A method and system for constructing a thyroid tumor evaluation model
Patent Information
- Application Number
- CN202411638064.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-16
- Publication Date
- 2026-10-09
AI Technical Summary
然而，现有的评估模型往往仅基于单一的检测手段(如超声成像或细胞学检查)，缺乏多模态数据的综合分析和利用，导致评估结果的准确性和可靠性有待提高
[0047]提高评估准确性：通过整合血浆检测样本的细胞学检查数据和多模态超声成像数据，利用机器学习和深度学习技术进行综合分析，提高了甲状腺肿瘤评估的准确性；
Smart Images
Figure CN122889348A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical information processing technology, and in particular relates to a method and system for constructing a thyroid tumor assessment model. Background Technology
[0002] Thyroid tumors are common endocrine system tumors, and accurate differentiation between benign and malignant tumors is crucial for developing appropriate treatment plans. Traditional thyroid tumor assessment mainly relies on the physician's clinical experience and pathological examination results, but this method suffers from high subjectivity, is time-consuming, and has limited accuracy. In recent years, with the development of medical informatics and artificial intelligence technologies, constructing thyroid tumor assessment models using big data and machine learning techniques has become a research hotspot. However, existing assessment models are often based on only a single detection method (such as ultrasound imaging or cytology examination), lacking comprehensive analysis and utilization of multimodal data, resulting in a need to improve the accuracy and reliability of the assessment results. Summary of the Invention
[0003] This invention proposes a method for constructing a thyroid tumor assessment model. This method integrates cytological examination data from plasma test samples and multimodal ultrasound imaging data, and utilizes machine learning and deep learning technologies to perform feature extraction, feature selection and optimization, model construction and training, and model validation and optimization, thereby constructing an efficient and accurate thyroid tumor assessment model.
[0004] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:
[0005] This invention relates to a method for constructing a thyroid tumor assessment model, comprising:
[0006] S1. Data collection and preprocessing: Cytological examination of the patient's plasma test sample is performed to obtain examination data, and multimodal ultrasound imaging data is obtained through ultrasound images. The data is then preprocessed.
[0007] S2. Feature Extraction: Machine learning algorithms are used to extract features from the preprocessed examination data and ultrasound imaging data. For ultrasound images, deep learning technology is used for feature extraction.
[0008] S3. Feature Selection and Optimization: The extracted features are screened through big data analysis and machine learning algorithms to retain features that are highly relevant to the differentiation of benign and malignant thyroid tumors, and the screened features are optimized.
[0009] S4. Weight coefficient determination: Assign a weight coefficient to each feature based on its importance;
[0010] S5. Model Building and Training: Using the selected and optimized features and their weight coefficients as input, an appropriate machine learning algorithm is selected to build a thyroid tumor assessment model, and cross-validation technology is used to train and validate the model.
[0011] S6. Model Validation and Optimization: Validate the trained model using an independent test set, verify the model's accuracy through comparative experiments, and fine-tune the model based on the evaluation results;
[0012] S7. Model Deployment and Application: Deploy the optimized model into the thyroid tumor assessment system to achieve automated assessment.
[0013] As a preferred embodiment of the present invention, step S1 specifically includes the following sub-steps:
[0014] Plasma samples from patients are collected for cytological examination to obtain data, including but not limited to the expression levels of specific genes, mutation status, and other biomarkers related to thyroid tumors.
[0015] Collect multimodal ultrasound imaging data from patients, including two-dimensional grayscale ultrasound, color Doppler ultrasound, and elastography;
[0016] The collected data is preprocessed, including data cleaning, normalization, noise reduction, and missing value handling, to ensure data quality.
[0017] As a preferred embodiment of the present invention, the specific machine learning algorithm used in step S5 is as follows:
[0018] Convolutional Neural Networks (CNNs) are used to process ultrasound image data;
[0019] Recurrent neural networks (RNNs) are used to process examination data collected from plasma test samples.
[0020] As a preferred embodiment of the present invention, the thyroid tumor assessment model constructed in step S5 includes the following architecture:
[0021] Input layer: The input layer is provided with multiple input channels, each accepting different data types;
[0022] Feature extraction layer: For ultrasound images, convolutional and pooling layers of CNN are used for feature extraction; for time series data, hidden layers of RNN are used for feature learning.
[0023] Fusion layer: Fuses features extracted from different channels to form a unified feature representation;
[0024] Output layer: Design the corresponding output layer according to the requirements of the evaluation task.
[0025] As a preferred embodiment of the present invention, step S5 further includes a data augmentation step during model training to improve the generalization ability of the model; at the same time, a suitable loss function and optimization algorithm are used for model training, and the loss function value and performance indicators are monitored during the training process to avoid overfitting and underfitting.
[0026] As a preferred embodiment of the present invention, step S5 further includes the following step:
[0027] The system includes a self-learning module that can periodically acquire new data for self-training; it also incorporates transfer learning technology to quickly adapt to new data, improve update efficiency, and set update cycles and strategies to ensure that the model can reflect the latest clinical data and research findings in a timely manner.
[0028] The present invention also provides a thyroid tumor assessment system, comprising:
[0029] The data acquisition module is used to collect cytological examination data and multimodal ultrasound imaging data from plasma test samples.
[0030] The data preprocessing module is used to preprocess the collected data;
[0031] The feature extraction module is used to extract features from the preprocessed data. For ultrasound images, deep learning technology is used for feature extraction.
[0032] The feature selection and optimization module is used to filter and optimize the extracted features;
[0033] The model building and training module is used to build and train thyroid tumor assessment models;
[0034] The model evaluation and optimization module is used to evaluate and optimize the trained model;
[0035] The results display module is used to show the assessment results of thyroid tumors.
[0036] As a preferred embodiment of the present invention, the feature extraction module further includes:
[0037] The deep learning model library submodule stores a variety of deep learning models, including but not limited to convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants, to adapt to different types of data feature extraction needs;
[0038] The feature fusion submodule is used to fuse features extracted from different data sources to form a comprehensive feature vector for use by subsequent models.
[0039] As a preferred embodiment of the present invention, the model evaluation and optimization module further includes:
[0040] The performance evaluation submodule uses multiple performance metrics to comprehensively evaluate the model, ensuring its effectiveness in practical applications.
[0041] The model iteration and optimization submodule automatically or manually adjusts model parameters, feature sets, or algorithm selection based on evaluation results to continuously improve model performance.
[0042] As a preferred embodiment of the present invention, it further includes a user interface module, specifically comprising:
[0043] The user authentication submodule ensures that only authorized users can access the system.
[0044] The visualization report generation submodule presents the assessment results to users in the form of intuitive charts and reports, including the predicted classification of thyroid tumors, confidence levels, and possible follow-up recommendations.
[0045] The user feedback collection submodule allows users to provide feedback on evaluation results and the system in order to continuously optimize and improve the system.
[0046] The present invention has the following beneficial effects:
[0047] Improving assessment accuracy: By integrating cytological examination data from plasma test samples and multimodal ultrasound imaging data, and using machine learning and deep learning technologies for comprehensive analysis, the accuracy of thyroid tumor assessment has been improved.
[0048] Improved assessment efficiency: The automated assessment of thyroid tumors has been achieved, which has greatly shortened the assessment time and improved the assessment efficiency.
[0049] Enhanced model generalization ability: Through techniques such as data augmentation, cross-validation, autonomous learning, and transfer learning, the model's generalization ability has been enhanced, enabling it to adapt to different datasets and clinical scenarios;
[0050] Promote continuous model optimization: continuously iterate and optimize the model based on user feedback and performance evaluation results to continuously improve model performance.
[0051] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0053] Figure 1 This is a schematic diagram of the method flow of the present invention;
[0054] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation
[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0056] Example 1
[0057] like Figure 1 As shown: This invention provides a method for constructing a thyroid tumor assessment model, including the following steps:
[0058] S1. Data collection and preprocessing: Cytological examination of the patient's plasma test samples is performed to obtain examination data. At the same time, multimodal ultrasound imaging data is obtained through ultrasound images, and the data is preprocessed.
[0059] S2. Feature Extraction: Machine learning algorithms are used to extract features from the preprocessed examination data and ultrasound imaging data. For ultrasound images, deep learning technology is used for feature extraction.
[0060] S3. Feature Selection and Optimization: The extracted features are screened through big data analysis and machine learning algorithms, retaining features that are highly relevant to the differentiation of benign and malignant thyroid tumors, and the screened features are optimized.
[0061] S4. Weight coefficient determination: Assign a weight coefficient to each feature based on its importance;
[0062] S5. Model Building and Training: Using the selected and optimized features and their weight coefficients as input, an appropriate machine learning algorithm is selected to build a thyroid tumor assessment model, and cross-validation technology is used to train and validate the model.
[0063] S6. Model Validation and Optimization: Validate the trained model using an independent test set, verify the model's accuracy through comparative experiments, and fine-tune the model based on the evaluation results;
[0064] S7. Model Deployment and Application: Deploy the optimized model into the thyroid tumor assessment system to achieve automated assessment.
[0065] This embodiment describes in detail the method for constructing the thyroid tumor assessment model provided by the present invention, and its steps and key points are as follows:
[0066] S1. Data Collection and Preprocessing:
[0067] Objective: To collect and prepare the basic data for model building.
[0068] Procedure: Cytological examination data: obtained through plasma test samples, including biomarker information such as the expression level and mutation status of specific genes, which is crucial for understanding the mechanisms of tumor occurrence and development.
[0069] Multimodal ultrasound imaging data includes two-dimensional grayscale ultrasound, color Doppler ultrasound, and elastography. These techniques can reflect the structural and functional characteristics of thyroid tumors from different perspectives.
[0070] Data preprocessing: a key step in ensuring data quality, including data cleaning (removing invalid or erroneous data), normalization (making different features comparable on the same scale), noise reduction (reducing random errors in the data), and missing value handling (filling in or deleting missing data).
[0071] S2, Feature Extraction:
[0072] Objective: To extract valuable features for the assessment of thyroid tumors from preprocessed data.
[0073] Procedure: For cytology data, machine learning algorithms are used for feature extraction. For ultrasound imaging data, deep learning technology is used for more refined feature extraction, such as tumor edge features and internal echo patterns.
[0074] S3. Feature Selection and Optimization:
[0075] Objective: To screen out and optimize features that are highly relevant to the differentiation between benign and malignant thyroid tumors.
[0076] Operation: The extracted features are filtered using big data analytics and machine learning algorithms (such as feature selection algorithms and correlation analysis) to retain those most valuable for tumor assessment. Simultaneously, the filtered features are optimized to improve their accuracy and robustness.
[0077] S4. Determining the weighting coefficients:
[0078] Objective: To assign a weight coefficient to each feature to reflect its importance in the assessment of thyroid tumors.
[0079] Operation: Assign a weight coefficient to each feature based on its importance (e.g., obtained through a feature importance evaluation algorithm). These weight coefficients will be used in subsequent model building.
[0080] S5. Model Building and Training:
[0081] Objective: To construct a thyroid tumor assessment model using screened and optimized features and their weighting coefficients.
[0082] Operation: Convolutional Neural Network (CNN) is used to process ultrasound image data, leveraging its powerful image processing capabilities to extract image features;
[0083] Use recurrent neural networks (RNNs) to process time-series data (such as cytology examination data) and capture temporal dependencies in the data;
[0084] The constructed thyroid tumor assessment model has multiple input channels, enabling it to process different types of data simultaneously, and integrates these features into a unified feature representation through a fusion layer.
[0085] S6. Model Validation and Optimization:
[0086] Objective: To verify the accuracy of the model and to fine-tune the model based on the evaluation results.
[0087] Procedure: Validate the trained model using an independent test set, and evaluate its accuracy, sensitivity, and specificity through comparative experiments. Based on the evaluation results, fine-tune the model as needed to improve its performance.
[0088] S7. Model Deployment and Application:
[0089] Objective: To deploy the optimized model into a thyroid tumor assessment system to achieve automated assessment.
[0090] Operation: The trained and optimized model is integrated into the thyroid tumor assessment system, which can automatically assess the patient's thyroid tumors and provide doctors with accurate diagnostic support.
[0091] In summary, this embodiment provides a comprehensive and systematic method for constructing a thyroid tumor assessment model. Through steps such as data collection and preprocessing, feature extraction, feature selection and optimization, determination of weight coefficients, model construction and training, model validation and optimization, and model deployment and application, accurate assessment of thyroid tumors is achieved.
[0092] Example 2
[0093] Based on Embodiment 1, the difference in this embodiment is:
[0094] In step S5, the model training process also includes a data augmentation step to improve the model's generalization ability; at the same time, an appropriate loss function and optimization algorithm are used for model training, and the loss function value and performance indicators are monitored during the training process to avoid overfitting and underfitting.
[0095] Step S5 also includes the following steps: setting up a self-learning module that can periodically acquire new data for self-training; introducing transfer learning technology that can quickly adapt to new data and improve update efficiency; and setting update cycles and update strategies to ensure that the model can reflect the latest clinical data and research results in a timely manner.
[0096] This embodiment further optimizes and expands upon Embodiment 1, particularly in model training and data updating, to improve the generalization ability and update efficiency of the thyroid tumor assessment model. The following is a detailed explanation of the additions and improvements in this embodiment:
[0097] Optimization in Step S5
[0098] Data augmentation: Data augmentation is a technique that increases the diversity of a dataset by applying random transformations (such as rotation, scaling, translation, flipping, etc.). During model training, data augmentation of the original data can generate more training samples, thereby helping the model learn richer feature representations and improving the model's generalization ability.
[0099] Loss Function and Optimization Algorithm: The loss function measures the difference between the model's predictions and the actual results. During model training, a suitable loss function needs to be selected to guide the model's optimization direction. Optimization algorithms are used to update the model's parameters to minimize the loss function value. Commonly used optimization algorithms include stochastic gradient descent (SGD) and Adam. Monitoring the loss function value and performance metrics (such as accuracy, specificity, and sensitivity) during training is crucial for avoiding overfitting and underfitting. By continuously adjusting hyperparameters such as the learning rate and regularization parameters, the model training process can be optimized, improving model performance.
[0100] The self-learning module is configured to periodically acquire new data from clinical databases or other reliable data sources for self-training. By continuously introducing new data, the model can learn more feature representations, improving its prediction accuracy and generalization ability. The self-learning module can also be combined with online learning technologies to achieve real-time model updates and optimization.
[0101] Introducing Transfer Learning: Transfer learning is a technique that transfers knowledge learned on one task to another. In the updating process of a thyroid tumor assessment model, introducing transfer learning can quickly adapt to new data and improve the model's update efficiency. Through transfer learning, the model can utilize feature representations pre-trained on large amounts of data to quickly fine-tune on new datasets, thereby saving training time and computational resources.
[0102] Setting update cycles and strategies: To ensure the model reflects the latest clinical data and research findings in a timely manner, it is necessary to set reasonable update cycles and strategies. For example, the model can be updated regularly (e.g., quarterly or annually), or an update can be triggered when a certain amount of new data is available.
[0103] This embodiment further improves the generalization ability and update efficiency of the thyroid tumor assessment model by introducing data augmentation, selecting appropriate loss functions and optimization algorithms, setting up a self-learning module, and introducing transfer learning techniques. These optimization measures enable the model to better adapt to different datasets and clinical scenarios, providing doctors with more accurate and reliable diagnostic support. Simultaneously, by regularly updating the model, it is ensured that it can promptly reflect the latest clinical data and research findings, providing stronger technical support for the early diagnosis and personalized treatment of thyroid tumors.
[0104] like Figure 2 As shown: This invention also provides a comprehensive thyroid tumor assessment system, aiming to improve the accuracy and efficiency of thyroid tumor assessment by integrating multiple technologies. The following is a detailed analysis of the system's functions and features:
[0105] Data acquisition module: Collects cytological examination data and multimodal ultrasound imaging data from plasma test samples. Cytological examination can provide information on abnormalities at the cellular level, while multimodal ultrasound imaging can display the structure, morphology, and blood flow characteristics of tumors. The combination of the two provides more comprehensive information for tumor assessment.
[0106] Data preprocessing module: Cleans, formats, and standardizes the collected data. Preprocessing is the foundation for subsequent analysis, ensuring data quality and reducing the impact of noise and outliers on model training.
[0107] Feature extraction module: Extracts key features from preprocessed data, using deep learning techniques for feature extraction from ultrasound images. This includes a deep learning model library submodule: storing various deep learning models, such as CNN and RNN, to adapt to different types of data feature extraction needs; and a feature fusion submodule: fusing features from different data sources to form a comprehensive feature vector, improving the model's generalization ability.
[0108] Feature selection and optimization module: Filters and optimizes the extracted features, removes redundant and noisy features, and retains the features most valuable for tumor assessment.
[0109] Model building and training module: This module builds and trains a thyroid tumor assessment model, and validates it to ensure its accuracy and stability. Model building is the core of the system, and the training and optimization processes directly affect the accuracy of the assessment results.
[0110] The model evaluation and optimization module comprehensively evaluates the trained model and adjusts model parameters, feature sets, or algorithm selection based on the evaluation results to continuously improve model performance. This includes a performance evaluation submodule, which evaluates the model using various performance metrics (such as accuracy, recall, and F1 score); and a model iterative optimization submodule, which automatically or manually adjusts the model based on the evaluation results.
[0111] Results Display Module: This module presents the thyroid tumor assessment results to users in an intuitive way. The results display serves as a bridge between the system and the user; an intuitive and clear presentation helps users understand and accept the assessment results.
[0112] The user interface module provides user authentication, visualization report generation, and user feedback collection functions. It includes a user authentication submodule to ensure system security and prevent unauthorized access; a visualization report generation submodule to present evaluation results in charts and reports, including predicted classifications, confidence levels, and follow-up recommendations; and a user feedback collection submodule to collect user feedback for system optimization and improvement.
[0113] In summary, the thyroid tumor assessment system of this embodiment integrates multiple technologies to achieve a complete process from data acquisition, preprocessing, feature extraction to model construction, validation, optimization, and result display. Simultaneously, the user interface module provides a safe and intuitive user experience and allows user feedback for continuous system optimization. This system is expected to provide strong support for the early detection and treatment of thyroid tumors.
[0114] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A method for constructing a thyroid tumor assessment model, characterized in that, Includes the following steps: S1. Data collection and preprocessing: Cytological examination of the patient's plasma test sample is performed to obtain examination data, and multimodal ultrasound imaging data is obtained through ultrasound images. The data is then preprocessed. S2. Feature Extraction: Machine learning algorithms are used to extract features from the preprocessed examination data and ultrasound imaging data. For ultrasound images, deep learning technology is used for feature extraction. S3. Feature Selection and Optimization: The extracted features are screened through big data analysis and machine learning algorithms, retaining features that are highly relevant to the differentiation of benign and malignant thyroid tumors, and the screened features are optimized. S4. Weight coefficient determination: Assign a weight coefficient to each feature based on its importance; S5. Model Building and Training: Using the selected and optimized features and their weight coefficients as input, an appropriate machine learning algorithm is selected to build a thyroid tumor assessment model, and cross-validation technology is used to train and validate the model. S6. Model Validation and Optimization: Validate the trained model using an independent test set, verify the model's accuracy through comparative experiments, and fine-tune the model based on the evaluation results; S7. Model Deployment and Application: Deploy the optimized model into the thyroid tumor assessment system to achieve automated assessment.
2. The method for constructing a thyroid tumor assessment model according to claim 1, characterized in that, Specifically, step S1 includes the following sub-steps: Plasma samples from patients are collected for cytological examination to obtain data, including but not limited to the expression levels of specific genes, mutation status, and other biomarkers related to thyroid tumors. Collect multimodal ultrasound imaging data from patients, including two-dimensional grayscale ultrasound, color Doppler ultrasound, and elastography; The collected data is preprocessed, including data cleaning, normalization, noise reduction, and missing value handling, to ensure data quality.
3. The method for constructing a thyroid tumor assessment model according to claim 1, characterized in that, In step S5, the specific machine learning algorithm used is: Convolutional Neural Networks (CNNs) are used to process ultrasound image data; Recurrent neural networks (RNNs) are used to process examination data collected from plasma test samples.
4. The method for constructing a thyroid tumor assessment model according to claim 3, characterized in that, In step S5, the constructed thyroid tumor assessment model includes the following architecture: Input layer: The input layer is provided with multiple input channels, each accepting different data types; Feature extraction layer: For ultrasound images, convolutional and pooling layers of CNN are used for feature extraction; for time series data, hidden layers of RNN are used for feature learning. Fusion layer: Fuses features extracted from different channels to form a unified feature representation; Output layer: Design the corresponding output layer according to the requirements of the evaluation task.
5. The method for constructing a thyroid tumor assessment model according to claim 4, characterized in that, In step S5, the model training process also includes a data augmentation step to improve the model's generalization ability; at the same time, a suitable loss function and optimization algorithm are used for model training, and the loss function value and performance indicators are monitored during the training process to avoid overfitting and underfitting.
6. The method for constructing a thyroid tumor assessment model according to claim 5, characterized in that, Step S5 also includes the following steps: The system includes a self-learning module that can periodically acquire new data for self-training; it also incorporates transfer learning technology to quickly adapt to new data, improve update efficiency, and set update cycles and strategies to ensure that the model can reflect the latest clinical data and research findings in a timely manner.
7. A thyroid tumor assessment system for implementing the method described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to collect cytological examination data and multimodal ultrasound imaging data from plasma test samples. The data preprocessing module is used to preprocess the collected data; The feature extraction module is used to extract features from the preprocessed data. For ultrasound images, deep learning technology is used for feature extraction. The feature selection and optimization module is used to filter and optimize the extracted features; The model building and training module is used to build and train thyroid tumor assessment models; The model evaluation and optimization module is used to evaluate and optimize the trained model; The results display module is used to show the assessment results of thyroid tumors.
8. A thyroid tumor assessment system according to claim 7, characterized in that, The feature extraction module further includes: The deep learning model library submodule stores a variety of deep learning models, including but not limited to convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants, to adapt to different types of data feature extraction needs; The feature fusion submodule is used to fuse features extracted from different data sources to form a comprehensive feature vector for use by subsequent models.
9. A thyroid tumor assessment system according to claim 7, characterized in that, The model evaluation and optimization module also includes: The performance evaluation submodule uses multiple performance metrics to comprehensively evaluate the model, ensuring its effectiveness in practical applications. The model iteration and optimization submodule automatically or manually adjusts model parameters, feature sets, or algorithm selection based on evaluation results to continuously improve model performance.
10. A thyroid tumor assessment system according to claim 7, characterized in that, It also includes a user interface module, specifically including: The user authentication submodule ensures that only authorized users can access the system. The visualization report generation submodule presents the assessment results to users in the form of intuitive charts and reports, including the predicted classification of thyroid tumors, confidence levels, and possible follow-up recommendations. The user feedback collection submodule allows users to provide feedback on evaluation results and the system in order to continuously optimize and improve the system.