Prostate cancer endocrine therapy drug resistance prediction system based on multi-model artificial intelligence

Abstract

The invention discloses a prostate cancer endocrine therapy drug resistance prediction system based on multi-model artificial intelligence. The prostate cancer endocrine therapy drug resistance prediction system comprises a data acquisition and preprocessing module, a single omics prediction module, a multi-modal feature integration module, a model training module and a prediction result output module. Multi-omics data of a patient is collected and preprocessed, data representations of omics data of different labels are generated by using priori knowledge, and an optimal single-omics prediction model is matched for the omics data of different labels based on the data representations after knowledge enhancement. And constructing a multi-omics prediction model of the prostatic cancer endocrine treatment drug resistance by using the optimal single-omics prediction model corresponding to the different label medical data by adopting Stacking, predicting the drug resistance risk of the patient within a preset time, outputting an early warning, and adjusting the treatment scheme of the patient according to the drug resistance risk. According to the method, the multi-omics data are integrated, the accurate multi-omics prediction model for the drug resistance of the endocrine therapy of the prostate cancer is constructed, and the accuracy and clinical applicability of drug resistance prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of drug resistance prediction, and more specifically, to a prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence. Background Art

[0002] Prostate cancer is a common malignancy in men. Endocrine therapy (such as androgen deprivation therapy (ADT)) is the mainstay of treatment for advanced prostate cancer. However, most patients eventually develop castration-resistant prostate cancer (CRPC), which is characterized by resistance to endocrine therapy. The mechanisms of resistance are complex, involving the interplay of multiple factors, including genomic variation, epigenetic regulation, and alterations in the tumor microenvironment. This limits the accuracy and generalizability of traditional single-model predictions.

[0003] In recent years, the application of multi-model artificial intelligence (AI) technology in the medical field has provided new insights for predicting drug resistance in prostate cancer. By integrating multimodal data such as genomics, transcriptomics, proteomics, clinicopathological features, and imaging, multi-model AI can more comprehensively capture biomarkers and molecular pathways associated with drug resistance. For example, deep learning-based convolutional neural networks (CNNs) can analyze pathological image features, graph neural networks (GNNs) can model gene interaction networks, and ensemble learning methods (such as random forests and XGBoost) can fuse heterogeneous data to improve predictive robustness. Furthermore, the introduction of transfer learning and federated learning can help address issues such as insufficient sample size and privacy protection in medical data. Current challenges include the high-dimensional heterogeneity of multimodal data, insufficient model interpretability, and the cross-center validation required for clinical translation. Therefore, how to use multi-model AI to improve the accuracy of predicting drug resistance in prostate cancer endocrine therapy is an urgent issue that needs to be addressed. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence, which integrates multi-omics data (genome, transcriptome, proteome, clinical pathological characteristics and imaging data) to construct an accurate multi-omics prediction model for prostate cancer endocrine therapy resistance, thereby improving the accuracy and clinical applicability of resistance prediction.

[0005] The present invention provides a prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence, including a data acquisition and preprocessing module, a single-omics prediction module, a multimodal feature integration module, a model training module and a prediction result output module; The data acquisition and preprocessing module is responsible for collecting the multi-omics data of the patient and preprocessing the collected multi-omics data; The single-omics prediction module uses prior knowledge to generate data representations of omics data with different labels, matches the omics data with different labels with the best single-omics prediction model based on the knowledge-enhanced data representation, and trains the best single-omics prediction model corresponding to the medical data with different labels; The multimodal feature integration module uses stacking to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels, and predicts the risk of drug resistance in patients within a preset time. The model training module obtains training data, uses the data-balanced training data to train the best single-omics prediction model, and introduces the concept of federated learning to update the multi-omics prediction model for prostate cancer endocrine therapy resistance; The prediction result output module outputs an early warning of the patient's drug resistance risk within a preset time and adjusts the patient's treatment plan based on the drug resistance risk.

[0006] In this solution, the data acquisition and preprocessing module is specifically: Obtain medical records based on the patient's identity information, cleanse and standardize the data, build a medical data extraction model based on the BERT model combined with the Bi-LSTM network, obtain relevant medical data through preset data category labels, and perform span detection and category classification training on the medical data extraction model; Import the preprocessed medical record data into the trained medical data extraction model, use the BERT model for word segmentation to generate text vectors, encode the text vectors using the Bi-LSTM network, detect entity words in the medical record data, and use continuous entity words as candidate spans; An attention layer is introduced into the Bi-LSTM network to calculate the similarity between the text vector sequence and the key value in the candidate span to obtain an attention weight, and the vector is weighted using the attention weight. The weighted result is used as a feature vector, and Softmax is used for classification to extract data with a preset data category label from the medical record data; Corresponding medical data subsets are constructed according to preset data category labels to generate multi-omics data of patients, and different data preprocessing is performed on different medical data subsets.

[0007] In this solution, in the single-omics prediction module, prior knowledge is used to generate data representations of omics data with different labels, specifically: Obtain multi-omics data from patients, construct a temporal encoder to encode medical data subsets with different data category labels, obtain the temporal dependency of different data, and generate hidden layer representations of different data in the multi-omics data; Using similarity calculation based on the latent layer representations of different data corresponding to the patient, obtain a prostate cancer endocrine therapy resistance record instance whose similarity meets preset requirements, and retrieve a patient group similar to the patient from the prostate cancer endocrine therapy resistance record instance to construct a context learning sample; The spatial distribution of the latent layer representations and contextual learning examples corresponding to different patient data is aligned, and then imported into a large medical model to generate personalized medical knowledge for each data category label in the task of predicting endocrine therapy resistance. The latent representations of different data and personalized medical knowledge are encoded into embedding vectors, and the knowledge-enhanced data representations of omics data with different labels are output.

[0008] In this solution, based on the knowledge-enhanced data representation, the best single-omics prediction model is matched to the omics data with different labels, and the best single-omics prediction model corresponding to the medical data with different labels is trained. Specifically: Obtain prediction model categories associated with different data category labels in the task of predicting prostate cancer endocrine therapy resistance based on the knowledge-enhanced data representation of patient multi-omics data; Using different data category labels, prediction model category labels, and associations to construct a relationship graph, using Metapath random walk to pre-train the relationship graph, and generate metapaths containing different data category labels as head nodes; A graph attention network is introduced to represent meta-paths with different data category labels as head nodes. The predicted model category nodes in the meta-path are used as neighbor nodes of the head node. The node feature vectors are obtained through graph convolution, and the attention weights of the neighbor nodes on the head node in different meta-paths are calculated. The neighbor nodes in each meta-path are sorted by the attention weight, the best single-omics prediction model in each meta-path is determined according to the sorting result, and the best single-omics prediction model for training sample training matching is constructed by using medical data subsets with different data category labels.

[0009] In this solution, in the multimodal feature integration module, stacking is used to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels. Specifically: Obtain training samples with different data category labels to construct a base training dataset, train each best single-omics prediction model, use a 5-fold cross-validation strategy, splice the prediction probabilities output by each best single-omics prediction model into a complete meta-feature matrix, and use the meta-feature matrix as a new training dataset to train the meta-model based on the SVM network; During the meta-selector training process, the meta-model hyperparameters are adjusted through grid search, and each optimal single-omics prediction model is retrained using the entire training dataset. The contribution of each optimal single-omics prediction model is calculated using a non-negative least squares algorithm to generate meta-model weight information. The meta-model prediction results are combined with the weight information to generate the final prediction results of endocrine therapy resistance in prostate cancer; An independent test dataset is used to validate the model. When the final prediction results of prostate cancer endocrine therapy resistance meet the preset test criteria, a multi-omics prediction model for prostate cancer endocrine therapy resistance is output; Based on the patient's medical condition information, the patient's medication data within the preset time is determined, and the data representation of the patient's multi-omics data after knowledge enhancement is used as the model input to predict the patient's probability of drug resistance to the medication data within the preset time, and output the patient's drug resistance risk.

[0010] In this solution, in the model training module, training data is obtained and the optimal single-omics prediction model is trained using the balanced training data. Specifically: Extract features from prostate cancer endocrine therapy resistance records corresponding to different data category labels, construct resistance training data and non-resistance training data based on the true labels in the examples, and define the majority class training sample set and minority class training sample set based on the number of samples in the resistance training data and non-resistance training data; Select a training sample from the minority class training samples as a root sample, obtain K training samples closest to the root sample to generate a neighbor sample set, and select any one neighbor sample from the neighbor sample set as an auxiliary sample for sample synthesis; New training samples are synthesized by random interpolation between the root sample and the auxiliary sample, the new training samples are added to a new training sample set, the new training sample set is used to expand the minority class training samples, and the training data after data balance is used to train the best single-omics prediction model.

[0011] In this solution, federated learning is introduced into the model training module to update the multi-omics prediction model for prostate cancer endocrine therapy resistance. Specifically: The multi-omics prediction model for prostate cancer endocrine therapy resistance was used as the initial global model. Training data from hospital nodes in different regions was extracted and local data was standardized. The federated parameters were initialized and training and updating were performed using a federated learning algorithm. In federated training, the central server sends the parameters of the initial global model to hospital nodes in different regions to build local models. Each hospital node uses local data to update the local model. The hospital node uploads the trained local model, performs weighted aggregation in the central server, and retrains the aggregated and updated global model using the balanced training data. The updated global model is re-issued, and the local model parameters are updated. After each hospital node performs drug resistance prediction, federal training is carried out using the drug resistance prediction results and confidence guidance. The global model is verified on the central server. After verification, the multi-omics prediction model of prostate cancer endocrine therapy resistance is updated.

[0012] A second aspect of the present invention provides a method for predicting resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence, comprising the following steps: Collect multi-omics data of patients and pre-process the collected multi-omics data; Use prior knowledge to generate data representations for omics data with different labels. Based on the knowledge-enhanced data representations, match the best single-omics prediction model for omics data with different labels, and train the best single-omics prediction model corresponding to medical data with different labels. Stacking is used to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels. Federated learning is introduced to train the model and predict the risk of drug resistance in patients within a preset time. Output an early warning of the patient's drug resistance risk within a preset time, and adjust the patient's treatment plan based on the drug resistance risk.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This study leverages multi-model artificial intelligence technology to integrate multi-omics data (genomic, transcriptomic, proteomic, clinicopathological, and imaging data) to construct a precise prediction model for prostate cancer resistance to endocrine therapy. The system employs a modular design encompassing data preprocessing, feature extraction, multi-model fusion, predictive analysis, and visualization output to enhance the accuracy and clinical applicability of resistance prediction.

[0014] This invention integrates multi-dimensional data such as genomic mutations, transcriptome expression, and clinical indicators to overcome the limitations of a single data source. The combination of base models such as XGBoost and neural networks and meta-models can effectively identify complex biomarker interactions and improve robustness. The system also allows for future access to multi-center data (each hospital trains the base model locally and only shares meta-features), meeting data privacy requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or exemplary descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to these drawings without paying any creative work.

[0016] Figure 1 A block diagram of a multi-model artificial intelligence-based prostate cancer endocrine therapy resistance prediction system is shown.

[0017] Figure 2 A flowchart showing the matching of the best single-omics prediction model for omics data with different labels; Figure 3 A flowchart showing the construction of a multi-omics prediction model for endocrine therapy resistance in prostate cancer is shown; Figure 4 A flowchart of a multi-model artificial intelligence-based method for predicting resistance to endocrine therapy in prostate cancer is shown; DETAILED DESCRIPTION In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0018] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0019] Figure 1 A block diagram of a multi-model artificial intelligence-based prostate cancer endocrine therapy resistance prediction system is shown.

[0020] like Figure 1 As shown, the first embodiment of the present invention provides a prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence, including: a data acquisition and preprocessing module 101, a single-omics prediction module 102, a multimodal feature integration module 103, a model training module 104 and a prediction result output module 105; The data acquisition and preprocessing module is responsible for collecting the multi-omics data of the patient and preprocessing the collected multi-omics data; The single-omics prediction module uses prior knowledge to generate data representations of omics data with different labels, matches the omics data with different labels with the best single-omics prediction model based on the knowledge-enhanced data representation, and trains the best single-omics prediction model corresponding to the medical data with different labels; The multimodal feature integration module uses stacking to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels, and predicts the risk of drug resistance in patients within a preset time. The model training module obtains training data, uses the data-balanced training data to train the best single-omics prediction model, and introduces the concept of federated learning to update the multi-omics prediction model for prostate cancer endocrine therapy resistance; The prediction result output module outputs an early warning of the patient's drug resistance risk within a preset time and adjusts the patient's treatment plan based on the drug resistance risk.

[0021] It should be noted that medical record data is obtained based on patient identity information, cleansed, and standardized. A medical data extraction model is constructed based on the BERT model combined with a Bi-LSTM network. Relevant medical data is obtained using pre-set data category labels, and the model is trained for span detection and category classification. These pre-set data category labels include clinical data (patient age, PSA level, Gleason score, treatment regimen, and drug resistance (CRPC progression time)), genomic data (resistance-related variants such as AR mutations, PTEN deletions, and TP53 mutations), transcriptomic data (differentially expressed genes such as NKX3-1 and ERG, and non-coding RNAs (lncRNAs and miRNAs)), proteomic data (AR-V7, PSMA, and PTEN protein expression levels), and imaging data (tumor volume, ADC value, and metabolic profiles). Medical data with pre-set data category labels serves as the source domain dataset for the extraction model, which is trained in the span detection and category classification stages. Contrastive learning is used to improve the accuracy and stability of the extraction model.

[0022] In the target domain, the preprocessed medical record data is imported into the trained medical data extraction model. The BERT model is used for word segmentation to generate text vectors. These text vectors are then encoded using a Bi-LSTM network to detect entity words in the medical record data, using continuous entity words as candidate spans. The BERT model is used to extract word vectors from the preprocessed medical record data, fully preserving the semantic information of the case data and improving the ability to extract bidirectional contextual features. The Bi-LSTM network, composed of a forward LSTM and a backward LSTM, is trained from two directions, enabling it to memorize both past and future information, addressing the difficulty of capturing contextual information. The addition of an attention mechanism to the Bi-LSTM network distinguishes the importance of different features, ignoring unimportant ones and focusing on important ones, thereby improving classification accuracy. An attention layer is introduced into the Bi-LSTM network. The similarity between the text vector sequence and the key value in the candidate span is calculated to obtain attention weights. The vectors are weighted using the attention weights, and the weighted results are used as feature vectors for classification using Softmax. Data with preset data category labels are extracted from the medical record data. Corresponding medical data subsets are constructed based on the preset data category labels to generate multi-omics data for patients. Different data preprocessing is performed on different medical data subsets. For example, variation detection (GATK, STAR) is performed on genome and transcriptome data to obtain variation annotation (ANNOVAR), and key genes (|logFC|>1, FDR<0.05) are screened through differential expression analysis (DESeq2 / edgeR). Image data is preprocessed through image registration, ROI segmentation, and feature extraction.

[0023] It should be noted that in the single-omics prediction module, the multi-omics data of the patient is obtained, and a temporal encoder is constructed to encode the medical data subsets with different data category labels to obtain the time dependency of different data, and generate the hidden layer representation of different data in the multi-omics data. Preferably, HiTANet is used as the encoding model for multi-omics data. Based on the latent representations of different patient data, similarity calculations are performed to obtain prostate cancer endocrine therapy resistance records that meet preset similarity requirements. From these prostate cancer endocrine therapy resistance records, a patient group similar to the patient is retrieved to construct contextual learning examples. The large-scale model is guided to learn underlying patterns based on the relevant information of similar patients. The prediction model used by medical data with different data category labels is then used to spatially align the latent representations of the different patient data and the contextual learning examples. This is then imported into large-scale medical models (such as BioGPT, Med-PaLM, and ClinicalBERT) to generate personalized medical knowledge for the endocrine therapy resistance prediction task for each data category label. By leveraging the rich medical knowledge within the large-scale model, the distribution differences between the two data types are aligned and the global information of the two data types is combined to generate valuable personalized medical knowledge for the prostate cancer endocrine therapy resistance prediction task. The latent representations of the different data types and the personalized medical knowledge are encoded into embedding vectors, and the output is a knowledge-enhanced data representation of the omics data with different labels. For example, the input may include EMR text, AR mutation detection report, and MRI image of patient A. After enhancing the knowledge of large models in the medical field, the output is: text embedding: prompting "rapid increase in PSA + bone metastasis" → high risk of drug resistance; gene embedding: AR mutation is mapped to "androgen receptor signaling pathway activation"; image embedding: PI-RADS 5 points → highly aggressive tumor.

[0024] Figure 2 A flowchart showing the matching of the best single-omics prediction model for omics data with different labels.

[0025] According to an embodiment of the present invention, based on the knowledge-enhanced data representation, the best single-omics prediction model is matched to the omics data with different labels, and the best single-omics prediction model corresponding to the medical data with different labels is trained, specifically: S202, obtaining prediction model categories associated with different data category labels in the task of predicting prostate cancer endocrine therapy resistance based on the data representation of the patient's multi-omics data after knowledge enhancement; S204, constructing a relationship graph using different data category labels, prediction model category labels, and associations, pre-training the relationship graph using Metapath random walk, and generating metapaths including head nodes with different data category labels; S206: Introduce a graph attention network to represent meta-paths with different data category labels as head nodes, use the prediction model category nodes in the meta-paths as neighbor nodes of the head nodes, obtain node feature vectors through graph convolution, and calculate the attention weights of the neighbor nodes in different meta-paths to the head node; S208, sorting the neighbor nodes in each meta-path by the attention weight, determining the best single-omics prediction model in each meta-path according to the sorting result, and constructing the best single-omics prediction model for training sample training matching by using medical data subsets with different data category labels.

[0026] It should be noted that Metapath2vec++ pre-training is used to represent the relationship graph in high dimensions, define path parameters, generate walk sequences, and construct corresponding meta-paths. A graph attention network is used to represent the meta-paths, and graph convolution is used to learn the feature information of neighbor entities. The weights of neighbor entities in each meta-path are obtained through the attention mechanism to characterize the importance of neighbor entities to the head entity. The neighbor node with the largest weight in each meta-path is selected as the optimal single-omics prediction model. For example, for genomic / transcriptomic data, a gene interaction network (STRING database) is constructed, and GraphSAGE is used to learn key pathways for prediction. For clinical data, XGBoost or LightGBM is used for prediction, and SHAP value analysis is used to screen important features and predict drug resistance risk scores. For imaging data, 3D CNN + Transformer is used for drug resistance prediction, and 3D ResNet is used to extract spatial features, and ViT (Vision Transformer) is used to capture long-range dependencies.

[0027] Figure 3 A flowchart showing the construction of a multi-omics prediction model for endocrine therapy resistance in prostate cancer.

[0028] According to an embodiment of the present invention, in the multimodal feature integration module, stacking is used to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels. Specifically, S302, obtaining training samples with different data category labels to construct a base training dataset, training each best single-omics prediction model, using a 5-fold cross-validation strategy, splicing the prediction probabilities output by each best single-omics prediction model into a complete meta-feature matrix, and using the meta-feature matrix as a new training dataset to train a meta-model based on the SVM network; S304, adjusting meta-model hyperparameters through grid search during the meta-selector training process, retraining each optimal single-omics prediction model using the entire training data set, calculating the contribution of each optimal single-omics prediction model using a non-negative least squares algorithm, generating meta-model weight information, and combining the meta-model prediction results with the weight information to generate a final prostate cancer endocrine therapy resistance prediction result; S306, using an independent test data set to perform model validation, and when the final prediction result of prostate cancer endocrine therapy resistance meets the preset test criteria, outputting a multi-omics prediction model for prostate cancer endocrine therapy resistance; S308, determining the patient's medication data within a preset time based on the patient's condition information, using the patient's multi-omics data representation after knowledge enhancement as a model input, predicting the patient's drug resistance probability to the medication data within the preset time, and outputting the patient's drug resistance risk.

[0029] It should be noted that Stacking consists of two layers: a base model and a meta-model. The base model trains the best single-omics prediction model for different omics data. The meta-model trains the final classifier based on the output of the base models. A 5-fold cross-validation is applied to each base model, and the cross-validated prediction probabilities of the base models are concatenated into a complete meta-feature matrix. This meta-feature matrix is ​​used as the new training set, retaining the original true labels. The SVM network is selected as the meta-model, and meta-model hyperparameters (such as learning rate and tree depth) are optimized through grid search. The best single-omics predictions are then retrained using the entire training data. Stacking is used to construct a multi-omics prediction model for endocrine therapy resistance in prostate cancer by combining the best single-omics prediction models corresponding to medical data with different labels. This model integrates multi-dimensional information from genomic mutations, gene expression, and clinical indicators. When the quality of a particular omics data is poor, other omics can provide compensation. The contribution weight of each omics is output to assist physicians in understanding the basis for the prediction and supports the flexible addition of other modalities such as imaging omics and proteomics.

[0030] It should be noted that in the model training module, features are extracted from prostate cancer endocrine therapy resistance record instances corresponding to different data category labels, and resistance training data and non-resistance training data are constructed according to the real labels in the instances. The data are divided into training set and independent test set at a ratio of 7:3. In order to ensure that the ratio of resistance / non-resistance samples in the two groups is consistent, the majority class training sample set and the minority class training sample set are defined by the number of samples of resistance training data and non-resistance training data; a training sample is selected as the root sample in the minority class training sample, and the K training samples closest to the root sample are obtained to generate a neighbor sample set, and one neighbor sample is selected from the neighbor sample set as an auxiliary sample for sample synthesis; a new training sample is synthesized by random difference between the root sample and the auxiliary sample, and the new training sample is added to the new training sample set. The minority class training sample is expanded using the new training sample set, and the training data after data balance is used to train the best single-omics prediction model.

[0031] In the model training module, federated learning was introduced to update a multi-omics prediction model for prostate cancer endocrine therapy resistance. This model was used as the initial global model. Training data from hospital nodes in different regions was extracted and standardized locally. A data isolation zone was established, and all computations were performed in a secure local environment. Initializing federated parameters, including the aggregation algorithm (FedAvg), communication rounds (T), and the minimum number of participating nodes (K ≥ 3), allows for training and updating using the federated learning algorithm. During federated training, the central server sends the initial global model parameters to hospital nodes in different regions to construct local models. Each hospital node then updates the local model using local data. The trained local models are then uploaded to the central server for weighted aggregation. The updated global model is then retrained using the balanced training data, alleviating data sparsity to a certain extent. Data augmentation also protects patients' real-world data from being leaked. The updated global model is then redistributed, and local model parameters are updated. After each hospital node performs drug resistance prediction, federated training is conducted using the drug resistance prediction results and confidence levels. The global model is validated on the central server, and upon successful validation, the multi-omics prediction model for prostate cancer endocrine therapy resistance is updated. The dynamic update mechanism for the multi-omics prediction model uses model distillation (DistilFL) for rapid adaptation to accommodate new hospital additions. Federated training is automatically triggered monthly for model iteration.

[0032] It should be noted that a PDF report containing risk scores, decision-making basis and literature support is automatically generated in the prediction result output module (for example: the patient's PTEN deficiency + high Gleason score leads to an 87% probability of resistance to a certain drug (95% CI: 82-91%)). The system predicts the patient's risk of drug resistance within 1-2 years, adjusts the treatment plan in advance, and realizes visual early warning.

[0033] Figure 4 A flowchart of the method for predicting resistance to endocrine therapy in prostate cancer based on multi-model artificial intelligence is shown.

[0034] A second embodiment of the present invention provides a method for predicting resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence, comprising the following steps: S402, collecting multi-omics data of the patient and preprocessing the collected multi-omics data; S404, using prior knowledge to generate data representations of omics data with different labels, matching the omics data with different labels with the best single-omics prediction model based on the knowledge-enhanced data representation, and training the best single-omics prediction model corresponding to the medical data with different labels; S406 uses stacking to combine the best single-omics prediction models corresponding to medical data with different labels to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance. This model is trained using federated learning to predict the risk of drug resistance in patients within a preset timeframe. S408: Outputting a warning of the patient's drug resistance risk within a preset time, and adjusting the patient's treatment plan according to the drug resistance risk.

[0035] The third embodiment of the present invention provides a computer-readable storage medium, which includes a method program for predicting resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence. When the method program for predicting resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence is executed by a processor, it implements the steps of the method for predicting resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence.

[0036] In the several embodiments provided in this application, it should be understood that the disclosed methods can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0037] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0038] Alternatively, if the integrated modules of the present invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

[0039] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A multi-model artificial intelligence-based prostate cancer endocrine therapy resistance prediction system, characterized by: It includes data acquisition and preprocessing module, single-omics prediction module, multimodal feature integration module, model training module and prediction result output module; The data acquisition and preprocessing module is responsible for collecting the multi-omics data of the patient and preprocessing the collected multi-omics data; The single-omics prediction module uses prior knowledge to generate data representations of omics data with different labels, matches the omics data with different labels with the best single-omics prediction model based on the knowledge-enhanced data representation, and trains the best single-omics prediction model corresponding to the medical data with different labels; The multimodal feature integration module uses stacking to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels, and predicts the risk of drug resistance in patients within a preset time. The model training module obtains training data, uses the data-balanced training data to train the best single-omics prediction model, and introduces the concept of federated learning to update the multi-omics prediction model for prostate cancer endocrine therapy resistance; The prediction result output module outputs an early warning of the patient's drug resistance risk within a preset time and adjusts the patient's treatment plan based on the drug resistance risk.

2. The multi-model artificial intelligence-based prostate cancer endocrine therapy resistance prediction system according to claim 1, characterized in that: The data acquisition and preprocessing module is specifically: Obtain medical records based on the patient's identity information, cleanse and standardize the data, build a medical data extraction model based on the BERT model combined with the Bi-LSTM network, obtain relevant medical data through preset data category labels, and perform span detection and category classification training on the medical data extraction model; Import the preprocessed medical record data into the trained medical data extraction model, use the BERT model for word segmentation to generate text vectors, encode the text vectors using the Bi-LSTM network, detect entity words in the medical record data, and use continuous entity words as candidate spans; An attention layer is introduced into the Bi-LSTM network to calculate the similarity between the text vector sequence and the key value in the candidate span to obtain an attention weight, and the vector is weighted using the attention weight. The weighted result is used as a feature vector, and Softmax is used for classification to extract data with a preset data category label from the medical record data; Corresponding medical data subsets are constructed according to preset data category labels to generate multi-omics data of patients, and different data preprocessing is performed on different medical data subsets.

3. The prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence according to claim 1, characterized in that: In the single-omics prediction module, prior knowledge is used to generate data representations of omics data with different labels, specifically: Obtain multi-omics data from patients, construct a temporal encoder to encode medical data subsets with different data category labels, obtain the temporal dependency of different data, and generate hidden layer representations of different data in the multi-omics data; Using similarity calculation based on the latent layer representations of different data corresponding to the patient, obtain a prostate cancer endocrine therapy resistance record instance whose similarity meets preset requirements, and retrieve a patient group similar to the patient from the prostate cancer endocrine therapy resistance record instance to construct a context learning sample; The spatial distribution of the latent layer representations and contextual learning examples corresponding to different patient data is aligned, and then imported into a large medical model to generate personalized medical knowledge for each data category label in the task of predicting endocrine therapy resistance. The latent representations of different data and personalized medical knowledge are encoded into embedding vectors, and the knowledge-enhanced data representations of omics data with different labels are output.

4. The prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence according to claim 1, characterized in that: Based on the knowledge-enhanced data representation, the best single-omics prediction model is matched to the omics data with different labels, and the best single-omics prediction model corresponding to the medical data with different labels is trained. Specifically: Obtain prediction model categories associated with different data category labels in the task of predicting prostate cancer endocrine therapy resistance based on the knowledge-enhanced data representation of patient multi-omics data; Using different data category labels, prediction model category labels, and associations to construct a relationship graph, using Metapath random walk to pre-train the relationship graph, and generate metapaths containing different data category labels as head nodes; A graph attention network is introduced to represent meta-paths with different data category labels as head nodes. The predicted model category nodes in the meta-path are used as neighbor nodes of the head node. The node feature vectors are obtained through graph convolution, and the attention weights of the neighbor nodes on the head node in different meta-paths are calculated. The neighbor nodes in each meta-path are sorted by the attention weight, the best single-omics prediction model in each meta-path is determined according to the sorting result, and the best single-omics prediction model for training sample training matching is constructed by using medical data subsets with different data category labels.

5. The prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence according to claim 1, characterized in that: In the multimodal feature integration module, stacking is used to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels. Specifically: Obtain training samples with different data category labels to construct a base training dataset, train each best single-omics prediction model, use a 5-fold cross-validation strategy, splice the prediction probabilities output by each best single-omics prediction model into a complete meta-feature matrix, and use the meta-feature matrix as a new training dataset to train the meta-model based on the SVM network; During the meta-selector training process, the meta-model hyperparameters are adjusted through grid search, and each optimal single-omics prediction model is retrained using the entire training data set. The contribution of each optimal single-omics prediction model is calculated using a non-negative least squares algorithm to generate meta-model weight information. The meta-model prediction results are combined with the weight information to generate the final prediction result of endocrine therapy resistance in prostate cancer; An independent test dataset is used to validate the model. When the final prediction results of prostate cancer endocrine therapy resistance meet the preset test criteria, a multi-omics prediction model for prostate cancer endocrine therapy resistance is output; Based on the patient's medical condition information, the patient's medication data within the preset time is determined, and the data representation of the patient's multi-omics data after knowledge enhancement is used as the model input to predict the patient's probability of drug resistance to the medication data within the preset time, and output the patient's drug resistance risk.

6. The multi-model artificial intelligence-based prostate cancer endocrine therapy resistance prediction system according to claim 1, characterized in that: In the model training module, obtain training data and use the balanced training data to train the best single-omics prediction model. Specifically: Extract features from prostate cancer endocrine therapy resistance records corresponding to different data category labels, construct resistance training data and non-resistance training data based on the true labels in the examples, and define the majority class training sample set and minority class training sample set based on the number of samples in the resistance training data and non-resistance training data; Select a training sample from the minority class training samples as a root sample, obtain K training samples closest to the root sample to generate a neighbor sample set, and select any one neighbor sample from the neighbor sample set as an auxiliary sample for sample synthesis; New training samples are synthesized by random interpolation between the root sample and the auxiliary sample, the new training samples are added to a new training sample set, the new training sample set is used to expand the minority class training samples, and the training data after data balance is used to train the best single-omics prediction model.

7. The prostate cancer endocrine therapy resistance prediction system based on multi-model artificial intelligence according to claim 6, characterized in that: In the model training module, federated learning was introduced to update the multi-omics prediction model for prostate cancer endocrine therapy resistance. Specifically: The multi-omics prediction model for prostate cancer endocrine therapy resistance was used as the initial global model. Training data from hospital nodes in different regions was extracted and local data was standardized. The federated parameters were initialized and training and updating were performed using a federated learning algorithm. In federated training, the central server sends the parameters of the initial global model to hospital nodes in different regions to build local models. Each hospital node uses local data to update the local model. The hospital node uploads the trained local model, performs weighted aggregation in the central server, and retrains the aggregated and updated global model using the balanced training data. The updated global model is re-issued, and the local model parameters are updated. After each hospital node performs drug resistance prediction, federal training is carried out using the drug resistance prediction results and confidence guidance. The global model is verified on the central server. After verification, the multi-omics prediction model of prostate cancer endocrine therapy resistance is updated.

8. A method for predicting resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence, characterized in that: The system for predicting drug resistance to endocrine therapy for prostate cancer based on multi-model artificial intelligence as described in any one of claims 1 to 7 comprises the following steps: Collect multi-omics data of patients and pre-process the collected multi-omics data; Use prior knowledge to generate data representations for omics data with different labels. Based on the knowledge-enhanced data representations, match the best single-omics prediction model for omics data with different labels, and train the best single-omics prediction model corresponding to medical data with different labels. Stacking is used to construct a multi-omics prediction model for prostate cancer endocrine therapy resistance by combining the best single-omics prediction models corresponding to medical data with different labels. Federated learning is introduced to train the model and predict the risk of drug resistance in patients within a preset time. Output an early warning of the patient's drug resistance risk within a preset time, and adjust the patient's treatment plan based on the drug resistance risk.

Images