Tumor pathology comprehensive diagnosis and prognosis evaluation system based on artificial intelligence
By utilizing an AI-based comprehensive tumor pathology diagnostic system with multi-source data and deep learning algorithms, the system addresses the issues of subjectivity and insufficient detection capabilities in existing tumor pathology diagnoses, achieving efficient and accurate tumor diagnosis and prognostic assessment.
Patent Information
- Application Number
- CN202511295378.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-11
- Publication Date
- 2025-12-02
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing tumor pathology diagnostic methods are highly subjective, have limited detection capabilities, and are highly invasive, leading to inconsistent diagnostic results and missed or misdiagnosed cases, making large-scale screening difficult.
An AI-based comprehensive tumor pathology diagnosis and prognostic assessment system is adopted. Through multi-source data acquisition, data preprocessing, feature extraction and selection, diagnostic model construction and prognostic assessment model construction, combined with deep learning algorithms and machine learning algorithms, features are automatically extracted and personalized reports are generated.
It has improved the accuracy and consistency of diagnosis, reduced human intervention, increased diagnostic efficiency and prognostic accuracy, provided a basis for personalized treatment plans, and enhanced the efficiency of medical resource utilization.
Smart Images
Figure CN121054232A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tumor pathology diagnosis and assessment technology, and in particular to a comprehensive tumor pathology diagnosis and prognostic assessment system based on artificial intelligence. Background Technology
[0002] Tumor pathology is the science that studies the nature, development, morphology, cellular structure, and functional metabolic changes of tumors. Tumors are new growths formed when the body's cells, under the long-term synergistic effect of internal and external tumorigenic factors, experience differentiation and regulation disorders at the gene level, resulting in excessive proliferation and abnormal differentiation. Tumor pathology diagnosis, as the gold standard for tumor diagnosis, clarifies the benign or malignant nature, type, grade, and stage of tumors through microscopic observation and analysis of tumor tissues, providing crucial information for clinical treatment decisions.
[0003] Tumor pathology diagnosis plays a crucial role in the diagnosis and treatment of tumors. First, it clarifies the nature of the tumor, distinguishing between benign and malignant tumors, which is decisive for the selection of treatment plans and prognosis. Second, pathological diagnosis can determine the type and grade of the tumor. Different types of tumors have significantly different responses to treatment and patient prognoses, and accurate pathological diagnosis helps to develop personalized treatment plans. In addition, tumor pathology diagnosis can assess the extent of tumor invasion and metastasis, providing important basis for determining the scope of surgical resection and prognostic assessment. Finally, with the development of molecular biology technology, molecular pathology diagnosis is gradually becoming an important part of precision oncology. By detecting and analyzing tumor-related genes, it can provide guidance for the selection of new treatment methods such as targeted therapy and immunotherapy.
[0004] Existing methods for tumor pathological diagnosis mainly include cytological diagnosis, tissue biopsy, immunohistochemistry, and molecular biology techniques. However, these methods still have the following shortcomings in practical operation:
[0005] 1) Highly subjective. Traditional pathological diagnosis methods mainly rely on the experience of pathologists and microscopic observation. Diagnostic results may differ between different pathologists, affecting the accuracy and consistency of diagnosis.
[0006] 2) Limited detection capabilities: Current technologies are insufficient for detecting early-stage tumors and small lesions, easily leading to false negatives and false positives, resulting in missed or misdiagnosed tumors.
[0007] 3) Invasive diagnostic methods can cause discomfort to patients and may even cause them to resist, and large-scale screening is difficult to conduct. Summary of the Invention
[0008] In view of the problems of existing tumor pathology diagnostic methods, such as strong subjectivity, limited detection capabilities, and discomfort caused to patients by invasive diagnostic methods, this invention is proposed.
[0009] Therefore, the purpose of this invention is to provide an artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology. Its purpose is to: comprehensively consider multiple information such as pathological images, clinical data, and genetic data, overcome the limitations of diagnosis from a single data source, improve the accuracy of diagnosis and prognostic assessment, and at the same time, automatically extract features and build models through deep learning algorithms, reduce human intervention, reduce the influence of subjective factors on diagnostic results, and improve diagnostic efficiency.
[0010] To address the aforementioned technical problems, this invention provides the following technical solution: an artificial intelligence-based comprehensive tumor pathology diagnosis and prognostic assessment system, including a multi-source data acquisition module, used to acquire patients' pathological images, clinical data, and genetic data from multiple channels such as hospital information systems and laboratory equipment, providing a comprehensive data foundation for subsequent analysis;
[0011] The data preprocessing module is used to clean the collected data, remove noise and outliers, and standardize and normalize the data to make it meet the model input requirements.
[0012] The feature extraction and selection module is used to extract tumor-related features from the preprocessed data and select the most representative and discriminative features through a feature selection algorithm, thereby reducing data dimensionality and improving model efficiency.
[0013] The diagnostic model building module is used to input feature vectors into the model for training, obtain diagnostic results that can accurately identify tumor type, grade, and stage, and evaluate and optimize the model.
[0014] The prognostic assessment model building module is used to predict prognostic indicators such as patient survival and recurrence risk, and generate personalized prognostic assessment reports based on individual patient characteristics.
[0015] The results display and interaction module is used to present diagnostic results and prognostic assessment reports to doctors and patients in an intuitive and easy-to-understand way, such as through image annotations, charts, and text descriptions, and provides interactive functions.
[0016] As a preferred embodiment of the artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the multi-source data acquisition module includes a pathological image acquisition submodule, a clinical data acquisition submodule, and a gene data acquisition submodule.
[0017] As a preferred embodiment of the AI-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the data preprocessing module includes a data cleaning submodule, a data standardization submodule, and a data normalization submodule.
[0018] The data normalization submodule normalizes the data, mapping it to a specific range, such as [0, 1] or [-1, 1], to eliminate the influence of data units and improve model training performance. For example, the Min-Max normalization method can be used, with the following formula:
[0019] =
[0020] Where x represents the original data. and These are the minimum and maximum values of the data, respectively.
[0021] As a preferred embodiment of the artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the feature extraction and selection module includes an image feature extraction submodule, a clinical feature extraction submodule, a gene feature extraction submodule, and a feature selection submodule.
[0022] As a preferred embodiment of the artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the diagnostic model construction module includes a model training submodule and a model evaluation submodule.
[0023] The model training submodule uses deep learning algorithms to construct a tumor pathology diagnostic model. Feature vectors are input into the model for training, resulting in accurate diagnostic results that identify tumor type, grade, and stage. Convolutional Neural Networks (CNNs) can be used for model training; the formula for their backpropagation algorithm is as follows:
[0024] = (( )⊙ ( )
[0025] in For the first Layer error term, For the first Layer weight matrix ( ) is the first The derivative of the activation function of the layer;
[0026] The model evaluation submodule evaluates the trained diagnostic model, calculating metrics such as accuracy, recall, and F1 score to assess the model's performance and generalization ability. The formula for calculating accuracy is: Accuracy =
[0027] TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative.
[0028] Recall calculation formula: Recall= Recall rate represents the proportion of samples that were actually positive but were correctly identified as positive.
[0029] As a preferred embodiment of the artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the prognostic assessment model construction module includes a survival analysis submodule, a machine learning model training submodule, and a risk assessment submodule.
[0030] The Kaplan-Meier estimation formula in the survival analysis submodule is used to estimate the survival function. The formula is as follows:
[0031] (t) = (1- ),in For the first Event time For the first The number of deaths in each event For the first The number of people at risk for each event;
[0032] The formula for Random Forest Survival Analysis (RF-SRC) in the machine learning model training submodule: Random Forest Survival Analysis predicts survival probability by constructing multiple decision trees. The formula for predicting survival probability is as follows:
[0033] P( >t∣x)=
[0034] in For the number of decision trees, For the first The survival probability predicted by the decision tree.
[0035] As a preferred embodiment of the AI-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the result display and interaction module includes a result visualization submodule and an interactive feedback submodule.
[0036] As a preferred embodiment of the AI-based comprehensive tumor pathology diagnosis and prognostic assessment system of the present invention, the feature selection submodule adopts a feature selection method based on genetic algorithms. By simulating the biological evolution process, it automatically searches for the optimal feature subset, thereby improving the efficiency and accuracy of feature selection. Compared with traditional feature selection methods, it can more effectively screen out the features most valuable for tumor diagnosis and prognostic assessment.
[0037] As a preferred embodiment of the AI-based comprehensive diagnosis and prognostic assessment system for tumor pathology described in this invention, the diagnostic model construction module employs a combination of data augmentation and model regularization during the training process. By applying data augmentation techniques to the training data to increase sample diversity, and simultaneously using model regularization techniques to constrain the complexity of the model, overfitting is prevented, thereby improving the stability and accuracy of the model in practical applications.
[0038] The beneficial effects of this invention are:
[0039] 1. This invention acquires various information such as pathological images, clinical data, and genetic data through a multi-source data acquisition module, providing a comprehensive data foundation for diagnosis. The feature extraction and selection module can extract tumor-related features from these data and select the most representative and discriminative features through a feature selection algorithm, reducing data dimensionality and improving the efficiency and accuracy of the model. The diagnostic model construction module uses deep learning algorithms to build a diagnostic model, which can automatically learn and extract features, reducing human intervention, reducing the impact of subjective factors on diagnostic results, and improving the accuracy and consistency of diagnosis.
[0040] 2. The prognostic assessment model construction module of this invention combines multi-source patient data and uses survival analysis methods and machine learning algorithms to construct a prognostic assessment model, which can predict prognostic indicators such as patient survival time and recurrence risk, and generate personalized prognostic assessment reports based on individual patient characteristics. This provides a more targeted basis for clinical treatment decisions and helps improve patients' survival rate and quality of life.
[0041] 3. This invention automates the processes of data collection, preprocessing, feature extraction and selection, model training and evaluation, reducing the time and workload of manual operations and improving diagnostic efficiency. At the same time, the system can quickly process large amounts of data, providing clinicians with timely and accurate diagnostic results and prognostic assessment reports, which helps to accelerate the treatment process and improve the utilization efficiency of medical resources. Attached Figure Description
[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the 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. Wherein:
[0043] Figure 1 This is a schematic diagram of the overall framework of the artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology of this invention. Detailed Implementation
[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0046] Reference Figure 1 The first embodiment of the present invention provides an artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology. This artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology includes a multi-source data acquisition module, which is used to acquire patients' pathological images, clinical data and genetic data from multiple channels such as hospital information systems and laboratory equipment, so as to provide a comprehensive data foundation for subsequent analysis.
[0047] The data preprocessing module is used to clean the collected data, remove noise and outliers, and standardize and normalize the data to make it meet the model input requirements.
[0048] The feature extraction and selection module is used to extract tumor-related features from the preprocessed data and select the most representative and discriminative features through a feature selection algorithm, thereby reducing data dimensionality and improving model efficiency.
[0049] The diagnostic model building module is used to input feature vectors into the model for training, obtain diagnostic results that can accurately identify tumor type, grade, and stage, and evaluate and optimize the model.
[0050] The prognostic assessment model building module is used to predict prognostic indicators such as patient survival and recurrence risk, and generate personalized prognostic assessment reports based on individual patient characteristics.
[0051] The results display and interaction module is used to present diagnostic results and prognostic assessment reports to doctors and patients in an intuitive and easy-to-understand way, such as through image annotations, charts, and text descriptions, and provides interactive functions.
[0052] The multi-source data acquisition module includes a pathological image acquisition submodule: acquiring patient pathological slide image data from the hospital's image archiving and communication system (PACS) to provide basic data for subsequent image analysis;
[0053] Clinical data acquisition submodule: Extracts patients' clinical data, such as age, gender, medical history, symptoms, and treatment plans, from hospital information systems (HIS), electronic medical record systems (EMR). This data helps to comprehensively understand the patient's condition and physical status, and provides important reference for diagnosis and prognosis assessment.
[0054] Gene data acquisition submodule: Acquires patient gene data from devices such as gene sequencers, including the expression levels and mutation status of tumor-related genes. Gene data can reveal the molecular characteristics of tumors and provide a basis for precision medicine.
[0055] The data preprocessing module includes a data cleaning submodule: cleaning the collected data to remove noise and outliers. For example, for pathological image data, median filtering and other methods can be used to remove noise from the images; for clinical data and genetic data, the integrity and consistency of the data can be checked, and erroneous data records can be deleted or corrected.
[0056] The data standardization submodule converts data from different sources and in different formats into a unified standard format to facilitate subsequent processing and analysis. It converts pathological image data into standard image formats, such as DICOM format, and converts clinical and genetic data into structured tabular formats.
[0057] The data normalization submodule normalizes the data, mapping it to a specific range, such as [0, 1] or [-1, 1], to eliminate the influence of data units and improve model training performance. For example, the Min-Max normalization method can be used, with the following formula:
[0058] =
[0059] Where x represents the original data. and These are the minimum and maximum values of the data, respectively.
[0060] The feature extraction and selection module includes an image feature extraction submodule: extracting tumor-related features from pathological images, such as tumor morphology, texture, and color. Convolutional neural networks (CNNs) can be used to automatically extract image features, or traditional image processing methods, such as gray-level co-occurrence matrix (GLCM), can be used to extract texture features.
[0061] Clinical feature extraction submodule: Extracts tumor-related features from clinical data, such as patient age, gender, medical history, symptoms, etc. These features can be extracted and selected through data mining and statistical analysis methods;
[0062] Gene feature extraction submodule: Extracts tumor-related features from gene data, such as the expression level and mutation status of tumor-related genes. These features can be analyzed and extracted using bioinformatics methods.
[0063] Feature selection submodule: Selects the extracted features, filtering out the most representative and discriminative features, reducing data dimensionality, and improving the efficiency and accuracy of the model. Principal component analysis (PCA) can be used for feature dimensionality reduction, or recursive feature elimination (RFE) and other methods can be used for feature selection.
[0064] The diagnostic model construction module includes a model training submodule and a model evaluation submodule;
[0065] The model training submodule uses deep learning algorithms to construct a tumor pathology diagnostic model. Feature vectors are input into the model for training, resulting in accurate diagnostic results that identify tumor type, grade, and stage. Convolutional Neural Networks (CNNs) can be used for model training; the formula for their backpropagation algorithm is as follows:
[0066] = (( )⊙ ( )
[0067] in For the first Layer error term, For the first Layer weight matrix ( ) is the first The derivative of the activation function of the layer;
[0068] The model evaluation submodule evaluates the trained diagnostic model, calculating metrics such as accuracy, recall, and F1 score to assess the model's performance and generalization ability. The formula for calculating accuracy is: Accuracy =
[0069] TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative.
[0070] Recall calculation formula: Recall= Recall rate represents the proportion of samples that were actually positive but were correctly identified as positive.
[0071] The prognostic assessment model building module includes a survival analysis submodule and a machine learning model training submodule.
[0072] The Kaplan-Meier estimation formula in the survival analysis submodule is used to estimate the survival function. The formula is as follows:
[0073] (t) = (1- ),in For the first Event time For the first The number of deaths in each event For the first The number of people at risk for each event;
[0074] The formula for Random Forest Survival Analysis (RF-SRC) in the machine learning model training submodule: Random Forest Survival Analysis predicts survival probability by constructing multiple decision trees. The formula for predicting survival probability is as follows:
[0075] P( >t∣x)=
[0076] in For the number of decision trees, For the first The survival probability predicted by the decision tree;
[0077] Risk assessment submodule: Based on the output of the prognostic assessment model, calculate the patient's prognostic risk score and classify the patient into different risk groups, such as high-risk, medium-risk and low-risk groups, to provide a basis for clinical treatment decisions.
[0078] The results display and interaction module includes a results visualization sub-module: it displays the diagnostic results and prognostic assessment reports to doctors and patients in an intuitive and easy-to-understand way, such as through image annotation, charts, text descriptions, etc. It can use heat maps to display the characteristic distribution of tumors and bar charts to display the patient's survival probability and risk score.
[0079] Interactive Feedback Submodule: Provides interactive functions to facilitate doctors' further analysis and interpretation of results, as well as communication and exchange with patients. Doctors can provide feedback and corrections on diagnostic results and prognostic assessment reports, and the feedback information will be used for further optimization and updates of the model.
[0080] The feature selection submodule employs a feature selection method based on genetic algorithms. By simulating the biological evolution process, it automatically searches for the optimal feature subset, improving the efficiency and accuracy of feature selection. Compared with traditional feature selection methods, it can more effectively screen out the features most valuable for tumor diagnosis and prognosis assessment.
[0081] The diagnostic model construction module employs a combination of data augmentation and model regularization during training. By applying data augmentation techniques to the training data to increase sample diversity, and by using model regularization techniques to constrain the complexity of the model, overfitting is prevented, thereby improving the stability and accuracy of the model in practical applications.
[0082] This system acquires various information, including pathological images, clinical data, and genetic data, through a multi-source data acquisition module, providing a comprehensive data foundation for diagnosis. The feature extraction and selection module extracts tumor-related features from this data and uses a feature selection algorithm to filter out the most representative and discriminative features, reducing data dimensionality and improving model efficiency and accuracy. The diagnostic model construction module uses deep learning algorithms to build the diagnostic model, enabling automatic feature learning and extraction, reducing human intervention, minimizing the impact of subjective factors on diagnostic results, and improving diagnostic accuracy and consistency.
[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A comprehensive tumor pathology diagnosis and prognostic assessment system based on artificial intelligence, characterized in that, It includes a multi-source data acquisition module, which is used to acquire patients' pathological images, clinical data and genetic data from multiple channels such as hospital information systems and laboratory equipment, providing a comprehensive data foundation for subsequent analysis; The data preprocessing module is used to clean the collected data, remove noise and outliers, and standardize and normalize the data to make it meet the model input requirements. The feature extraction and selection module is used to extract tumor-related features from the preprocessed data and select the most representative and discriminative features through a feature selection algorithm, thereby reducing data dimensionality and improving model efficiency. The diagnostic model building module is used to input feature vectors into the model for training, obtain diagnostic results that can accurately identify tumor type, grade, and stage, and evaluate and optimize the model. The prognostic assessment model building module is used to predict prognostic indicators such as patient survival and recurrence risk, and generate personalized prognostic assessment reports based on individual patient characteristics. The results display and interaction module is used to present diagnostic results and prognostic assessment reports to doctors and patients in an intuitive and easy-to-understand way, such as through image annotations, charts, and text descriptions, and provides interactive functions.
2. The artificial intelligence-based comprehensive tumor pathology diagnosis and prognostic assessment system according to claim 1, characterized in that: The multi-source data acquisition module includes a pathological image acquisition submodule, a clinical data acquisition submodule, and a gene data acquisition submodule.
3. The artificial intelligence-based comprehensive tumor pathology diagnosis and prognostic assessment system according to claim 2, characterized in that: The data preprocessing module includes a data cleaning submodule, a data standardization submodule, and a data normalization submodule; The data normalization submodule normalizes the data, mapping it to a specific range, such as [0,1] or [-1, 1], to eliminate the influence of data units and improve model training performance. For example, the Min-Max normalization method can be used, with the following formula: = Where x represents the original data. and These are the minimum and maximum values of the data, respectively.
4. The artificial intelligence-based comprehensive tumor pathology diagnosis and prognostic assessment system according to claim 3, characterized in that: The feature extraction and selection module includes an image feature extraction submodule, a clinical feature extraction submodule, a gene feature extraction submodule, and a feature selection submodule.
5. The artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology according to claim 4, characterized in that: The diagnostic model construction module includes a model training submodule and a model evaluation submodule; The model training submodule uses deep learning algorithms to construct a tumor pathology diagnostic model. Feature vectors are input into the model for training, resulting in accurate diagnostic results that identify tumor type, grade, and stage. Convolutional Neural Networks (CNNs) can be used for model training; the formula for their backpropagation algorithm is as follows: = (( )⊙ ( ) in For the first Layer error term, For the first Layer weight matrix ( ) is the first The derivative of the activation function of the layer; The model evaluation submodule evaluates the trained diagnostic model, calculating metrics such as accuracy, recall, and F1 score to assess the model's performance and generalization ability. The accuracy calculation formula is: Accuracy = TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative. Recall calculation formula: Recall= Recall rate represents the proportion of samples that were actually positive but were correctly identified as positive.
6. The artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology according to claim 5, characterized in that: The prognostic assessment model building module includes a survival analysis submodule, a machine learning model training submodule, and a risk assessment submodule. The Kaplan-Meier estimation formula in the survival analysis submodule is used to estimate the survival function. The formula is as follows: (t) = (1- ),in For the first Event time For the first The number of deaths in each event For the first The number of people at risk for each event; The formula for Random Forest Survival Analysis (RF-SRC) in the machine learning model training submodule: Random Forest Survival Analysis predicts survival probability by constructing multiple decision trees. The formula for predicting survival probability is as follows: P( >t∣x)= in For the number of decision trees, For the first The survival probability predicted by the decision tree.
7. The artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology according to claim 6, characterized in that: The results display and interaction module includes a results visualization submodule and an interactive feedback submodule.
8. The artificial intelligence-based comprehensive tumor pathology diagnosis and prognostic assessment system according to claim 7, characterized in that: The feature selection submodule employs a feature selection method based on genetic algorithms. By simulating the biological evolution process, it automatically searches for the optimal feature subset, improving the efficiency and accuracy of feature selection. Compared with traditional feature selection methods, it can more effectively screen out the features most valuable for tumor diagnosis and prognosis assessment.
9. The artificial intelligence-based comprehensive diagnosis and prognostic assessment system for tumor pathology according to claim 8, characterized in that: The diagnostic model construction module employs a combination of data augmentation and model regularization during training. By applying data augmentation techniques to the training data to increase sample diversity, and by using model regularization techniques to constrain the complexity of the model, overfitting is prevented, thereby improving the stability and accuracy of the model in practical applications.