Local advanced rectal cancer neoadjuvant therapy curative effect prediction system based on AI

Abstract

The invention relates to the technical field of rectal cancer, and discloses an AI-based local advanced rectal cancer neoadjuvant therapy curative effect prediction system, which comprises a data collection and processing module, a prediction model construction and training module, a prediction model verification and optimization module and a clinical application module, the data collecting and processing module is used for collecting clinical data, image data, radiotherapy target area data, radiotherapy dose data, radiobiology data and genomics data of a patient and then transmitting the data to the prediction model building and training module, and the prediction model building and training module extracts and selects key features in the data. And constructing a tumor segmentation multi-modal fusion and calibration network prediction model based on the PET / CT image. According to the invention, the data of the patient is collected through the data collection and processing module, so that the accuracy of radiotherapy effect and side reaction prediction is improved through the constructed prediction model, and the problem of feature capture limitation caused by a single center and a single mode of an existing AI model is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of rectal cancer, and in particular to an AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer. Background Art

[0002] Radiotherapy is one of the important means of treating malignant tumors. About 70% of cancer patients need to receive radiotherapy at different stages of the disease. Therefore, screening patients who are sensitive to radiotherapy before treatment and giving them individualized radiotherapy plans can improve the efficacy of patients and reduce side effects, which can achieve the dual benefits of prolonging survival and improving the quality of life. In recent years, AI has shown unique advantages in automatic delineation and planning design of radiotherapy and has been put into clinical application.

[0003] Most of the existing AI prediction models used to predict the efficacy and side effects of tumor radiotherapy are single-center, retrospective data, and are mostly based on single-modality images such as imaging genomics, which leads to the problem of limited capture of disease characteristics. Summary of the invention

[0004] In order to make up for the above shortcomings, the present invention provides an AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer, which aims to improve the existing AI prediction model. Most of the studies on predicting the efficacy and side effects of tumor radiotherapy are single-center, retrospective data, and are mostly based on single-modality images such as imaging genomics, which will lead to the problem of limited feature capture of the disease.

[0005] In the first aspect, the present invention provides the following technical solutions: an AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer, comprising a data collection and processing module, a prediction model construction and training module, a prediction model verification and optimization module, and a clinical application module; the data collection and processing module is used to collect the patient's clinical, imaging, radiotherapy target and dose, radiobiology, and genomics data, and then transmit the data to the prediction model construction and training module; the prediction model construction and training module extracts and selects key features in the data, constructs a tumor segmentation multimodal fusion and calibration network prediction model based on PET / CT images, and optimizes the prediction model parameters through training; the prediction model verification and optimization module is used to evaluate the prediction model performance through external validation, cross-validation, and sensitivity analysis, introduce more data based on clinical feedback and data accumulation, improve feature extraction methods, and adjust prediction model parameters; the clinical application module is used to apply the prediction model to actual clinical scenarios, assist doctors in formulating radiotherapy plans, monitor treatment efficacy in real time, and evaluate prognosis.

[0006] Preferably, the data collection and processing module includes a clinical data collection unit, an image data acquisition and processing unit, a radiotherapy target and dose distribution data acquisition unit, a radiobiological data detection unit and a genomic information acquisition unit. The clinical data collection unit is used to comprehensively collect the patient's basic information, radiotherapy efficacy, side effects and long-term prognosis information. The image data acquisition and processing unit is used to obtain the patient's imaging data and perform region of interest delineation and review. The radiotherapy target and dose distribution data acquisition unit is used to obtain dose distribution parameters of GTV, CTV, PTV and organs at risk. The radiobiological data detection unit is used to evaluate the tumor cell DNA damage repair ability through HRD score detection, comet experiment and AP site detection. The genomic information acquisition unit is used to detect rectal cancer tissue using whole exome sequencing technology to obtain genomic data.

[0007] Preferably, the prediction model construction and training module includes a feature extraction and selection unit, a prediction model construction unit and a prediction model training unit, the feature extraction and selection unit is used to mine and screen key features related to rectal cancer from the data, the prediction model construction unit is used to construct a multimodal fusion downsampling block with a residual structure and design a multimodal mutual calibration block based on the initial structure, and the prediction model training unit is used to input the processed multimodal data into the prediction model for training, and adjust the prediction model parameters to optimize the prediction model performance.

[0008] Preferably, the prediction model verification and optimization module includes an external verification unit, a cross-validation and sensitivity analysis unit, and a prediction model optimization unit. The external verification unit is used to test the trained prediction model through independent data, the cross-validation and sensitivity analysis unit is used to perform cross-validation and sensitivity analysis operations to determine the stability and reliability of the prediction model, and the prediction model optimization unit is used to optimize the prediction model by introducing new data, improving feature extraction methods, and adjusting prediction model parameters.

[0009] Preferably, the clinical application module includes a radiotherapy plan formulation auxiliary unit, a therapeutic efficacy monitoring and prognosis evaluation unit, and a unit for integrating with a clinical decision support system. The radiotherapy plan formulation auxiliary unit is used to provide doctors with selection suggestions for radiotherapy dose and irradiation range parameters based on the prediction results of the prediction model. The therapeutic efficacy monitoring and prognosis evaluation unit is used to use the prediction model to monitor the therapeutic efficacy in real time during the treatment process and to evaluate the patient's prognosis. The unit for integrating with the clinical decision support system is used to integrate the prediction model with the clinical decision support system through a data interface.

[0010] Preferably, the clinical data collection unit includes a patient basic information collection subunit, a therapeutic effect evaluation subunit, a side effect evaluation subunit and a prognosis information collection subunit. The patient basic information collection subunit is used to collect age, gender, maximum tumor diameter, pathological type, and clinical stage. The therapeutic effect evaluation subunit is used to evaluate the therapeutic effect of radiotherapy one month after radiotherapy through the RECI ST standard (version 1.1). The side effect evaluation subunit is used to evaluate radiotherapy-related side effects through TCAE version 5.0. The prognosis information collection subunit is used to obtain the patient's long-term prognosis information through medical records and follow-up records.

[0011] Preferably, the feature extraction and selection unit includes a clinical feature processing subunit, an imaging genomics feature processing subunit and a radiobiological feature processing subunit. The clinical feature processing subunit is used to collect gender, age, maximum tumor diameter, pathological type, target area and organ-at-risk DVH parameters, radiotherapy toxicity reaction grading, hematological index changes and gene expression status information, and perform univariate and multivariate Cox regression analysis screening. The imaging genomics feature processing subunit is used to extract shape, first-order, texture and wavelet transform features, and perform dimensionality reduction processing using minimum shrinkage and selection operator correlation analysis.

[0012] In a second aspect, the present invention provides the following technical solution, the method for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer based on AI comprises the following steps:

[0013] S1. Data collection: collect patient clinical data, obtain imaging data and have it outlined and reviewed by physicians, collect radiotherapy target area and dose distribution data, detect radiobiological data, and obtain genomic information through the data collection and processing module;

[0014] S2. Feature extraction and selection: The feature extraction and selection unit is used to mine and screen key features of clinical features, imaging features, and radiobiological features;

[0015] S3. Prediction model construction: Develop a multimodal fusion and calibration network for tumor segmentation based on PET / CT images through the prediction model construction unit based on key features, propose a multimodal fusion downsampling block (MFDB) to fuse multimodal image features and retain their unique features, and design a multimodal mutual calibration block (MMCB) to extract multi-scale pathological features with the help of an attention mechanism to guide the network to focus on the tumor area;

[0016] S4, prediction model training: repeatedly adjust the prediction model parameters using sample data;

[0017] S5. Validation and optimization of prediction models: Conduct external validation based on prospective, multi-center clinical research cohorts using external independent data to clarify the robustness and versatility of the prediction model, conduct cross-validation and sensitivity analysis to evaluate stability and reliability, and then improve the accuracy and practicality of the prediction model by introducing more modality data, improving feature extraction and selection methods, and adjusting parameters based on clinical feedback and data accumulation;

[0018] S6. Clinical application: Predict the results through the prediction model, assist doctors to accurately determine the radiotherapy dose and irradiation range, monitor the efficacy in real time during treatment, adjust the strategy in time and evaluate the prognosis, and integrate the prediction model with the clinical decision support system.

[0019] In the third aspect, the invention provides the following technical solution: a computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the above-mentioned AI-based method for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer when executing the computer program.

[0020] In a fourth aspect, the present invention provides the following technical solution: a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned AI-based method for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer is implemented.

[0021] The present invention has the following beneficial effects:

[0022] 1. In the present invention, the prediction model constructed by collecting the patient's clinical, imaging, radiotherapy target and dose, radiobiology and genomics data through the data collection and processing module improves the accuracy of radiotherapy efficacy and side effect prediction, thereby improving the existing AI prediction model. Most of the studies on predicting the efficacy and side effects of tumor radiotherapy are single-center, retrospective data, and are mostly based on single-modality images such as imaging genomics, which will lead to the problem of limited capture of disease characteristics.

[0023] 2. In the present invention, through the clinical application module, doctors are provided with suggestions on the selection of radiotherapy dose and irradiation range parameters. During the treatment process, the prediction model is used to monitor the therapeutic effect in real time, and the patient's prognosis is evaluated at the same time, so as to further optimize the radiotherapy dose of tumor radiotherapy and the limit of normal tissues and organs, so as to achieve individualized precision radiotherapy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a system architecture diagram of the AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer proposed by the present invention; DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Embodiment 1

[0027] Reference Figure 1 In the first embodiment of the present invention, the present invention provides an AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer, including a data collection and processing module, a prediction model construction and training module, a prediction model verification and optimization module, and a clinical application module. The data collection and processing module is used to collect the patient's clinical, imaging, radiotherapy target and dose, radiobiology and genomics data, and then transmit the data to the prediction model construction and training module. The prediction model construction and training module extracts and selects key features in the data, constructs a tumor segmentation multimodal fusion and calibration network prediction model based on PET / CT images, and optimizes the prediction model parameters through training. The prediction model verification and optimization module is used to evaluate the prediction model performance through external validation, cross-validation and sensitivity analysis, introduce more data based on clinical feedback and data accumulation, improve feature extraction methods and adjust prediction model parameters. The clinical application module is used to apply the prediction model to actual clinical scenarios, assist doctors in formulating radiotherapy plans, monitor treatment efficacy in real time, and evaluate prognosis.

[0028] Specifically, the data collection and processing module can integrate multi-dimensional data such as clinical, imaging, radiotherapy target, radiobiology and genomics. Through standardized evaluation, professional operation, complex testing and fine screening, it can provide a comprehensive, accurate and targeted data foundation for subsequent prediction model construction and clinical application. The prediction model construction and training module can develop an effective network prediction model based on multimodal data fusion calibration, and train it to learn data features and patterns to improve the accuracy of predicting the efficacy of neoadjuvant treatment for locally advanced rectal cancer. The prediction model verification and optimization module can use external independent data to test the robustness and versatility of the prediction model, evaluate stability and reliability through cross-validation and sensitivity analysis, and optimize the prediction model based on clinical feedback to enhance its accuracy and practicality. The clinical application module can assist doctors in accurately formulating radiotherapy plans, monitoring efficacy and prognosis in real time, and can also be integrated with clinical decision support systems to help them make scientific and reasonable medical decisions.

[0029] The data collection and processing module includes a clinical data collection unit, an imaging data acquisition and processing unit, a radiotherapy target and dose distribution data acquisition unit, a radiobiological data detection unit, and a genomic information acquisition unit. The clinical data collection unit is used to comprehensively collect the patient's basic information, radiotherapy efficacy, side effects, and long-term prognosis information. The imaging data acquisition and processing unit is used to obtain the patient's imaging data and to delineate and review the region of interest. The radiotherapy target and dose distribution data acquisition unit is used to obtain the dose distribution parameters of GTV, CTV, PTV, and organs at risk. The radiobiological data detection unit is used to evaluate the DNA damage repair ability of tumor cells through HRD score detection, comet assay, and AP site detection. The genomic information acquisition unit is used to detect rectal cancer tissues using whole exome sequencing technology and obtain genomic data.

[0030] Specifically, the clinical data collection unit can obtain key data such as patient basic information, radiotherapy efficacy and side effects, and long-term prognosis, laying a solid information foundation for subsequent diagnosis and treatment evaluation, prediction model training, and clinical decision-making. The image data acquisition and processing unit can obtain patient imaging data from relevant medical centers and process them through professional physicians to provide accurate and reliable imaging basis for subsequent prediction model construction and clinical analysis. The radiotherapy target and dose distribution data acquisition unit can collect key target and dose distribution parameters during radiotherapy, providing accurate radiotherapy data support for subsequent prediction model training, radiotherapy plan optimization, and efficacy evaluation. The radiobiology data detection unit can obtain HRD scores, DNA damage repair capacity-related data, AP s ite detection data, etc., providing key radiobiology data support for evaluating radiotherapy efficacy, building and optimizing prediction models. The genomic information acquisition unit can obtain genomic data such as whole exome sequencing of rectal cancer tissues, including mutations and polypeptide sites, providing important genetic basis for subsequent feature extraction, prediction model construction, and precision medical decision-making.

[0031] The prediction model construction and training module includes a feature extraction and selection unit, a prediction model construction unit and a prediction model training unit. The feature extraction and selection unit is used to mine and screen key features related to rectal cancer from the data. The prediction model construction unit is used to construct a multimodal fusion downsampling block with a residual structure and design a multimodal mutual calibration block based on the initial structure. The prediction model training unit is used to input the processed multimodal data into the prediction model for training and adjust the prediction model parameters to optimize the prediction model performance.

[0032] Specifically, the feature extraction and selection unit can extract key features from clinical, imaging genomics, radiobiology and other data, remove redundant information, and provide the most representative and relevant data feature combination for building an accurate and efficient prediction model. The prediction model construction unit can develop a network prediction model architecture that integrates multimodal data and effectively extracts pathological features, building a core framework for accurately predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer. The prediction model training unit can use a large amount of multimodal processing data to adjust the parameters of the constructed prediction model, so that the prediction model can fully learn the patterns and features in the data, thereby effectively improving the accuracy of the prediction model for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer.

[0033] The prediction model verification and optimization module includes an external verification unit, a cross-validation and sensitivity analysis unit, and a prediction model optimization unit. The external verification unit is used to test the trained prediction model through independent data. The cross-validation and sensitivity analysis unit is used to perform cross-validation and sensitivity analysis operations to determine the stability and reliability of the prediction model. The prediction model optimization unit is used to optimize the prediction model by introducing new data, improving feature extraction methods, and adjusting prediction model parameters.

[0034] Specifically, the external validation unit can use independent data from prospective, multicenter clinical research cohorts to test the prediction model, so as to determine the robustness and versatility of the prediction model under different data sources and judge whether the prediction model can be widely applied in actual clinical scenarios. The cross-validation and sensitivity analysis unit can comprehensively evaluate the stability of the prediction model under different data divisions, as well as the sensitivity of changes in various factors to the prediction results of the prediction model, so as to determine the reliability and key influencing factors of the prediction model. The prediction model optimization unit can effectively improve the accuracy and practicality of the prediction model in predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer based on clinical feedback and data accumulation by introducing new data, improving feature extraction and selection methods, and adjusting prediction model parameters.

[0035] The clinical application module includes a radiotherapy plan formulation auxiliary unit, a efficacy monitoring and prognosis evaluation unit, and a unit integrated with a clinical decision support system. The radiotherapy plan formulation auxiliary unit is used to provide doctors with selection recommendations for radiotherapy dose and irradiation range parameters based on the prediction results of the prediction model. The efficacy monitoring and prognosis evaluation unit is used to use the prediction model to monitor the efficacy in real time during the treatment process and evaluate the patient's prognosis. The unit integrated with the clinical decision support system is used to integrate the prediction model with the clinical decision support system through a data interface.

[0036] Specifically, the radiotherapy plan formulation auxiliary unit can predict the results based on the prediction model, help doctors accurately determine the radiotherapy dose, reasonably plan the irradiation range and other key parameters, and improve the scientificity and adaptability of the radiotherapy plan. The efficacy monitoring and prognosis evaluation unit can use the prediction model to track the efficacy in real time during the treatment process, promptly identify problems and adjust the treatment strategy, and effectively evaluate the prognosis of the disease, providing support for the dynamic optimization of the treatment process and the prognosis of patients. By combining with the clinical decision support system, the unit can integrate the prediction model prediction information, provide doctors with comprehensive, systematic and in-depth data interpretation and diagnosis and treatment references, and help them make more scientific and accurate clinical decisions that fit the actual patient.

[0037] The clinical data collection unit includes a patient basic information collection subunit, an efficacy evaluation subunit, an adverse reaction evaluation subunit, and a prognosis information collection subunit. The patient basic information collection subunit is used to collect age, gender, maximum tumor diameter, pathological type, and clinical stage. The efficacy evaluation subunit is used to evaluate the efficacy of radiotherapy one month after radiotherapy using the RECIST standard (version 1.1). The adverse reaction evaluation subunit is used to evaluate radiotherapy-related adverse reactions using TCAE version 5.0. The prognosis information collection subunit is used to obtain patients' long-term prognosis information through medical records and follow-up records.

[0038] Specifically, the patient basic information collection subunit can collect basic information such as the patient's age, gender, and maximum tumor diameter, providing basic data support for subsequent evaluation of radiotherapy efficacy, side effects, construction of prediction models, and formulation of treatment plans. The efficacy evaluation subunit can evaluate the effect after radiotherapy (such as CR, PR, SD, PD) according to standards, and provide key efficacy data references for subsequent adjustment of treatment strategies, monitoring of disease progression, prediction of prognosis, and optimization of prediction models. The side effect evaluation subunit can systematically collect, analyze, and quantify the patient's side effect information after radiotherapy, providing accurate data support for side effects for timely adjustment of radiotherapy plans, relief of patient pain, optimization of the treatment process, and improvement of prediction model predictive factors. The prognosis information collection subunit can collect prognosis-related data such as patient survival time and recurrence, providing important basis for judging the development trend of the disease, evaluating the long-term effectiveness of treatment plans, and improving the prognostic factors in the prediction model.

[0039] The feature extraction and selection unit includes a clinical feature processing subunit, an imaging genomics feature processing subunit, and a radiobiological feature processing subunit. The clinical feature processing subunit is used to collect information on gender, age, maximum tumor diameter, pathological type, target volume and organ-at-risk DVH parameters, radiotherapy toxicity reaction grading, changes in hematological indicators, and gene expression status, and perform univariate and multivariate Cox regression analysis and screening. The imaging genomics feature processing subunit is used to extract shape, first-order, texture, and wavelet transform features, and perform dimensionality reduction using minimum shrinkage and selection operator correlation analysis.

[0040] Specifically, the clinical feature processing subunit can organize, classify and quantify the clinical features of patients (such as symptoms, test results, etc.), providing standardized and valuable clinical data input for the subsequent construction of prediction models, accurate evaluation of radiotherapy effects, analysis of side effects and prognosis. The imaging genomics feature processing subunit can screen, quantify and standardize a large number of complex features extracted from medical images, providing key image-based quantitative data support for the construction of accurate prediction models, auxiliary radiotherapy plan decisions, and evaluation of efficacy and prognosis. The radiobiological feature processing subunit can integrate, analyze and standardize radiobiological related data (such as cell survival curves, DNA damage repair indicators, etc.), providing key data support based on radiobiological characteristics for radiotherapy efficacy prediction, side effect evaluation, prognosis judgment and prediction model construction.

[0041] Embodiment 2:

[0042] Reference Figure 1 In a second embodiment of the present invention, the present invention provides an AI-based method for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer, comprising the following steps:

[0043] S1. Data collection: collect patient clinical data, obtain imaging data and have it outlined and reviewed by physicians, collect radiotherapy target area and dose distribution data, detect radiobiological data, and obtain genomic information through the data collection and processing module;

[0044] S2. Feature extraction and selection: The feature extraction and selection unit is used to mine and screen key features of clinical features, imaging features, and radiobiological features;

[0045] S3. Prediction model construction: Develop a multimodal fusion and calibration network for tumor segmentation based on PET / CT images through the prediction model construction unit based on key features, propose a multimodal fusion downsampling block (MFDB) to fuse multimodal image features and retain their unique features, and design a multimodal mutual calibration block (MMCB) to extract multi-scale pathological features with the help of an attention mechanism to guide the network to focus on the tumor area;

[0046] S4, prediction model training: repeatedly adjust the prediction model parameters using sample data;

[0047] S5. Validation and optimization of prediction models: Conduct external validation based on prospective, multi-center clinical research cohorts using external independent data to clarify the robustness and versatility of the prediction model, conduct cross-validation and sensitivity analysis to evaluate stability and reliability, and then improve the accuracy and practicality of the prediction model by introducing more modality data, improving feature extraction and selection methods, and adjusting parameters based on clinical feedback and data accumulation;

[0048] S6. Clinical application: Predict the results through the prediction model, assist doctors to accurately determine the radiotherapy dose and irradiation range, monitor the efficacy in real time during treatment, adjust the strategy in time and evaluate the prognosis, and integrate the prediction model with the clinical decision support system.

[0049] Embodiment 3

[0050] The third embodiment of the present invention is based on the same inventive concept. The present invention proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the AI-based neoadjuvant treatment efficacy prediction method for locally advanced rectal cancer in the above-mentioned embodiment.

[0051] Embodiment 4

[0052] The fourth embodiment of the present invention is based on the same inventive concept. A computer device proposed by the present invention includes: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory to execute the AI-based neoadjuvant treatment efficacy prediction method for locally advanced rectal cancer of the above embodiment.

[0053] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0054] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer, including a data collection and processing module, a prediction model construction and training module, a prediction model verification and optimization module, and a clinical application module, characterized by: The data collection and processing module is used to collect the patient's clinical, imaging, radiotherapy target and dose, radiobiology and genomics data, and then transmit the data to the prediction model construction and training module. The prediction model construction and training module extracts and selects key features in the data, constructs a tumor segmentation multimodal fusion and calibration network prediction model based on PET / CT images, and optimizes the prediction model parameters through training. The prediction model verification and optimization module is used to evaluate the prediction model performance through external validation, cross-validation and sensitivity analysis, introduce more data based on clinical feedback and data accumulation, improve feature extraction methods and adjust prediction model parameters. The clinical application module is used to apply the prediction model to actual clinical scenarios, assist doctors in formulating radiotherapy plans, monitor treatment efficacy in real time and evaluate prognosis.

2. The AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to claim 1, characterized in that: The data collection and processing module includes a clinical data collection unit, an image data acquisition and processing unit, a radiotherapy target and dose distribution data acquisition unit, a radiobiological data detection unit and a genomic information acquisition unit. The clinical data collection unit is used to comprehensively collect the patient's basic information, radiotherapy efficacy, side effects and long-term prognosis information. The image data acquisition and processing unit is used to obtain the patient's imaging data and to delineate and review the region of interest. The radiotherapy target and dose distribution data acquisition unit is used to obtain the dose distribution parameters of GTV, CTV, PTV and organs at risk. The radiobiological data detection unit is used to evaluate the tumor cell DNA damage repair ability through HRD score detection, comet experiment and AP site detection. The genomic information acquisition unit is used to detect rectal cancer tissue using whole exome sequencing technology to obtain genomic data.

3. The AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to claim 1, characterized in that: The prediction model construction and training module includes a feature extraction and selection unit, a prediction model construction unit and a prediction model training unit. The feature extraction and selection unit is used to mine and screen key features related to rectal cancer from data. The prediction model construction unit is used to construct a multimodal fusion downsampling block with a residual structure and design a multimodal mutual calibration block based on an initial structure. The prediction model training unit is used to input the processed multimodal data into the prediction model for training and adjust the prediction model parameters to optimize the prediction model performance.

4. The AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to claim 1, characterized in that: The prediction model verification and optimization module includes an external verification unit, a cross-validation and sensitivity analysis unit, and a prediction model optimization unit. The external verification unit is used to test the trained prediction model through independent data, the cross-validation and sensitivity analysis unit is used to perform cross-validation and sensitivity analysis operations to determine the stability and reliability of the prediction model, and the prediction model optimization unit is used to optimize the prediction model by introducing new data, improving feature extraction methods, and adjusting prediction model parameters.

5. The AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to claim 1, characterized in that: The clinical application module includes a radiotherapy plan formulation auxiliary unit, an efficacy monitoring and prognosis evaluation unit, and a unit for combining with a clinical decision support system. The radiotherapy plan formulation auxiliary unit is used to provide doctors with selection suggestions for radiotherapy dose and irradiation range parameters based on the prediction results of the prediction model. The efficacy monitoring and prognosis evaluation unit is used to use the prediction model to monitor the efficacy in real time during the treatment process and evaluate the patient's prognosis. The unit for combining with the clinical decision support system is used to integrate the prediction model with the clinical decision support system through a data interface.

6. The AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to claim 2, characterized in that: The clinical data collection unit includes a patient basic information collection subunit, a therapeutic effect evaluation subunit, a side effect evaluation subunit and a prognosis information collection subunit. The patient basic information collection subunit is used to collect age, gender, maximum tumor diameter, pathological type, and clinical stage. The therapeutic effect evaluation subunit is used to evaluate the therapeutic effect of radiotherapy one month after radiotherapy according to the RECIST standard (version 1.1). The side effect evaluation subunit is used to evaluate radiotherapy-related side effects according to TCAE version 5.

0. The prognosis information collection subunit is used to obtain the patient's long-term prognosis information through medical records and follow-up records.

7. The AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to claim 3, characterized in that: The feature extraction and selection unit includes a clinical feature processing subunit, an imaging genomics feature processing subunit and a radiobiological feature processing subunit. The clinical feature processing subunit is used to collect gender, age, maximum tumor diameter, pathological type, target area and organ-at-risk DVH parameters, radiotherapy toxicity reaction grading, hematological index changes and gene expression status information, and perform univariate and multivariate Cox regression analysis screening. The imaging genomics feature processing subunit is used to extract shape, first-order, texture and wavelet transform features, and perform dimensionality reduction processing using minimum shrinkage and selection operator correlation analysis.

8. A prediction method for the neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer based on AI, characterized in that: Using the AI-based neoadjuvant therapy efficacy prediction system for locally advanced rectal cancer according to any one of claims 1 to 7, comprising the following steps: S1. Data collection: collect patient clinical data, obtain imaging data and have it outlined and reviewed by physicians, collect radiotherapy target area and dose distribution data, detect radiobiological data, and obtain genomic information through the data collection and processing module; S2. Feature extraction and selection: The feature extraction and selection unit is used to mine and screen key features of clinical features, imaging features, and radiobiological features; S3. Prediction model construction: Develop a multimodal fusion and calibration network for tumor segmentation based on PET / CT images through the prediction model construction unit based on key features, propose a multimodal fusion downsampling block (MFDB) to fuse multimodal image features and retain their unique features, and design a multimodal mutual calibration block (MMCB) to extract multi-scale pathological features with the help of an attention mechanism to guide the network to focus on the tumor area; S4, prediction model training: repeatedly adjust the prediction model parameters using sample data; S5. Validation and optimization of prediction models: Conduct external validation with external independent data based on prospective, multi-center clinical research cohorts to clarify the robustness and versatility of the prediction model, conduct cross-validation and sensitivity analysis to evaluate stability and reliability, and then improve the accuracy and practicality of the prediction model by introducing more modal data, improving feature extraction and selection methods, and adjusting parameters based on clinical feedback and data accumulation; S6. Clinical application: Predict the results through the prediction model, assist doctors to accurately determine the radiotherapy dose and irradiation range, monitor the efficacy in real time during treatment, adjust the strategy in time and evaluate the prognosis, and integrate the prediction model with the clinical decision support system.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer based on AI as described in claim 8 is implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by the processor, the method for predicting the efficacy of neoadjuvant therapy for locally advanced rectal cancer based on AI as claimed in claim 8 is implemented.

Images