Automatic prediction system for tumor chemoradiotherapy reaction

Abstract

The invention discloses a tumor chemoradiotherapy reaction automatic prediction system, and particularly relates to the technical field of medical automation, the tumor chemoradiotherapy reaction automatic prediction system comprises a data acquisition module, a data preprocessing module, a feature extraction and selection module, a model training module, a prediction analysis module, an early warning module and a feedback module; the multi-source data acquisition module comprehensively gathers patient information, the quality is improved through the preprocessing module, a foundation is built for the feature extraction and selection module to accurately screen key features, the data dimension is greatly reduced, and the model training efficiency is remarkably improved; the model training module applies a machine learning algorithm to deeply mine a data complex relationship, so that the prediction is more accurate and reliable; on the basis, the prediction analysis module outputs detailed chemoradiotherapy reaction prediction and provides confidence evaluation to help doctors to make decisions and select; and the early warning module is responsible for predicting, warning abnormity in time and helping doctors to plan intervention in advance.

Description

Technical Field

[0001] The present invention relates to the field of medical automation technology, and more particularly to an automatic prediction system for tumor radiotherapy and chemotherapy responses. Background Art

[0002] In the field of cancer treatment, radiotherapy and chemotherapy are important treatment methods, but different patients have great differences in their responses to radiotherapy and chemotherapy. This diversity of individual responses makes the formulation of treatment plans challenging. In the past, doctors mainly relied on their own clinical experience, limited traditional detection indicators and simple imaging observations to estimate patients' radiotherapy and chemotherapy responses. However, these methods have many limitations.

[0003] Based on clinical experience, there may be subjective differences in the judgments of different doctors on similar cases, and there is a lack of objective and unified standards. It is difficult to accurately capture the impact of each patient's unique physical characteristics and disease characteristics on the response to radiotherapy and chemotherapy. In terms of traditional detection indicators, such as common tumor marker tests, although they can provide a certain reference, a single marker often cannot fully reflect the complex biological behavior of the tumor and the overall physical condition of the patient. For example, although carcinoembryonic antigen (CEA) tends to increase in some tumor patients, other factors such as inflammation may also cause it to increase, making the accuracy of using this single indicator to predict the response to radiotherapy and chemotherapy poor.

[0004] Imaging examinations also have shortcomings. Traditional imaging methods such as CT and MRI mainly focus on observing changes in the morphology and size of tumors, and it is difficult to deeply reveal molecular-level information such as metabolic activities and gene expression within the tumor. For tumor heterogeneity, that is, the differences in genes, metabolism, etc. between cells in different regions of the same tumor, imaging examinations are difficult to accurately present. These heterogeneities play a key role in the response to radiotherapy and chemotherapy, because tumor cells with different characteristics have different sensitivities to radiotherapy and chemotherapy.

[0005] In addition, there is currently a lack of technical systems that can integrate multi-source data for efficient and accurate analysis. Patients' clinical data, imaging data, genetic data, etc. are often stored in a scattered manner and difficult to integrate and utilize. This makes it impossible to explore the potential correlations between the data from a holistic level, and thus it is impossible to accurately predict patients' responses to radiotherapy and chemotherapy, which is not conducive to the formulation of personalized treatment plans. To overcome these problems, an automatic prediction system for tumor radiotherapy and chemotherapy responses is proposed. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an automatic prediction system for tumor radiotherapy and chemotherapy reactions. The multi-source data acquisition module comprehensively collects patient information, improves the quality through the preprocessing module, lays a solid foundation for the feature extraction and selection module to accurately screen key features, greatly reduces the data dimension, and significantly improves the model training efficiency. The model training module uses machine learning algorithms to deeply explore the complex relationships in the data, making the prediction more accurate and reliable. On this basis, the prediction analysis module outputs a detailed prediction of radiotherapy and chemotherapy reactions and provides a confidence evaluation to assist doctors in decision-making and selection. The early warning module is responsible for making predictions and promptly warning of abnormalities to help doctors plan interventions in advance. The feedback module collects actual treatment feedback to optimize the model, enabling the system to keep up with clinical changes, continuously improve the prediction accuracy, and strongly promote the accurate formulation of radiotherapy and chemotherapy plans for tumor patients, improving the treatment effect and the quality of medical services.

[0007] To achieve the above object, the present invention provides the following technical solutions: It includes a data acquisition module, a data preprocessing module, a feature extraction and selection module, a model training module, a prediction analysis module, an early warning module, and a feedback module, where:

[0008] Data acquisition module: Collects patient clinical data, imaging data, and genetic data from the hospital information system, imaging equipment, and genetic testing instruments, as well as relevant data during the treatment process;

[0009] Data preprocessing module: Cleans the collected raw data, normalizes numerical data, and performs natural language processing on text data;

[0010] Feature extraction and selection module: Extracts clinical, imaging, and gene-related features from the preprocessed data and uses feature selection algorithms to screen features;

[0011] Model training module: Utilizes the selected feature data, adopts machine learning algorithms to train the model based on historical patient data, and optimizes the parameters through cross-validation;

[0012] Prediction analysis module: Inputs new patient data into the trained model to predict radiotherapy and chemotherapy reactions and conducts a confidence evaluation;

[0013] Early warning module: Compares the prediction result with a pre-set risk threshold, and issues a warning signal if the result exceeds the threshold;

[0014] Feedback module: Collects feedback information during the treatment process, analyzes and organizes it for updating and optimizing the prediction model.

[0015] In a preferred embodiment: The data acquisition module is connected to the hospital information system through a standardized interface to regularly collect information on the patient's past medical history and family medical history in the medical record. It obtains and converts DICOM format image data from imaging devices of different brands and models into the unified format of the system through a special data conversion and adaptation program, and receives gene sequence data and gene expression data generated by gene detection instruments in real time through a network communication interface. It collects treatment parameter data such as radiotherapy dose, number of radiotherapy sessions, types of chemotherapy drugs, and dosage of chemotherapy drugs recorded by radiotherapy and chemotherapy equipment.

[0016] In a preferred embodiment: The data preprocessing module uses data mining algorithms to clean duplicate data records according to the established data rule library, correct incorrect data according to data logical relationships and medical knowledge rules, fill in missing values using multiple imputation methods combined with the data characteristics of similar patients, map numerical data to the [0, 1] interval using the maximum-minimum normalization method, and convert text data into vector form using the bag-of-words model in natural language processing and perform part-of-speech tagging and named entity recognition operations to extract key medical terms and related information.

[0017] In a preferred embodiment: The feature extraction and selection module extracts clinical features, including age, gender, tumor stage, body mass index, white blood cell count, red blood cell count, platelet count in blood routine, alanine aminotransferase and aspartate aminotransferase related to liver and kidney function, and serum creatinine. It extracts image features, including three-dimensional size of the tumor, irregularity of shape, roughness of internal texture, and contrast, using image segmentation algorithms and texture analysis algorithms. It extracts gene features such as mutation types and gene copy number variations of genes related to radiotherapy and chemotherapy sensitivity, such as TP53 and BRCA1, and uses a recursive feature elimination algorithm combined with the feature importance evaluation of a random forest model to screen the feature subset that contributes the most to the prediction of radiotherapy and chemotherapy response.

[0018] In a preferred embodiment: For data with image features, the model training module preferably uses a convolutional neural network to construct the model structure and sets multiple convolutional layers and pooling layers to extract deep image features. For data with sequence features, it uses the long short-term memory network model in the recurrent neural network. The training data is randomly divided according to the ratio of 70% as the training set, 15% as the validation set, and 15% as the test set. The five-fold cross-validation method is used to adjust the learning rate, number of iterations, and number of hidden layer nodes of the model during the training process using the validation set, and the hyperparameters are adjusted to make the model achieve the best prediction performance on the test set.

[0019] In a preferred embodiment: when predicting the chemoradiotherapy response, the prediction analysis module outputs classification results of complete tumor remission, partial remission, stable disease, and disease progression, as well as probability values for each category, details common adverse reactions such as nausea, vomiting, hair loss, and myelosuppression, and grades the occurrence probability of each adverse reaction as low probability, medium probability, and high probability. The Bootstrap method is used to construct a model by sampling training data to calculate the variance of the prediction result to measure the confidence level.

[0020] In a preferred embodiment: based on a large amount of clinical research data and authoritative medical guidelines, the early warning module sets risk thresholds for different tumor types and chemoradiotherapy regimens respectively. If the predicted high-probability occurrence of nausea and vomiting exceeds 50%, the predicted probability of disease progression in the treatment effect exceeds 30%, and the situation exceeds the threshold, a message is pushed to the work terminal of the attending physician through the hospital internal information system and a warning message is sent to the monitoring system of the nurse station.

[0021] In a preferred embodiment: the feedback module collects feedback information on the difference between the prediction result and the actual treatment effect during the treatment process by doctors, as well as feedback information on the patient's subjective feelings. The collected feedback information is classified and sorted, and statistical analysis methods are used to analyze the key data characteristics in the feedback information. The analysis results are used to update and optimize the parameters and structure of the prediction model to improve the prediction accuracy and practicality of the system.

[0022] The technical effects and advantages of the present invention: through the multi-source data collection module, patient information can be comprehensively collected, and the preprocessing module effectively improves the data quality, laying a solid foundation for subsequent analysis; the feature extraction and selection module accurately screens key features, greatly reducing the data dimension and improving the model training efficiency; the model training module constructed by machine learning algorithms can deeply explore the complex relationships in the data, making the prediction results more accurate and reliable; the prediction analysis module can not only give a detailed prediction of chemoradiotherapy response, but also provide confidence assessment, providing strong support for doctors' decision-making; the early warning module, based on accurate predictions, promptly warns of abnormal situations, helping doctors plan intervention measures in advance and reduce risks; the feedback module collects actual treatment feedback and uses it to optimize the model, enabling the system to continuously adapt to clinical changes and continuously improve the prediction accuracy. Overall, it provides strong assistance for the precise formulation of chemoradiotherapy regimens for tumor patients and the improvement of treatment effects, effectively improving the quality of medical services and having extremely high clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the working system module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] Example embodiments will now be described more fully with reference to the accompanying drawings. An automatic prediction system for tumor radiotherapy and chemotherapy reactions includes a data acquisition module, a data preprocessing module, a feature extraction and selection module, a model training module, a prediction analysis module, an early warning module, and a feedback module, wherein:

[0026] Data acquisition module: Collects patient clinical data, imaging data, and genetic data, as well as relevant data during the treatment process, from the hospital information system, imaging equipment, and genetic testing instruments.

[0027] Data preprocessing module: Cleans the collected raw data, normalizes numerical data, and performs natural language processing on text data.

[0028] Feature extraction and selection module: Extracts clinical, imaging, and genetic-related features from the preprocessed data, and uses feature selection algorithms to screen features.

[0029] Model training module: Utilizes the screened feature data, and trains a model based on historical patient data using machine learning algorithms, and optimizes the parameters through cross-validation.

[0030] Prediction analysis module: Inputs new patient data into the trained model to predict radiotherapy and chemotherapy reactions, and performs confidence evaluation.

[0031] Early warning module: Compares the prediction result with a pre-set risk threshold, and issues an early warning signal if the result exceeds the threshold.

[0032] Feedback module: Collects feedback information during the treatment process, analyzes and organizes it for updating and optimizing the prediction model.

[0033] The data acquisition module is connected to the hospital information system through a standardized interface to regularly collect information on past medical history and family medical history in the patient's medical record. It obtains and converts DICOM format image data from different brands and models of imaging equipment into a unified system format through a special data conversion adapter. It receives gene sequence data and gene expression data generated by genetic testing instruments in real time through a network communication interface, and collects radiotherapy dose, number of radiotherapy sessions, type of chemotherapy drugs, and treatment parameter data recorded by radiotherapy and chemotherapy equipment. It is closely connected to the hospital information system through the FHIR standard interface, which can not only obtain basic information such as age, gender, height, and weight of patients in real time, but also deeply explore the detailed past medical history of patients. For each past disease, the time of first onset t is accurately recorded. onset , diagnosis time t diagnosis , duration of treatment t treatment and treatment evaluation effect E; at the same time, the system automatically collects multiple blood test results such as white blood cell count WBC, red blood cell count RBC, platelet count PLT and the corresponding test time t test , construct time series data {WBC1,t test1 ;WBC2,t test2 ;…;WBC n ,t testn}, using the moving average algorithm for vital signs such as heart rate HR.

[0034] The data is preliminarily processed and the k-period moving average is calculated This not only effectively smooths data, reduces interference from accidental factors, and more accurately reflects the patient's heart rate change trend, but also provides a rich and accurate data basis for subsequent analysis of changes in the patient's physical condition over time, greatly improving the availability and analytical value of clinical data.

[0035] Connected to various imaging devices (CT, MRI, PET-CT) based on the DICOM standard protocol, when acquiring CT images, in addition to obtaining conventional scanning parameters (tube voltage kV, tube current mA, slice thickness TH, etc.), the image noise level is accurately measured by calculating the standard deviation of pixel values in uniform areas of the image. (M, N are the number of rows and columns in the uniform area, I ij is the pixel value in the area, is the average pixel value of the region); for PET-CT images, the injection dose D of the radiotracer and the injection time t are recorded in detail. inject , Scan start time t scan And the patient's blood glucose level BG, and according to the formula Corrected standardized uptake value (SUV); The above-mentioned imaging data acquisition and correction method provides high-quality data for subsequent accurate analysis of tumor morphology, metabolism and other characteristics, significantly improving the reference value of imaging data in tumor diagnosis and treatment plan formulation.

[0036] Genetic data is transmitted with professional genetic testing institutions using the secure HTTPS protocol. Gene sequence data is stored in FASTQ format. During the collection process, the sequencing quality indicator base quality score Q = -10log is accurately recorded. 10 P (P is the probability of base misidentification, the higher the Q value, the higher the sequencing accuracy), and the coverage of whole genome sequencing (WGS) data (L covered is the length of the genome covered by sequencing, L genome The system integrates high-quality genetic data collection (D) and detailed quality metrics, including the total genome length (the length of the entire genome) and depth D (calculated by dividing the sum of all base sequenced times by the genome length). This provides a solid foundation for in-depth molecular research into tumor characteristics and chemoradiotherapy responses, and helps identify more targeted tumor treatment targets and biomarkers. The system is deeply integrated with the Radiotherapy Planning System (TPS) to comprehensively capture detailed radiotherapy parameters.

[0037] For intensity modulated radiotherapy (IMRT), accurately record the angle θ of each irradiation field i , weight w i 、Number of sub-fields n i And the area A of each sub-field ij , and use the Monte Carlo simulation algorithm to accurately calculate the radiotherapy dose distribution, simulating the transport process of a large number of particles in human tissue to obtain the absorbed dose D of each voxel voxel In terms of chemotherapy data, the pharmacokinetic parameters of chemotherapy drugs, such as drug half-life t, are accurately obtained from the hospital pharmacy management system and electronic medical record system. 1 / 2 , clearance rate CL, etc., and according to the formula Calculate the cumulative dose of each drug in the combination chemotherapy regimen (D i is the i-th medication dose, f i The above-mentioned treatment process data collection method provides a key basis for evaluating the implementation effect of the treatment plan and subsequent optimization, and effectively guarantees the guiding role of treatment process data in tumor treatment.

[0038] Secondly, the data acquisition module is connected to the hospital information system through a standardized interface to regularly collect information on the patient's past medical history and family medical history in the medical record. It obtains and converts DICOM format image data from imaging devices of different brands and models into the unified format of the system through a special data conversion and adaptation program. It receives gene sequence data and gene expression data generated by gene detection instruments in real time through a network communication interface, and collects treatment parameter data such as radiotherapy dose, number of radiotherapy sessions, types of chemotherapy drugs, and dosage recorded by radiotherapy and chemotherapy devices;

[0039] For missing values in clinical data, a multiple imputation method based on random forest is adopted. A random forest consists of multiple decision trees. Let the feature where the missing value is located be Y, and other features be X1, X2, …, X m ; First, divide the data into two parts: with and without missing values. Train a random forest model Y = RF(X1, X2, …, X m ) on the data without missing values, and then use the trained model to predict the data with missing values multiple times (randomly perturbing the input features each time), and obtain the final result by synthesizing multiple imputed values; Through this method, the complex relationships between a large number of features can be used to accurately impute missing values, greatly reducing the negative impact of data missing on subsequent analysis, and improving the integrity and reliability of clinical data. For outliers, they are identified using the Local Outlier Factor (LOF) algorithm; Calculate the local reachability density of each data point p and the local outlier factor N k (p) is the set of k-nearest neighbors of p, reach-dist k (p, o) is the reachable distance from p to o. When LOF(p) > 2, it is determined as an outlier and corresponding processing is performed, effectively avoiding the interference of outliers on data analysis; For numerical data normalization, the Min-Max normalization formula is used where X is the original value, X min 、X max are the maximum and minimum values of the feature, making the data of different features comparable and significantly improving the usability of clinical data in subsequent analysis and modeling.

[0040] Non-local means filtering (NLM) algorithm is used for CT image denoising. Let the gray value of pixel i in the image be I i , and after filtering it is weight (Ω is the set of image pixels, N i 、N jis the pixel neighborhood, and h controls the filtering intensity), effectively removing image noise and greatly improving image quality. Image segmentation uses the MaskR-CNN model based on deep learning. This model adds a segmentation branch on the basis of Faster R-CNN and can accurately identify the tumor area and generate a segmentation mask through training with a large number of labeled image data, providing strong support for subsequent accurate extraction of image features; The PET-CT image registration uses a registration algorithm based on mutual information to maximize the mutual information between PET and CT images (P PET 、P CT are the probability distributions of PET and CT images respectively, and P PET,CT (i,j) is the joint probability distribution) to achieve accurate registration, further improving the accuracy and consistency of image data in tumor analysis.

[0041] Next, in terms of clinical feature extraction, in addition to basic features, the change in tumor size is calculated to reflect the disease progression speed. For example, if the sizes of the tumor at times t1 and t2 are S1 and S2, then provides a key reference for treatment plan selection; By recording the number of times the patient takes medicine on time n on-time and the total number of times of taking medicine n total , the treatment compliance is calculated fully considering its important impact on the treatment effect. Lifestyle features are extracted, such as the smoking index SI = the number of cigarettes smoked per day × the number of years of smoking. The clinical features comprehensively reflect the patient's physical condition and treatment compliance, providing multi-angle information support for subsequent analysis and prediction.

[0042] In addition to morphological and texture features, the box-counting method is used to calculate the fractal dimension of the tumor in CT images (∈ is the box size, and N(∈) is the number of boxes containing the tumor boundary), describing the complexity of the tumor boundary; The tumor metabolic heterogeneity feature entropy is extracted from PET-CT images (p i is the metabolic probability of the i-th voxel in the tumor region, calculated through the SUV value), accurately depicting the uneven degree of tumor metabolism. These unique image features can more accurately describe the morphology, metabolism, and boundary of the tumor, significantly improving the depth and accuracy of image data in tumor analysis;

[0043] When extracting gene features, the gene mutation frequency is calculated (n var is the number of samples with mutations, n total(where [Total Sample Size] is the total number of samples, reflecting the degree of gene variation in the tumor population. The enrichment score ES is calculated using the Gene Set Enrichment Analysis (GSEA) method to determine whether a predefined gene set is significantly enriched in the sorted gene list, and to explore important biological processes and signaling pathways related to tumors. A Bayesian network algorithm is used to construct a gene regulatory network, and by calculating the conditional probability relationships between genes, the interaction relationships between genes are revealed. These in-depth gene features help to thoroughly reveal the mechanism of tumor occurrence and development at the molecular level, providing more accurate molecular targets and biomarkers for tumor treatment. Feature selection uses a genetic algorithm-based method, randomly generating a set of feature subsets as the initial population. Each feature subset is represented by a binary vector (1 indicates selection of the feature, 0 indicates non-selection). A fitness function F based on the classification accuracy of a classifier (such as a support vector machine) is defined. In each generation, a new population is generated through selection, crossover, and mutation operations until the termination condition is met (such as reaching the maximum number of iterations). Finally, the feature subset with the highest fitness is selected. This method effectively screens out the most representative and predictive feature subsets, significantly reducing the data dimension, and remarkably improving the training efficiency and prediction accuracy of subsequent models, laying a solid foundation for constructing an efficient and accurate prediction model.)

[0044] In addition, a multi-layer perceptron (MLP) model consists of an input layer, multiple hidden layers, and an output layer. The n neurons in the input layer correspond to n features, the m neurons in the l-th hidden layer l neurons, and the k neurons in the output layer correspond to k classification results. The input to the j-th neuron in the l-th hidden layer is the weight, is the output of the previous layer, is the bias), and the output is calculated through the ReLU activation function The output layer uses the Softmax activation function to convert the output into a probability distribution. Model training uses the Stochastic Gradient Descent (SGD) algorithm to update the weights and biases by minimizing the cross-entropy loss function (N is the number of samples, y i k * is the true label, and y i k is the predicted probability).

[0045] This model can effectively process multi-dimensional feature data and is suitable for predicting chemoradiotherapy responses by integrating multi-source features, showing excellent prediction ability in complex data environments. The convolutional neural network model is for image data and includes a convolutional layer, a pooling layer, and a fully connected layer. In the convolutional layer, the convolution kernel K convolves with the input image I to obtain a feature map Among them, M and N are the sizes of the convolution kernels. The pooling layer (such as max pooling) reduces the size of the feature map, and the fully connected layer integrates features for classification prediction. When the CNN model processes image data, it can automatically learn and extract key features in the image, especially showing excellent performance in identifying the morphology, structure, etc. of tumors, greatly improving the accuracy of predicting the radiotherapy and chemotherapy responses of tumors based on image data. To select the optimal model, the cross-validation method is used to evaluate the MLP, CNN, and RNN models. The dataset is divided into k non-overlapping subsets. Each time, k - 1 subsets are used for training and 1 subset is used for validation. The average accuracy, recall rate, F1 value, and other metrics are calculated after repeating k times.

[0046] It should be noted that when the prediction analysis module predicts the radiotherapy and chemotherapy response, it outputs the classification results of complete remission, partial remission, stable disease, and disease progression of the tumor and the probability values of each category, details the common adverse reactions such as nausea, vomiting, hair loss, and myelosuppression, and grades the occurrence probability of each adverse reaction as low probability, medium probability, and high probability. The Bootstrap method is used to construct a model by sampling the training data to calculate the variance of the prediction result to measure the confidence level. According to a large amount of clinical research data and historical case analysis, risk thresholds are set for different prediction indicators. For the prediction of radiotherapy and chemotherapy adverse reactions, such as the occurrence probability of nausea and vomiting, thresholds P nausea-threshold and P vomiting-threshold are set. When the predicted occurrence probability of nausea P nausea >P nausea-threshold or the occurrence probability of vomiting P vomiting >P vomiting-threshold , it is determined as high risk. For the prediction of treatment effect, such as the probability of tumor progression P progression , a threshold P progression-threshold is set. When P progression >P progression-threshold , it is considered that the treatment plan may have poor effects and there is a relatively high risk of tumor progression. Reasonable risk threshold setting can accurately identify potential high-risk situations based on historical experience and clinical evidence, providing a reliable basis for subsequent early warning and intervention.

[0047] Finally, the warning module sets risk thresholds for different tumor types and chemoradiotherapy regimens based on a large amount of clinical research data and authoritative medical guidelines. If the predicted high probability of nausea and vomiting exceeds 50%, the predicted probability of disease progression in the treatment effect exceeds 30%, and the situation exceeds the threshold, a message is pushed to the work terminal of the attending doctor through the hospital's internal information system and a warning message is sent to the monitoring system of the nurse station. The feedback module collects the feedback information on the difference between the predicted result and the actual treatment effect during the treatment process by doctors, as well as the feedback information on the patient's subjective feelings. The collected feedback information is classified and sorted, and statistical analysis methods are used to analyze the key data characteristics in the feedback information. The analysis results are used to update and optimize the parameters and structure of the prediction model to improve the prediction accuracy and practicality of the system. During the patient's treatment process, feedback information is collected through multiple channels. After each ward round, the doctor records the actual treatment effect of the patient, the occurrence of adverse reactions, and the evaluation of the prediction result through the electronic medical record system. The patient can feedback their subjective feelings, such as pain level, fatigue, appetite changes, etc., through the hospital's online questionnaire system or mobile application.

[0048] At the same time, the system also automatically collects the actual treatment parameters recorded by the treatment equipment, such as the actual output of radiotherapy dose, the actual infusion volume of chemotherapy drugs, etc. The comprehensive and multi-channel feedback information collection method can obtain the real situation of the patient's treatment process from different angles and provide rich data sources for subsequent analysis and model optimization.

[0049] Then, during the patient's treatment process, feedback information is collected through multiple channels. After each ward round, the doctor records the actual treatment effect of the patient, the occurrence of adverse reactions, and the evaluation of the prediction result through the electronic medical record system. The patient can feedback their subjective feelings, such as pain level, fatigue, appetite changes, etc., through the hospital's online questionnaire system or mobile application. At the same time, the system also automatically collects the actual treatment parameters recorded by the treatment equipment, such as the actual output of radiotherapy dose, the actual infusion volume of chemotherapy drugs, etc. The comprehensive and multi-channel feedback information collection method can obtain the real situation of the patient's treatment process from different angles and provide rich data sources for subsequent analysis and model optimization.

[0050] The collected feedback information is classified and sorted, and data mining and machine learning methods are used for analysis. For the evaluation information of doctors, text mining technology is used to extract key information, such as the deviation direction and degree between the prediction result and the actual situation. For the subjective feeling information of patients, sentiment analysis methods are used to judge the patient's emotional state and the severity of symptoms. For the actual treatment parameter information, the difference between it and the parameters used in the prediction model is analyzed. The in-depth analysis method can extract valuable content from the massive feedback information and accurately find out the problems and deficiencies of the prediction model.

[0051] The above embodiments can be implemented in whole or in part by software, hardware, firmware, and any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions and computer programs. When the computer instructions and computer programs are loaded and executed on a computer, the processes and functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, and data center to another website, computer, server, and data center via wired (such as infrared, wireless, microwave) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a server, data center data storage device that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0052] It should be understood that in various embodiments of the present application, the sequence numbers of the above processes do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0053] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0054] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0055] In addition, in each embodiment of the present application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0056] When the above-mentioned functions are implemented in the form of software functional units and sold and used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs, and various media that can store program codes.

[0057] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes and substitutions, which should all be covered within the protection scope of this application.

Claims

1. An automatic prediction system for tumor radiotherapy and chemotherapy reactions, characterized in that, It includes a data acquisition module, a data preprocessing module, a feature extraction and selection module, a model training module, a prediction and analysis module, an early warning module, and a feedback module, where: Data acquisition module: Collects patients' clinical data, imaging data, and genetic data, as well as relevant data during the treatment process, from the hospital information system, imaging devices, and genetic testing instruments; Data preprocessing module: Cleans the collected raw data, normalizes numerical data, and performs natural language processing on text data; Feature extraction and selection module: Extracts clinical, imaging, and genetic-related features from the preprocessed data, and uses feature selection algorithms to screen features; Model training module: Utilizes the screened feature data, adopts machine learning algorithms to train models based on historical patient data, and optimizes parameters through cross-validation; Prediction and analysis module: Inputs new patient data into the trained model to predict the radiotherapy and chemotherapy response, and conducts confidence evaluation; Early warning module: Compares the prediction results with a pre-set risk threshold, and issues an early warning signal if the results exceed the threshold; Feedback module: Collects feedback information during the treatment process, analyzes and organizes it for updating and optimizing the prediction model.

2. The automatic prediction system for tumor radiotherapy and chemotherapy reactions according to claim 1, wherein: The data acquisition module is connected to the hospital information system through a standardized interface to regularly collect information on the patient's past medical history and family medical history in the medical record. It obtains and converts DICOM format imaging data from imaging devices of different brands and models into the unified format of the system through a special data conversion and adaptation program, and receives gene sequence data and gene expression data generated by genetic testing instruments in real time through a network communication interface. It collects treatment parameter data such as radiotherapy dose, radiotherapy times, chemotherapy drug types, and usage doses recorded by radiotherapy and chemotherapy devices.

3. The automatic prediction system for tumor radiotherapy and chemotherapy reactions according to claim 1, characterized in that: The data preprocessing module uses data mining algorithms to clean duplicate data records according to the established data rule library, corrects incorrect data based on data logical relationships and medical knowledge rules, fills in missing values using the multiple imputation method combined with the data characteristics of similar patients, maps numerical data to the [0, 1] interval using the maximum-minimum normalization method, and converts text data into vector form using the bag-of-words model in natural language processing and performs part-of-speech tagging and named entity recognition operations to extract key medical terms and relevant information.

4. The automatic tumor radiotherapy and chemotherapy reaction prediction system according to claim 1, characterized in that: The feature extraction and selection module extracts clinical features, including age, gender, tumor stage, body mass index, white blood cell count, red blood cell count, platelet count in blood routine, alanine aminotransferase, aspartate aminotransferase, and serum creatinine related to liver and kidney function. It extracts imaging features, including the three-dimensional size of the tumor, shape irregularity, internal texture roughness, and contrast, using image segmentation algorithms and texture analysis algorithms. It extracts gene features such as the mutation type and gene copy number variation of genes related to radiotherapy and chemotherapy sensitivity, such as TP53 and BRCA1, and adopts the recursive feature elimination algorithm combined with the feature importance evaluation of the random forest model to screen the feature subset that contributes the most to the prediction of radiotherapy and chemotherapy response.

5. The automatic prediction system for tumor radiotherapy and chemotherapy reactions according to claim 1, wherein: The model training module preferentially uses a convolutional neural network to construct a model structure for data with image features, and sets multiple convolutional layers and pooling layers to extract deep-level image features. For data with sequence features, it uses the long short-term memory network model in the recurrent neural network. The training data is randomly divided according to the ratio of 70% as the training set, 15% as the validation set, and 15% as the test set. The five-fold cross-validation method is used to adjust the learning rate, number of iterations, and number of hidden layer nodes of the model during the training process, and the hyperparameters are used to make the model achieve the best prediction performance on the test set.

6. The automatic prediction system for tumor radiotherapy and chemotherapy reactions according to claim 1, wherein: When the prediction and analysis module predicts the chemoradiotherapy response, it outputs the classification results of complete remission, partial remission, stable disease, and disease progression of the tumor and the probability values of each category, details the common adverse reactions such as nausea, vomiting, hair loss, and myelosuppression, and grades the occurrence probability of each adverse reaction as low probability, medium probability, and high probability. The Bootstrap method is used to construct a model by sampling the training data to calculate the variance of the prediction result to measure the confidence level.

7. An automatic prediction system for tumor radiotherapy and chemotherapy reactions according to claim 1, characterized in that: The warning module sets risk thresholds for different tumor types and chemoradiotherapy regimens based on a large amount of clinical research data and authoritative medical guidelines. If the predicted high-probability occurrence of nausea and vomiting exceeds 50%, the predicted probability of disease progression in the treatment effect exceeds 30%, and the situation exceeds the threshold, a message is pushed to the work terminal of the attending doctor through the hospital internal information system and a warning message is sent to the monitoring system of the nurse station.

8. The automatic prediction system for tumor radiotherapy and chemotherapy reactions according to claim 1, characterized in that: The feedback module collects the feedback information on the difference between the prediction result and the actual treatment effect during the treatment process by doctors, as well as the feedback information on the patients' subjective feelings. The collected feedback information is classified and sorted, and statistical analysis methods are used to analyze the key data features in the feedback information. The analysis results are used to update and optimize the parameters and structure of the prediction model to improve the prediction accuracy and practicality of the system.

Images