormal cell metabolism; the macroscopic tissue changes include irregular gland size, disordered arrangement, basement membrane destruction, perineural invasion, accompanied by necrosis and bleeding, and interstitial fibrosis, etc.

[0030] The multi-scale feature fusion sub-module fuses the feature information at three objective lens magnification scales, performs weighted averaging on the biochemical recurrence probabilities obtained at different objective lens magnifications, and finally generates the predicted risk score for the patient to obtain the recurrence or non-recurrence prediction result.

[0031] Multi-scale fusion is crucial in pathological image analysis because features at different scales can complement each other and provide more comprehensive information. For example, a low-magnification objective lens (such as 10×) helps identify the overall distribution, scope of cancerous lesions, and their relationship with surrounding tissues, such as the size, proportion, and invasion of cancerous lesions. A high-magnification objective lens (such as 40×) reveals morphological details at the cellular level, such as the morphology of cancer cells, nuclear atypia, and nuclear-cytoplasmic ratio. Through multi-scale fusion, the model can better understand the hierarchical structure and improve the ability to identify complex features. In addition, this method also enhances the robustness of the model, reduces the impact of noise and information loss, and can simulate the observation methods of pathologists at different magnifications to improve diagnostic accuracy.

[0032] (4) Fusion prediction module

[0033] The fusion prediction module aims to achieve accurate stratification and personalized prediction of the risk of biochemical recurrence (BCR) of prostate cancer. Through a multi-modal feature deep fusion strategy, this module integrates deep learning features at the panoramic pathological section scan image level, clinical CAPRA-S scores, and key clinical information to construct an end-to-end artificial intelligence risk prediction system.

[0034] Specifically, the fusion prediction module first fuses the CAPRA-S score, age, clinical pathological stage, seminal vesicle invasion, surgical margin status, preoperative PSA level, and biochemical recurrence prediction score at the pathological level through the Xgboost model, and highlights the weights of key features through the attention mechanism; then calculates the attention scores (W i × score) of each feature, performs weighted fusion on the features, and obtains high-risk, medium-risk, and low-risk three-level risk grades based on the calculation and analysis results; then uses the survival:aft accelerated failure time objective function for modeling, and predicts the survival time of the patient based on the relationship between the clinical features learned by the model and the survival time distribution.

[0035] In the present invention, the Softmax algorithm is used to implement the dynamic weight allocation of the above features, that is, the attention mechanism automatically learns the weight of each feature in the current task by calculating the correlation or importance between features. The calculation formula is as follows:

[0036] Among them, Q and K are the representations of features, K T refers to the transpose of matrix K, and d is the dimension.

[0037] The assignment of weights depends on the overall distribution of the input data, such as the combined relationship of the above-mentioned characteristics of a certain patient, etc.

[0038] In the code implementation, first, through the Xgboost extreme gradient boosting tree model, seven features such as CAPRA-S score, age, clinical pathological stage, seminal vesicle invasion, margin status, PSA level, and biochemical recurrence prediction score at the pathological level are integrated. The survival:aft accelerated failure time objective function is used for modeling, and parameter optimization is carried out through grid search. The parameter selection is based on cross-validation and ROC-AUC evaluation metrics, and its calculation formula is: 。

[0039] Based on the relationship between features and survival time, a prediction model under the Xgboost framework is established. This model takes clinical features as input and outputs a recurrence risk score. The objective function of the Xgboost model is survival:aft, which is an Accelerated Failure Time (AFT) model specifically for survival analysis. The AFT model directly models the survival time, assuming that covariates (i.e., clinical features) affect the survival time by accelerating or decelerating time. In this case, if BCR = 1: BCR_TimeMonths represents the actually observed biochemical recurrence time. If BCR = 0: BCR_TimeMonths represents that no biochemical recurrence is observed during the follow-up period, so this is regarded as a censored data, and BCR_TimeMonths corresponds to the censoring time. The model predicts the survival time of each patient by learning the relationship between clinical features and the survival time distribution. Due to the use of the AFT model, the model actually predicts the acceleration factor, which describes how clinical features affect the survival time. Therefore, the finally predicted risk score actually reflects the survival time of the patient, that is, the time when the patient has biochemical recurrence predicted by the model. The lower the score, the shorter the recurrence time of the patient and the higher the recurrence risk; on the contrary, the higher the risk score, the lower the recurrence risk of the patient.

[0040] (5) Interpretability module

[0041] The interpretability module is an important part of the AI-based multimodal prostate cancer biochemical recurrence risk stratification prediction system, including a pathological image interpretability sub-module, a fusion model interpretability sub-module, and a similar case comparison and analysis sub-module. Its core design aims to improve the transparency and clinical credibility of the prediction results. This module consists of three subsystems working in coordination, aiming to provide in-depth interpretations from three aspects: pathological images, fusion models, and similar cases, enabling clinicians to understand the basis of the model's judgment from multiple dimensions such as vision, quantitative analysis, and analogical reasoning, significantly enhancing the transparency and credibility of AI-assisted decision-making, and providing stronger support for individualized follow-up and treatment plan formulation.

[0042] 1) The pathological image interpretability sub-module intelligently analyzes pathological images based on the Grad-CAM technique, accurately locates pathological regions that contribute significantly to the prediction score through multi-scale heatmaps, presents important morphological features at different magnifications (10×, 20×, 40×), and the key pathological features extracted include: glandular structure, nuclear atypia, cell infiltration, etc. A built-in visualization calibration mechanism is used to ensure that the extraction results are basically consistent with the pathologist's judgment. For pathological regions with ambiguity or uncertainty, the system will fluoresce and bold the important regions in the heatmap to prompt the pathologist for review, so as to better integrate clinical experience and AI analysis and provide a more accurate pathological basis for subsequent risk prediction.

[0043] 2) The fusion model interpretability sub-module uses the SHAP value calculation method to quantitatively analyze the contribution degree of all input variables, generating waterfall charts and force diagrams to visually display the positive and negative impact degrees of each clinical index (PSA, Gleason score, margin status, etc.) and pathological features on the final risk prediction result. The contribution degrees of the key clinical indicators focused on cover the core variables composed of the CAPRA-S Score, such as patient age, preoperative PSA level, Gleason score, clinicopathological stage, margin status, etc. A built-in contribution degree verification mechanism is used to ensure the accuracy of the SHAP value calculation results. For indicators with abnormal contribution degrees or inconsistent with clinical experience, the system will fluoresce and bold the key variables in the waterfall chart to prompt the clinician for review, so as to better integrate clinical cognition and AI analysis and improve the credibility of the model prediction results.

[0044] 3) The similar case comparison and analysis sub-module automatically retrieves 30 patient cases with the most similar feature expressions and prediction scores from the internal training set based on the KNN algorithm, and presents their key clinical baseline indicators, treatment plans, and actual biochemical recurrence results in tabular form, forming a risk reference spectrum based on real cases. The key clinical baseline indicators for key reference include: age, preoperative PSA level, Gleason score, clinical TN stage, treatment plan, follow-up time, and biochemical recurrence time, etc., and a similarity verification mechanism is built in to ensure the accuracy of the retrieval results. For features with significant differences between similar cases and the target patient, the system will fluorescently mark the key variables in the bold table to prompt clinicians for review, so as to better integrate individual characteristics and group experience and provide a more comprehensive reference for formulating personalized treatment plans.

[0045] In the second aspect of the present invention, a construction method of the above prediction system and its application in clinical practice are provided, including the following steps:

[0046] (1) Dataset division: Clinically relevant data and whole-slide pathology data of patients after radical prostatectomy are retrospectively collected from multiple centers. The scanned images of HE-stained panoramic pathology sections of tissues determined by doctors are aggregated into a total dataset, and the total dataset is divided into a test set and a validation set;

[0047] (2) Clinical feature extraction and processing module, by accessing the large prediction model API, automatically extracts the clinical feature information of patients (age, preoperative PSA level, Gleason score, clinical pathological stage, surgical margin status, lymph node status, seminal vesicle invasion) from the original medical record reports and surgical pathology reports, and automatically determines and fills in missing values, and then performs risk stratification algorithm scoring based on the CAPRA-S scoring system on the filled information to form a standardized clinical feature vector;

[0048] (3) For the scanned images of HE-stained panoramic pathology sections of tissues after preprocessing, the pathological feature extraction and processing module uses a multi-instance algorithm based on the pseudo-packet strategy and cyclic cross-attention. First, it extracts the deep features of tumor images at different objective lens magnifications, and then generates feature representations at the level of panoramic pathology section scanned images in a weakly supervised network, extracts microscopic cell abnormalities and macroscopic tissue change features, obtains pathological feature representations, and fuses them to generate a biochemical recurrence prediction score at the pathological level;

[0049] (4) Fusion prediction module, performs multi-modal feature fusion on the pathological feature representation and the standardized clinical feature vector information to construct an end-to-end prostate cancer biochemical recurrence risk prediction model; highlights the weights of key features through an attention mechanism, and then calculates the attention scores of each feature (W iAfter calculating the CAPRA-S score (× score) and performing weighted fusion of features, the patients are classified into three risk levels: high risk, medium risk, and low risk based on the calculation and analysis results, and the key predictive factors are visualized.

[0050] (5) Train and validate the above evaluation system in multiple independent validation cohorts and compare it with traditional clinical prognosis evaluation models.

[0051] In the third aspect of the present invention, a non-transitory computer-readable storage medium is provided, which can be installed in communication tools such as computers. A computer program is stored thereon, and when the computer program is executed by a processor, the above-described steps are implemented.

[0052] Functions and effects of the invention

[0053] The present invention establishes an artificial intelligence-based multimodal biochemical recurrence risk stratification prediction system for prostate cancer. Based on the Xgboost framework, by performing end-to-end, multi-scale, multi-center, and large-sample analysis on the whole-slide pathology scan images of postoperative patients' pathology, it can maximize the utilization of pathological information and provide reliable clinical facts with a complete long-term follow-up process. At the same time, combining clinical indicators such as the CAPRA-S score can more comprehensively evaluate the prognosis risk of patients, having obvious advantages over traditional models. This method aims to better fit the hospital's usage scenarios in multiple aspects. By fusing the pathological slice features and clinical features after radical surgery, it can more efficiently and accurately predict the risk of prostate cancer recurrence in patients, helping to accurately predict the risk of recurrence in patients within 3 years and longer after surgery. Combining an interpretable module to assist doctors in interpreting the results, and then performing precise stratification and personalized follow-up of the BCR risk of prostate cancer patients, reducing the risks of over-treatment and missed diagnosis, and having good application prospects. Description of the drawings

[0054] Figure 1 Shows the composition framework diagram of the system of the present invention;

[0055] Figure 2 Shows the model process of the training and validation stages of the system of the present invention.

[0056] Figure 3 Shows the internal cohort training validation ROC curve.

[0057] Figure 4 Shows the external validation cohort ROC curves, which are the external validation ROC curves of the Chinese multi-center cohort (A), TCGA (B), and PLCO (C) from left to right.

[0058] Figure 5 Shows the KM survival prediction curves of four cohorts.

[0059] Figure 6Displays the analysis results of the interpretability module. Detailed implementation

[0060] The present invention will be described in detail below in conjunction with embodiments and drawings. However, the following embodiments should not be regarded as limiting the scope of the present invention.

[0061] I. System structure

[0062] The structure of the artificial intelligence-based multimodal prostate cancer biochemical recurrence risk stratification prediction system 100 in the present invention is shown in Figure 1 , and includes an input display module 1, a clinical feature extraction and processing module 2, a pathological feature extraction and processing module 3, a fusion prediction module 4, an interpretability module 5, a storage module 6, and a control module 7.

[0063] The input display module 1 is the data entry of the system, which is used to input the original medical record report of the patient and the surgical pathology report containing the panoramic pathological section scan image of the pathological section, and display the analysis results after the analysis is completed. This module is designed as an intelligent adaptive interface, aiming to realize the efficient and accurate collection of the patient's clinical data, and support the clinician to directly upload or paste the original medical record and surgical pathology report.

[0064] The clinical feature extraction and processing module 2 includes a clinical feature extraction sub-module 21 and a clinical data processing sub-module 22. The clinical feature extraction sub-module 21 is a large language model API integrated in the system (such as GPT-4o, Claude, etc.), which is used to efficiently collect and process the multi-dimensional clinical data of the patient, including the key indicators of CAPRA-SScore composed of the patient's age, preoperative PSA level, pathological Gleason score, clinical pathological stage, surgical margin status, seminal vesicle invasion, etc.; the clinical data processing sub-module 22 first fuses the patient's characteristics and the training cohort, and uses the multiple imputation chained equations (MICE) algorithm to accurately fill the missing values of the key clinical variables of the patient, and sends a warning when the number of missing values is not less than the preset number (such as >4 missing values); then, the filled data is scored by a risk stratification algorithm based on the CAPRA-S scoring system, and the PSA level, Gleason score, clinical pathological stage, surgical margin status, and lymph node status are scored and assigned values, and finally a continuous risk assessment of 0-7 points is realized, and the patient is accurately stratified into three levels: low risk (0-2 points), medium risk (3-5 points), and high risk (6 points and above).

[0065] The pathological feature extraction and processing module 3 includes a feature vector processing sub-module 31, a cyclic cross-attention sub-module 32, and a multi-scale feature fusion sub-module 33. The feature vector processing sub-module 31 first preprocesses the panoramic pathological slice scan image, and then crops the tumor region into non-overlapping image patches of 256×256 pixels at the magnification of 10×, 20×, and 40× objective lenses. Then, a self-supervised vision encoder UNI is used to automatically extract features from it, and the features are refined based on the pseudo-packet strategy, while capturing the local key region information and retaining the overall structural features, improving the model's recognition and discrimination ability for heterogeneous lesions in the image; the cyclic cross-attention sub-module 32 then performs four-dimensional feature reconstruction and convolution processing of different sizes on the feature representation at the panoramic pathological slice scan image level or the feature packet representation of the pseudo-packet and the parent packet. First, it reconstructs it into a four-dimensional tensor X reshaped : , and then uses convolution kernels of different sizes (7×7, 5×5, 3×3) to process the feature data to further extract features with spatial perception and context information; the multi-scale feature fusion sub-module 33 fuses the feature information at three objective lens magnification scales, performs weighted averaging on the biochemical recurrence probabilities obtained at different objective lens magnifications, and finally generates the predicted risk score of the patient to obtain the recurrence or non-recurrence prediction result.

[0066] The fusion prediction module 4 aims to achieve accurate stratification and personalized prediction of the risk of biochemical recurrence (BCR) of prostate cancer. Through a multi-modal feature deep fusion strategy, this module integrates the deep learning features at the panoramic pathological slice scan image level, the clinical CAPRA-S score, and key clinical information to construct an end-to-end artificial intelligence risk prediction system.

[0067] Specifically, first, the CAPRA-S score, age, clinical pathological stage, seminal vesicle invasion, surgical margin status, preoperative PSA level, and biochemical recurrence prediction score at the pathological level are fused through the Xgboost model, and the weights of key features are highlighted through the attention mechanism; then, the attention scores (W i ×score) of each feature are calculated, and after weighted fusion of the features, high-risk, medium-risk, and low-risk three-level risk grades are obtained based on the calculation and analysis results; then, the survival:aft accelerated failure time objective function is used for modeling, and based on the relationship between the clinical features learned by the model and the survival time distribution, the survival time of the patient is predicted.

[0068] Preferably, the Softmax algorithm is used to implement the dynamic weight allocation of the above features, that is, the attention mechanism automatically learns the weight of each feature in the current task by calculating the correlation or importance between features. The calculation formula is as follows:

[0069] Among them, Q and K are representations of features, and K T refers to the transpose of matrix K, and d is the dimension.

[0070] The assignment of weights depends on the overall distribution of the input data, such as the combined relationship of the above-mentioned features of a certain patient, etc.

[0071] In the code implementation, first, through the Xgboost extreme gradient boosting tree model, 7 features such as CAPRA-S score, age, clinical and pathological stage, seminal vesicle invasion, margin status, PSA level, and biochemical recurrence prediction score at the pathological level are integrated. The survival:aft accelerated failure time objective function is used for modeling, and parameter optimization is carried out through grid search. The parameter selection is based on cross-validation and ROC-AUC evaluation metrics, and its calculation formula is: .

[0072] Based on the relationship between features and survival time, a prediction model under the Xgboost framework is established. This model takes clinical features as input and outputs a recurrence risk score. The objective function of the Xgboost model is survival:aft, which is an Accelerated Failure Time (AFT) model specifically used for survival analysis. The AFT model directly models the survival time and assumes that covariates (i.e., clinical features) affect the survival time by accelerating or decelerating time. In this case, if BCR = 1: BCR_TimeMonths represents the actually observed biochemical recurrence time. If BCR = 0: BCR_TimeMonths represents that no biochemical recurrence is observed during the follow-up period, so this is regarded as a censored data, and BCR_TimeMonths corresponds to the censoring time. The model predicts the survival time of each patient by learning the relationship between clinical features and the survival time distribution. Due to the use of the AFT model, the model actually predicts the acceleration factor, which describes how clinical features affect the survival time. Therefore, the finally predicted risk score actually reflects the survival time of the patient, that is, the time when the patient has biochemical recurrence predicted by the model. The lower the score, the shorter the recurrence time of the patient and the higher the recurrence risk; conversely, the higher the risk score, the lower the recurrence risk of the patient.

[0073] The interpretability module 5 is an important part of the artificial intelligence-based multimodal prostate cancer biochemical recurrence risk stratification prediction system, including the pathological image interpretability sub-module 51, the fusion model interpretability sub-module 52, and the similar case comparison and analysis sub-module 53. Its core design is to improve the transparency and clinical credibility of the prediction results. This module consists of three subsystems that work together to provide in-depth interpretations from three aspects: pathological images, fusion models, and similar cases, enabling clinicians to understand the basis of the model's judgment from multiple dimensions of vision, quantitative analysis, and analogical reasoning, significantly enhancing the transparency and credibility of artificial intelligence-assisted decision-making, and providing stronger support for individualized follow-up and treatment plan formulation.

[0074] II. Model Training and Construction of the Artificial Intelligence-Based Multimodal Prostate Cancer Biochemical Recurrence Risk Stratification Prediction System

[0075] See the modeling flowchart in Figure 2 , and the specific process is as follows:

[0076] 1. Data collection and processing: Retrospectively collect the original medical record reports of patients after radical prostatectomy and the surgical pathology reports containing panoramic pathological section scan images with HE staining sections from multiple centers, and perform data cleaning and preprocessing.

[0077] This invention included a total of 2522 patients after radical prostatectomy from 4 independent centers. Among them, 849 cases in the training set were from Renji Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, and the validation set included 564 cases from the Chinese multi-center cohort, 341 cases from the TCGA cohort, and 768 cases from the PLCO cohort.

[0078] Inclusion criteria: (1) Patients pathologically confirmed to have prostate cancer; (2) Underwent radical prostatectomy; (3) Did not receive neoadjuvant treatment before surgery and did not receive adjuvant treatment within 6 months after surgery; (4) Had at least 1 HE-stained whole pathological section image; (5) Had complete follow-up data and biochemical recurrence outcome information.

[0079] Exclusion criteria: (1) Poor specimen preservation, affecting feature extraction; (2) Too small tumor area to extract at least 1 image patch under 10× magnification; (3) Section contamination, seriously affecting outcome interpretation.

[0080] Label the tumor areas of the 4368 diagnostic HE-stained sections included in the model in the QuPath software, and export the labeling information in GeoJSON format. After excluding the background areas, cut the original panoramic pathological section scan images into non-overlapping image patches of 256×256 pixels at three scales of 10×, 20×, and 40×, and extract the tumor area patches according to the labeling as digital pathological sections of prostate cancer for subsequent analysis.

[0081] 2. Clinical Feature Extraction and Processing

[0082] Based on the system - integrated large - language model API, multi - dimensional clinical data of patients are collected and processed, including key indicators that make up the CAPRA - S Score, such as patient age, preoperative PSA level, pathological Gleason score, clinicopathological stage, surgical margin status, seminal vesicle invasion, etc. During the extraction process, the accuracy of the extraction results is ensured based on the built - in data verification mechanism. For fields with ambiguity or uncertainty, the system will fluorescently mark the clinical features in the bolded table to prompt clinicians for review, so as to better achieve the intelligent and standardized integration of clinical and pathological data, laying a solid data foundation for subsequent artificial intelligence risk prediction.

[0083] Subsequently, the patient characteristics are fused with the training cohort. Through the Multiple Imputation by Chained Equations (MICE) algorithm, the missing values of the patient's key clinical variables are accurately filled, and a warning is sent when the number of missing values is not less than the preset number (e.g., > 4 missing values); subsequently, the filled data is scored by a risk stratification algorithm based on the CAPRA - S scoring system, and values are assigned to the PSA level, Gleason score, clinicopathological stage, surgical margin status, and lymph node status as follows: The PSA level scoring range is 0 - 4 points (corresponding to < 6, 6 - 10, 10 - 20, 20 - 30, ≥ 30 respectively), the Gleason score range is 0 - 3 points (corresponding to ISUP 1 - 5 levels, where ISUP 4 and 5 levels are 3 points), the clinicopathological stage is 0 - 1 point (0, T1 / T2 stage; 1, T3 / T4 stage), the surgical margin status is 0 - 2 points (0, R0 status; 1, R1 status; 2, R2 status), and lymph node involvement is 0 - 1 point (0, N0 status; 1, N1 status). Finally, a continuous risk assessment of 0 - 7 points is achieved, and patients are accurately stratified into three grades: low - risk (0 - 2 points), medium - risk (3 - 5 points), and high - risk (6 points and above).

[0084] 3. Pathological Feature Extraction and Processing Module

[0085] The self - supervised visual encoder UNI is used to automatically extract features from the above - mentioned image patches of 256×256 pixel size, generating a feature representation at the panoramic pathological section scan image level, with a size of {Ns×1024}, where N represents the number of image patches at this scale and s ∈ {10×, 20×, 40×}; subsequently, the pseudo - packet strategy is used for key feature extraction; then the feature packet representation is reconstructed into a four - dimensional tensor X reshaped : , and then convolutional kernels of different sizes (7×7, 5×5, 3×3) are used to process the feature data to further extract features with spatial perception and context information.

[0086] Complementary convolution kernels of different scales are used, and new feature maps with dense and rich context information are iteratively generated based on cross-attention step by step, enabling the model to simultaneously identify microscopic cell abnormalities and macroscopic tissue changes; finally, the feature information at three objective lens magnification scales is fused, the weights of key features are highlighted through the attention mechanism, and the biochemical recurrence probabilities obtained at different objective lens magnification multiples are weighted and averaged to finally generate the biochemical recurrence prediction score at the pathological level of the patient, and the recurrence or non-recurrence prediction result is obtained.

[0087] 4. Fusion of clinical features and pathological features to construct a multimodal biochemical recurrence risk prediction model

[0088] First, the CAPRA-S score, age, clinical pathological stage, seminal vesicle invasion, surgical margin status, preoperative PSA level, and biochemical recurrence prediction score at the pathological level are fused through the Xgboost model, and the weights of key features are highlighted through the attention mechanism; then the attention scores (W i × score) of each feature are calculated, and after weighted fusion of the features, high-risk, medium-risk, and low-risk three-level risk grades are obtained based on the calculation and analysis results; then the survival:aft accelerated failure time objective function is used for modeling, and based on the relationship between the clinical features learned by the model and the survival time distribution, the survival time of the patient is predicted.

[0089] In the present invention, both fusion operations use the Softmax algorithm to achieve the dynamic weight assignment of the above features, that is, the attention mechanism automatically learns the weight of each feature in the current task by calculating the correlation or importance between features, and the calculation formula is as follows:

[0090] Among them, Q and K are the representations of features, and K T refers to the transpose of matrix K, and d is the dimension.

[0091] The assignment of weights depends on the overall distribution of the input data, such as the combined relationship of the above features of a certain patient, etc.

[0092] 5. Verification

[0093] Evaluate the prediction performance of the model on multiple independent validation cohorts and compare it with traditional clinical prognosis evaluation models.

[0094] Use the C-index, the area under the receiver operating characteristic curve (AUC), and the KM survival prediction curve to evaluate the prediction performance of the model on the internal validation set of Renji Hospital, the Chinese multi-center validation cohort, the TCGA validation cohort, and the PLCO validation cohort.

[0095] Meanwhile, the PBRC-S model constructed in the present invention was compared with the traditional clinical prognosis assessment model CAPRA-S score. As shown in Table 1 below, in the training cohort, the C-index value of using the CAPRA-S score alone was 0.783, while the C-index of the fusion model of the present invention reached 0.834, higher than that of CAPRA-S. In addition, in different validation cohorts, the C-index of the fusion model was higher than the CAPRA-S score, indicating that integrating pathological image information and clinical indicators can significantly improve the accuracy of prognosis prediction.

[0096] The ROC results showed that in the four cohorts, the AUC area of the PBRC-S model constructed in the present invention was higher than that of the traditional CAPRA-S score and the single-scale AI model, and the prediction performance was better ( Figure 3 、 Figure 4 ).

[0097] The KM survival prediction curve analysis showed that the PBRC-S model could more accurately trisect the total population and different subgroup populations in each cohort into high-risk, medium-risk, and low-risk patients according to the recurrence risk ( Figure 5 ).

[0098] Table 1 Validation effects of the model in different cohorts

[0099]

[0100] 6. Model interpretability analysis

[0101] To improve the transparency and clinical credibility of the model prediction results, the present invention constructed a multi-level interpretability analysis module. First, for pathological images, the Grad-CAM technology was used to intelligently analyze pathological images, and multi-scale heatmaps were generated to accurately locate the pathological regions that significantly contributed to the prediction score. Second, for the fusion model, the SHAP value calculation method was used to quantitatively analyze the contribution degrees of all input variables, and waterfall plots and force plots were generated to intuitively display the positive and negative influence degrees of each clinical indicator (such as PSA, Gleason score, surgical margin status, clinical pathological stage, etc.) and pathological features on the final risk prediction result. At the same time, a built-in contribution degree verification mechanism was used to ensure the accuracy of the SHAP value calculation results. In addition, based on KNN, patient cases with the most similar feature expressions and prediction scores were automatically retrieved from the internal training set, and their key clinical baseline indicators, treatment plans, and actual biochemical recurrence results were presented in tabular form to form a risk reference spectrum based on real cases.

[0102] For example, for the 561st patient in the test set, the model predicted that it had a high risk of biochemical recurrence. The heatmap of the pathological image found that the model might focus on high-grade features and inflammatory cell regions, which are closely related to the invasiveness of tumors ( Figure 6A). Combining clinical information, we further used SHAP values to analyze the impact of each feature on the prediction results. The results showed that the score of the pathological prediction module was the most critical risk factor, followed by CAPRA_S_score, psa_value, age_at_diagnosis, and pathologic_T_stage. Higher values of these features all increased the risk of biochemical recurrence ( Figure 6 B). Overall, factors such as pathologic_T_stage = 2, age_at_diagnosis = 70, the score of the pathological prediction module = 0.309, and psa_value = 32.9 acted together to push the patient's predicted risk score from the baseline value of 1.9 to 3.8. CAPRA_S_score = 10 largely reduced the patient's recurrence time and increased the recurrence risk ( Figure 6 C). Finally, the model predicted that the patient had a high risk of biochemical recurrence.

[0103] This interpretable model used the KNN algorithm to find 30 cases most similar to the patient's characteristics. The average age of these cases was 69.43 years, the median PSA value was 20.75 ng / mL, the average Gleason score was 9.03, the median CAPRA-S score was 8.00, the median score of the pathological prediction module was 0.27, and the median BCR_TimeMonths was 24.50. Compared with this patient, the PSA values, CAPRA-S scores, and scores of the pathological prediction module of these similar cases were lower, and their average recurrence time was also shorter than that of this patient. Considering that the PSA value and the score of the pathological prediction module of this patient were both higher than those of the similar cases, and the CAPRA_S_score was higher, it indicated that the recurrence risk might still be relatively high. For specific information, see Table 2 below:

[0104] Table 2 Comparison of the characteristics of the test patient and 30 similar patients selected by the KNN algorithm in the training set

[0105]

[0106] In summary, compared with other methods, this method established a multimodal prediction model for the risk stratification of biochemical recurrence of prostate cancer. By analyzing whole pathological sections based on the Xgboost framework through end-to-end, multi-scale, multi-center, and large-sample methods, it could maximize the use of pathological information. Combining the CAPRA-S score could more comprehensively evaluate the prognostic risk of patients and had a complete long-term follow-up process to provide reliable clinical facts, showing obvious advantages over traditional models. This method could assist doctors in the precise stratification and personalized follow-up of prostate cancer, was expected to reduce the risks of over-treatment and missed diagnosis, and had good application prospects.

[0107] The basic principles, main features and significant advantages of the present invention have been elaborated in detail above. The purpose of the embodiments and the descriptions in the specification is only to deeply explain the operating principle of the present invention. Without departing from the core idea and the legally protected scope of the present invention, the present invention allows and covers various reasonable changes and improvements. These changes and improvements will all be regarded as falling within the scope of protection claimed by the present invention. Specifically, the scope of protection of the present invention will be clearly and strictly defined by the appended claims and their equivalents with the same legal effect.

Claims

1. A multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence, characterized in that, Including: An input display module, which is used to input the original medical record report of the patient and the surgical pathology report containing the panoramic scan image of the pathological section, and display the analysis result after the analysis is completed; A clinical feature extraction and processing module, which automatically extracts the clinical feature information of the patient from the original medical record report and the surgical pathology report, automatically determines and fills in the missing values, and then performs risk stratification algorithm scoring based on the CAPRA-S scoring system for the filled information to form a standardized clinical feature vector; A pathological feature extraction and processing module. After the image is preprocessed, a multi-instance algorithm based on the pseudo-packet strategy and cyclic cross-attention is used to first extract the deep features of tumor images with different objective lens magnifications, and then generate a feature representation at the panoramic pathological section scan image level in a weakly supervised network, extract the microscopic cell abnormality and macroscopic tissue change features, obtain the pathological feature representation, and fuse and generate a biochemical recurrence prediction score at the pathological level; A fusion prediction module, which performs multi-modal feature fusion on the pathological feature representation and the standardized clinical feature vector information to construct an end-to-end risk prediction model for biochemical recurrence of prostate cancer; highlights the weights of key features through the attention mechanism, analyzes the pathological feature representation and the standardized clinical feature vector information, classifies the patient into three risk levels of high risk, medium risk, and low risk, and visualizes the key prediction factors; An interpretability module, which provides in-depth interpretations from three aspects of pathological images, fusion models, and similar cases, quantitatively displays the contribution degrees of each clinical index and pathological feature to the prediction result, and provides a basis for understanding the model's judgment from multiple dimensions of vision, quantitative analysis, and analogical reasoning.

2. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 1, characterized in that, It also includes a storage module, which is used to store the information processed by each module.

3. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 1, characterized in that, The clinical feature extraction and processing module includes a clinical feature extraction sub-module and a clinical data processing sub-module. The clinical feature extraction sub-module is a large language model API integrated in the system, and at least extracts the following clinical feature information: age, preoperative PSA level, Gleason score, clinical pathological stage, surgical margin status, lymph node status, seminal vesicle invasion; during the extraction process, based on the built-in data verification mechanism, the fields with ambiguity or uncertainty are marked to prompt the clinician to review. The clinical data processing sub-module first fuses the patient characteristics and the training cohort, accurately fills in the missing values of the key clinical variables of the patient through the MICE algorithm, and sends a warning when the number of missing values is not less than the preset number; then performs risk stratification algorithm scoring based on the CAPRA-S scoring system for the filled data, assigns values to the PSA level, Gleason score, clinical pathological stage, surgical margin status, and lymph node status, and accurately stratifies the patient into three levels of low risk, medium risk, and high risk based on the comprehensive scoring result.

4. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 1, wherein The pathological feature extraction and processing module includes a feature vector processing sub-module, a cyclic cross-attention sub-module, and a multi-scale feature fusion sub-module. The feature vector processing sub-module first preprocesses the panoramic pathological section scan image, then crops the tumor region into non-overlapping image patches of 256×256 pixels at the magnification of 10×, 20×, and 40× objective lenses, and automatically extracts features from them using the self-supervised vision encoder UNI to generate a feature representation at the panoramic pathological section scan image level containing low-dimensional feature sets under three different magnification levels; subsequently, the features are refined based on the pseudo-packet strategy, and four-dimensional reconstruction and convolution processing of different sizes are performed on the features. The cyclic cross-attention sub-module complements convolutional kernels of different scales and gradually generates new feature maps with dense and rich context information based on cross-attention iteration, enabling the model to simultaneously identify microscopic cell abnormalities and macroscopic tissue changes. The multi-scale feature fusion sub-module fuses the feature information at three objective lens magnification scales, and performs weighted averaging on the biochemical recurrence probabilities obtained at different objective lens magnifications to obtain the recurrence or non-recurrence prediction result.

5. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 4, wherein The preprocessing method of the panoramic pathological section scan image is as follows: by adjusting the mean and standard deviation of the image patch data, the data is converted into a distribution with zero mean and unit variance to achieve data standardization and image intensity normalization. After the last convolutional layer of UNI, 1024-dimensional feature vectors of each image patch are extracted respectively to generate a feature representation at the panoramic pathological section scan image level. Each panoramic pathological section scan image will generate a set of low-dimensional feature matrices at three magnification scales, with a size of {Ns×1024}, where N represents the number of image patches at this scale, and s∈{10×, 20×, 40×}. Each convolutional kernel size corresponds to a different spatial range, which are 7×7, 5×5, and 3×3 respectively.

6. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 1, wherein, The fusion prediction module first fuses the CAPRA-S score, age, clinical pathological stage, seminal vesicle invasion, surgical margin status, preoperative PSA level, and biochemical recurrence prediction score at the pathological level through the Xgboost model, and highlights the weights of key features through the attention mechanism. Then, the attention scores of each feature are calculated, and after weighted fusion of the features, high-risk, medium-risk, and low-risk three-level risk grades are obtained based on the calculation and analysis results; then, the survival:aft accelerated failure time objective function is used for modeling, and based on the relationship between the clinical features learned by the model and the survival time distribution, the survival time of the patient is predicted.

7. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 6, wherein The interpretability module includes a pathological image interpretability sub-module, a fusion model interpretability sub-module, and a similar case comparison and analysis sub-module. The pathological image interpretability sub-module intelligently analyzes the pathological image based on the Grad-CAM technology, accurately locates the pathological regions that contribute significantly to the prediction score through multi-scale heatmaps, and presents important morphological features at different magnifications. During the analysis process, the built-in visualization calibration mechanism is used to ensure that the extraction results are basically consistent with the judgment of the pathologist. For pathological regions with ambiguity or uncertainty, the system marks the important regions in the heatmap and prompts the pathologist to review. The interpretable sub-module of the fusion model uses the SHAP value calculation method to quantitatively analyze the contribution of all input variables, and intuitively shows the positive and negative influence degrees of each clinical index and pathological feature on the final risk prediction result; during the analysis process, for the indexes with abnormal contribution or inconsistent with clinical experience, the system marks them and prompts the clinician to conduct a review. The similar case comparison and analysis sub-module automatically retrieves 30 patient cases with the most similar feature expressions and prediction scores from the internal training set, and presents their key clinical baseline indicators, treatment plans, and actual biochemical recurrence results in tabular form, forming a risk reference spectrum based on real cases. During the analysis process, for the features with significant differences between the similar cases and the target patient, the system marks them and prompts the clinician to conduct a review.

8. The multi-modal prostate cancer biochemical recurrence risk stratification prediction system based on artificial intelligence according to claim 7, wherein The key pathological features mainly extracted by the interpretable sub-module of pathological images include glandular structure, nuclear atypia, and cell infiltration. The contribution degree of the key clinical indicators mainly concerned by the interpretable sub-module of the fusion model covers the core variables constituting the CAPRA-S Score. The key clinical baseline indicators mainly referred to by the similar case comparison and analysis sub-module include age, preoperative PSA level, Gleason score, clinical pathological stage, treatment plan, follow-up time, and biochemical recurrence time.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it realizes the functions of the system according to any one of claims 1 to 8.

Images