Deep learning-based lung cancer early screening system

Abstract

The invention discloses a deep learning-based lung cancer early screening system, and particularly relates to the technical field of medical assistance, which comprises an intelligent data ecological module, an intelligent analysis and diagnosis module, an intelligent decision support module and a user interaction management module. According to the invention, cross-mechanism data sharing and cooperative training are realized by using data semantic understanding and federated sharing technologies, transparency and traceability of data are ensured through a block chain technology, and in an intelligent analysis and diagnosis module, a deep learning model comprehensively analyzes a multi-modal image and identifies molecular features of cancer cells, so that the accuracy of cancer diagnosis is improved. Potential early-stage lung cancer signs are found through an unsupervised learning algorithm, individual risks are dynamically evaluated, focus development is tracked, and an intelligent decision support module combines personalized treatment suggestions, optimizes clinical pathways and multidisciplinary collaboration, provides accurate decision suggestions and assists doctors in making scientific and effective treatment schemes.

Description

Technical Field

[0001] The present invention relates to the field of medical assistance technology, and in particular to an early screening system for lung cancer based on deep learning. Background Art

[0002] Lung cancer is one of the most common malignant tumors in the world, and its morbidity and mortality rate rank first among cancers. Since the early symptoms of lung cancer are not obvious, most patients are already in the late stage when diagnosed, making treatment difficult and the survival rate of patients low. Early detection and accurate screening are considered to be important means to improve the prognosis of lung cancer patients and increase the cure rate.

[0003] At present, lung cancer screening based on medical imaging has gradually been applied in clinical practice, and lesion detection is performed through imaging technologies such as CT and PET. However, due to factors such as the huge amount of imaging data, complex cancer characteristics, and differences in physician experience, traditional screening methods often have problems such as high misdiagnosis rate, high risk of missed diagnosis, and low efficiency. Due to the uneven distribution of medical resources, data island effects, and patient privacy issues, data sharing and collaboration between medical institutions face many obstacles, further limiting the promotion and application of early screening technology for lung cancer.

[0004] Existing lung cancer early screening systems mostly rely on single-modality medical imaging data for lesion analysis, lack the ability to comprehensively process multimodal data, and find it difficult to fully identify the complex characteristics of early lung cancer. The training of deep learning models in existing technologies usually needs to rely on data sets from a single institution. Due to data silos and patient privacy protection restrictions, the model's generalization ability is insufficient and cross-institutional collaborative optimization cannot be achieved.

[0005] Traditional systems mostly use supervised learning methods for data analysis and anomaly detection, which rely on the quality and quantity of labeled data. It is difficult to discover potential early signs of cancer and there is a risk of misdiagnosis and missed diagnosis. The existing screening systems are relatively simple in user interaction and health management functions, and fail to provide personalized treatment recommendations and dynamic risk assessment. Doctors' clinical decision-making assistance capabilities are limited, patients have low sense of participation, and health management tools are insufficient. Most screening systems lack real-time monitoring and intelligent feedback mechanisms, and are unable to provide timely warnings for abnormal data or efficiently archive normal data, limiting the continuous optimization and efficient use of the system. These problems have greatly reduced the efficiency and reliability of early screening for lung cancer in actual applications.

[0006] Taking all the above issues into consideration, we provide a lung cancer early screening system based on deep learning. Summary of the invention

[0007] The purpose of the present invention is to solve the shortcomings of the prior art and propose an early screening system for lung cancer based on deep learning.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A lung cancer early screening system based on deep learning, including an intelligent data ecology module, an intelligent analysis and diagnosis module, an intelligent decision support module, and a user interaction management module;

[0010] The intelligent data ecosystem module includes a multi-dimensional health data integration module for integrating health data from different channels to build a comprehensive health ecosystem, an environmental lifestyle monitoring module for collecting users' environmental factors and lifestyle data to assess their impact on lung cancer risk, a data semantic understanding module for parsing unstructured data using doctors' diagnostic reports, a data federation sharing module for achieving cross-institutional and cross-regional data sharing and collaborative training through federated learning technology, and a block data traceability module for using blockchain technology to record the data collection, transmission and use process;

[0011] The intelligent analysis and diagnosis module includes a multimodal image analysis module for comprehensive analysis using a deep learning model in combination with multiple medical images such as CT, X-ray, and PET, a molecular image recognition module for analyzing molecular image data, identifying the molecular characteristics of cancer cells, and assisting in early and accurate diagnosis, an abnormal pattern detection module for discovering abnormal patterns in images and data through unsupervised learning algorithms and identifying potential early signs of lung cancer, a dynamic risk assessment module for dynamically assessing an individual's lung cancer risk based on real-time data and historical data and generating a personalized risk report, and a lesion tracking and prediction module for long-term tracking of detected suspicious lesions, predicting their development trends, and assisting in clinical decision-making.

[0012] The present invention is further configured as follows: the intelligent decision support module includes a personalized treatment suggestion module for generating personalized treatment plan suggestions according to the specific conditions of the patient, a clinical pathway optimization module for analyzing a large amount of clinical data to optimize the clinical pathway of lung cancer screening and treatment, a multidisciplinary collaboration support module for integrating expert knowledge of multiple disciplines such as oncology, radiology, and pathology to provide comprehensive decision support, a simulation analysis and prediction module for predicting the potential results of different treatment plans using simulation models, and an intelligent knowledge graph module for constructing a knowledge graph related to lung cancer and integrating the latest research results and clinical data;

[0013] The user interaction management module includes an intelligent health assistant module for providing 24 / 7 intelligent health consulting services through natural language processing technology to answer users' questions about lung cancer screening and health management, a virtual reality education module for using VR technology to provide users and medical staff with immersive educational experiences in lung cancer prevention, screening and treatment, a real-time monitoring feedback module for real-time monitoring of users' health data and screening progress and providing instant feedback and health advice, a community support interaction module for building a support community for patients and their families to provide a communication platform, and an automatic report generation and distribution module for automatically generating detailed screening reports and distributing them to users and relevant medical staff through multiple channels.

[0014] The present invention is further configured as follows: the abnormal pattern detection module includes a data preprocessing module for cleaning and normalizing input data, processing missing values, noise and abnormal value operations, a feature extraction module for extracting meaningful features from the preprocessed data, a feature selection module for selecting the most relevant and informative features, reducing data dimensions and reducing computational complexity, an abnormality detection algorithm module for applying an unsupervised learning algorithm to analyze feature data and identify potential abnormal patterns, and a binary classification determination module for determining whether an abnormal pattern exists based on the output of the abnormality detection algorithm;

[0015] The binary classification determination module divides the condition into an abnormal processing branch for an abnormal state and a normal processing branch for a normal state;

[0016] The exception handling branch includes an exception confirmation analysis module for further verifying and analyzing detected exceptions, an alarm notification module for automatically sending alarms and notifications to relevant medical personnel upon confirmation of the existence of an exception, and a case record tracking module for recording the abnormality detection results in the patient's electronic health record and initiating a subsequent case tracking process;

[0017] The normal processing branch includes a continuous monitoring module for continuously monitoring data determined to be normal to ensure timely detection of possible abnormal changes in future data, a data archiving learning module for archiving and storing normal data as a reference for future model training and optimization, and a user feedback optimization module for collecting user feedback to evaluate the detection effect and user satisfaction of the system.

[0018] The present invention is further configured as follows: the anomaly detection algorithm combination formula is as follows:

[0019] Formula 1:

[0020] E(x)=||xx|| 2 -||xf(g(x))|| 2

[0021] x: original input data vector;

[0022] g(x): encoder function, which maps the input data x to a low-dimensional latent representation space;

[0023] f(g(x)): decoder function, which recovers a reconstructed version of the original data from the latent representation z

[0024] The reconstruction error represents the squared Euclidean distance between the original input data and the reconstructed data;

[0025] Formula 2:

[0026]

[0027] x: input data point to be detected;

[0028] h(x): the average path length of the data point x in formula 2. The path length refers to the number of nodes passed from the root node to the leaf node in the random tree;

[0029] c(n): The average path length of the tree. For a balanced binary tree with n samples, the formula is:

[0030]

[0031] Where H(i) is the i-th harmonic number, which is approximately ln(i)+γ, and γ is the Euler constant;

[0032] Normalization factor, used to adjust the path length to be in the range [0,1];

[0033] P(x): normalized path length score, reflecting the degree of abnormality of data point x. The shorter the path length, the more likely the data point is an abnormal point.

[0034] Formula 3:

[0035] A(x)=α·Normalize(E(x))-β·Normalize(P(x))

[0036] A(x): combined anomaly score, which comprehensively considers the reconstruction error of Formula 1 and the path length of Formula 2 to obtain the final anomaly score;

[0037] α and β: weight parameters used to balance the impact of reconstruction error and path length in the final score, usually satisfying α+β=1;

[0038] Normalize(E(x)): Normalize the reconstruction error E(x) to standardize its value range, usually between [0,1];

[0039] Normalize(P(x)): Normalizes the path length score P(x) to a standard range, usually between [0,1].

[0040] The present invention is further configured as follows: in the anomaly detection algorithm module, normal data features are learned through formula 1 to reduce the reconstruction error of normal data, while the abnormal data is less or does not exist in the training set, making it impossible to effectively reconstruct these data, resulting in a large reconstruction error; in the anomaly detection algorithm module, formula 2 calculates the isolated value of the data by identifying the data path length to identify normal or abnormal data; in the anomaly detection algorithm module, formula 3 combines the scores of formula 1 and formula 2 and adjusts the weight parameters α and β, so as to optimize the model performance according to the specific application scenario.

[0041] The present invention is further configured as follows: the anomaly detection algorithm module can train formula 1 with a large amount of normal data in daily life, optimize the parameters of g(x) and f(g(x)) in formula 1 to minimize the reconstruction error E(x), train formula 2 with the same data, and construct multiple random trees to calculate the path length score P(x); when the anomaly detection algorithm module is used, the reconstruction error E(x) and the path length score P(x) are calculated for the new input data x, and then E(x) and P(x) are normalized to calculate the combined anomaly score A(x), and compared with the set threshold value θ:

[0042] If A(x)>θ, then x is considered abnormal;

[0043] Otherwise, it is judged as normal.

[0044] The present invention is further configured as follows: the environmental lifestyle monitoring module evaluates the impact of environmental and lifestyle factors on lung cancer risk in the health data from different channels summarized by the multidimensional health data integration module; the data semantic understanding module parses the environmental and lifestyle data collected by the environmental lifestyle monitoring module; the data federation sharing module utilizes the structured data processed by the data semantic understanding module to realize cross-institutional and cross-regional data sharing and collaborative training; the block data traceability module records the collection, transmission and use process of the data shared by the data federation sharing module to ensure the transparency and traceability of the data.

[0045] The present invention is further configured as follows: the molecular imaging recognition module uses a multimodal imaging analysis module to perform a comprehensive analysis of various medical images such as CT, X-ray, PET, etc. to identify the molecular characteristics of cancer cells; the abnormal pattern detection module analyzes the cancer cell feature data identified by the molecular imaging recognition module, and uses an unsupervised learning algorithm to discover potential early signs of lung cancer; the dynamic risk assessment module evaluates the individual's lung cancer risk in real time and generates a personalized risk report based on the abnormal data detected by the abnormal pattern detection module; the lesion tracking and prediction module uses the risk assessment results generated by the dynamic risk assessment module to perform long-term tracking and development trend prediction of the detected suspicious lesions to assist clinical decision-making.

[0046] The present invention is further configured as follows: the clinical pathway optimization module analyzes the treatment plan recommendations for the specific conditions of the patients generated by the personalized treatment recommendation module, and optimizes the clinical pathway for lung cancer screening and treatment; the multidisciplinary collaboration support module integrates the clinical pathway optimized by the clinical pathway optimization module, and combines the expert knowledge of multiple disciplines such as oncology, radiology, and pathology to provide comprehensive decision support; the simulation analysis and prediction module uses the expert knowledge integrated by the multidisciplinary collaboration support module to predict the potential results of different treatment plans; the intelligent knowledge graph module integrates the prediction results of the simulation analysis and prediction module, and combines the latest research results and clinical data to further enhance decision support.

[0047] The present invention is further configured as follows: the virtual reality education module utilizes the 24 / 7 health consultation service provided by the intelligent health assistant module through natural language processing technology to provide an immersive education experience for users and medical staff; the real-time monitoring feedback module monitors the user's health data and screening progress in real time based on the educational content provided by the virtual reality education module, and provides instant feedback and health advice; the community support interaction module communicates and supports patients and their families through the health advice generated by the real-time monitoring feedback module; the automatic report generation and distribution module utilizes the communication platform constructed by the community support interaction module to automatically generate a detailed screening report, and distributes it to users and relevant medical staff through multiple channels.

[0048] The beneficial effects of the present invention are:

[0049] 1. The present invention realizes cross-institutional data sharing and collaborative training by utilizing data semantic understanding and federated sharing technology, and ensures data transparency and traceability through blockchain technology. In the intelligent analysis and diagnosis module, the deep learning model conducts a comprehensive analysis of multimodal images, identifies the molecular characteristics of cancer cells, discovers potential early signs of lung cancer through unsupervised learning algorithms, dynamically assesses individual risks and tracks lesion development. The intelligent decision support module combines personalized treatment recommendations, optimized clinical pathways and multidisciplinary collaboration to provide accurate decision-making recommendations and assist doctors in formulating scientific and effective treatment plans. The user interaction management module provides a friendly user interface and comprehensive health management tools through intelligent health assistants, virtual reality education, real-time monitoring feedback and community support.

[0050] 2. The present invention further verifies through the abnormal confirmation analysis module, then automatically sends an alarm to medical personnel, and records the results in the electronic health record to start follow-up tracking; for normal data, the continuous monitoring module performs real-time monitoring, the data archiving learning module stores and archives for future model optimization, and improves system performance through the user feedback optimization module. This working principle effectively improves the accuracy and timeliness of abnormal detection, reduces the risk of misdiagnosis and missed diagnosis, and improves the efficiency and reliability of early screening for lung cancer, thereby helping to achieve early detection and intervention, and improving the treatment effect and survival rate of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the system modules in the present invention.

[0052] Figure 2 It is a schematic diagram of the abnormal pattern detection module flow in the present invention.

[0053] Figure 3 It is a schematic diagram of the dynamic risk assessment module flow in the present invention. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0055] Example 1

[0056] like Figure 1-3As shown, a lung cancer early screening system based on deep learning includes an intelligent data ecology module for collecting, integrating and managing data from multiple sources, an intelligent analysis and diagnosis module for using advanced algorithms and technologies to conduct in-depth analysis of data and identify potential problems and risks, an intelligent decision support module for providing accurate decision suggestions based on analysis results to assist users in making scientific and effective decisions, and a user interaction management module for providing a friendly user interface and management tools;

[0057] The intelligent data ecosystem module includes a multi-dimensional health data integration module for integrating health data from different channels, including but not limited to medical images, genomic data, electronic health records (EHR), and wearable device data to build a comprehensive health ecosystem; an environmental lifestyle monitoring module for collecting users' environmental factors (such as air quality, radiation levels) and lifestyle data (such as eating habits and exercise volume) to assess their impact on lung cancer risk; a data semantic understanding module for parsing unstructured data using doctors' diagnostic reports; a data federation sharing module for achieving cross-institutional and cross-regional data sharing and collaborative training through federated learning technology; and a block data traceability module for using blockchain technology to record the data collection, transmission, and use process;

[0058] The intelligent analysis and diagnosis module includes a multimodal image analysis module for combining CT, X-ray, PET and other medical images with a deep learning model for comprehensive analysis, a molecular image recognition module for analyzing molecular image data, identifying the molecular characteristics of cancer cells, and assisting in early and accurate diagnosis, an abnormal pattern detection module for discovering abnormal patterns in images and data through unsupervised learning algorithms and identifying potential early signs of lung cancer, a dynamic risk assessment module for dynamically assessing an individual's lung cancer risk based on real-time data and historical data and generating a personalized risk report, and a lesion tracking and prediction module for long-term tracking of detected suspicious lesions, predicting their development trends, and assisting clinical decision-making;

[0059] The intelligent decision support module includes a personalized treatment suggestion module for generating personalized treatment plan suggestions based on the patient's specific conditions (gene information, lesion characteristics, etc.) to assist doctors in making decisions, a clinical pathway optimization module for analyzing a large amount of clinical data and optimizing the clinical pathway for lung cancer screening and treatment, a multidisciplinary collaboration support module for integrating expert knowledge from multiple disciplines such as oncology, radiology, and pathology to provide comprehensive decision support, a simulation analysis and prediction module for using simulation models to predict the potential results of different treatment plans, and an intelligent knowledge graph module for building a knowledge graph related to lung cancer and integrating the latest research results and clinical data;

[0060] The user interaction management module includes an intelligent health assistant module for providing 24 / 7 intelligent health consulting services through natural language processing technology to answer users' questions about lung cancer screening and health management, a virtual reality education module for using VR technology to provide users and medical staff with immersive educational experience in lung cancer prevention, screening and treatment, a real-time monitoring feedback module for real-time monitoring of users' health data and screening progress, providing instant feedback and health advice, a community support interaction module for building a support community for patients and their families to provide a communication platform, and an automatic report generation and distribution module for automatically generating detailed screening reports and distributing them to users and relevant medical staff through multiple channels;

[0061] The environmental lifestyle monitoring module evaluates the impact of environmental and lifestyle factors on lung cancer risk in the multidimensional health data integration module that summarizes health data from different channels; the data semantic understanding module parses the environmental and lifestyle data collected by the environmental lifestyle monitoring module; the data federation sharing module uses the structured data processed by the data semantic understanding module to achieve cross-institutional and cross-regional data sharing and collaborative training; the block data traceability module records the collection, transmission and use process of the data shared by the data federation sharing module to ensure the transparency and traceability of the data;

[0062] The molecular image recognition module uses the multimodal image analysis module to comprehensively analyze various medical images such as CT, X-ray, PET, etc. to identify the molecular characteristics of cancer cells; the abnormal pattern detection module analyzes the characteristic data of cancer cells identified by the molecular image recognition module and uses an unsupervised learning algorithm to discover potential early signs of lung cancer; the dynamic risk assessment module evaluates the individual's lung cancer risk in real time and generates a personalized risk report based on the abnormal data detected by the abnormal pattern detection module; the lesion tracking prediction module uses the risk assessment results generated by the dynamic risk assessment module to conduct long-term tracking of the detected suspicious lesions and predict their development trends to assist clinical decision-making;

[0063] The clinical pathway optimization module analyzes the treatment plan recommendations for specific patient conditions generated by the personalized treatment recommendation module to optimize the clinical pathway for lung cancer screening and treatment; the multidisciplinary collaboration support module integrates the clinical pathway optimized by the clinical pathway optimization module, combines the expert knowledge of multiple disciplines such as oncology, radiology, and pathology, and provides comprehensive decision support; the simulation analysis and prediction module uses the expert knowledge integrated by the multidisciplinary collaboration support module to predict the potential results of different treatment plans; the intelligent knowledge graph module integrates the prediction results of the simulation analysis and prediction module, combines the latest research results and clinical data, and further enhances decision support;

[0064] The virtual reality education module uses the 24 / 7 health consultation service provided by the intelligent health assistant module through natural language processing technology to provide an immersive education experience for users and medical staff; the real-time monitoring feedback module monitors the user's health data and screening progress in real time based on the educational content provided by the virtual reality education module, and provides instant feedback and health advice; the community support interaction module communicates and supports patients and their families through the health advice generated by the real-time monitoring feedback module; the automatic report generation and distribution module uses the communication platform constructed by the community support interaction module to automatically generate detailed screening reports, and distributes them to users and relevant medical staff through multiple channels.

[0065] In the above embodiment, the system realizes cross-institutional data sharing and collaborative training by utilizing data semantic understanding and federated sharing technology, and ensures data transparency and traceability through blockchain technology. In the intelligent analysis and diagnosis module, the deep learning model comprehensively analyzes multimodal images, identifies the molecular characteristics of cancer cells, and discovers potential early signs of lung cancer through unsupervised learning algorithms, dynamically assesses individual risks and tracks lesion development. The intelligent decision support module combines personalized treatment recommendations, optimizes clinical pathways and multidisciplinary collaboration to provide accurate decision-making recommendations to assist doctors in formulating scientific and effective treatment plans. The user interaction management module provides a friendly user interface and comprehensive health management tools through intelligent health assistants, virtual reality education, real-time monitoring feedback and community support. In actual applications, the system can achieve early and efficient lung cancer detection, provide personalized treatment plans, improve data security and sharing efficiency, enhance user experience and the scientific nature of medical decision-making, and significantly improve the overall effect of lung cancer screening and management.

[0066] Among them, the dynamic risk assessment module sets the risk threshold through the medical monitoring system, and confirms the limit value set by the staff and inputs it into the system. Then the staff can choose the monitoring type as needed. They can define real-time monitoring or scheduled monitoring according to the patient's condition. If scheduled monitoring is selected, the monitoring frequency setting needs to be adjusted according to the system preset, and then the data jump platform is monitored, and the data obtained from real-time or scheduled monitoring is transmitted to the data status evaluation for evaluation, and the data from the data status evaluation is returned to the data jump platform to confirm the time period when the problem occurred and the specific data display of the problem segment, and the confirmation result is synchronized to the medical monitoring system for subsequent retrieval.

[0067] The data status assessment divides the patient data into the following four states:

[0068] 1. High risk

[0069] Feedback generation module: Generates high-risk alert information.

[0070] Intervention 1: Take immediate emergency intervention measures, such as activating emergency response plans.

[0071] 2. Medium risk

[0072] User interaction module: informs relevant personnel to make evaluations and decisions.

[0073] Intervention 2: Implement intermediate interventions, such as adjusting operating parameters.

[0074] 3. Low risk

[0075] Data storage and analysis module: record low-risk status and conduct long-term data analysis.

[0076] Intervention 3: Perform routine monitoring and preventive maintenance.

[0077] 4. Abnormal status

[0078] Feedback generation module: Generates abnormal status report.

[0079] Intervention 4: Conduct special investigations and corrective measures to prevent potential risks.

[0080] Example 2

[0081] like Figure 1-2 As shown, a lung cancer early screening system based on deep learning, the abnormal pattern detection module includes a data preprocessing module for cleaning and normalizing input data, processing missing values, noise and outliers, a feature extraction module for extracting meaningful features from the preprocessed data, a feature selection module for selecting the most relevant and informative features, reducing data dimensions and reducing computational complexity, an anomaly detection algorithm module for applying an unsupervised learning algorithm to analyze feature data and identify potential abnormal patterns, and a binary classification determination module for determining whether an abnormal pattern exists according to the output of the anomaly detection algorithm;

[0082] The two-classification determination module divides the condition into an abnormal processing branch for abnormal conditions and a normal processing branch for normal conditions;

[0083] The exception handling branch includes an exception confirmation analysis module for further verifying and analyzing detected exceptions, an alarm notification module for automatically sending alarms and notifications to relevant medical personnel upon confirmation of the existence of an exception, and a case record tracking module for recording the abnormality detection results in the patient's electronic health record and initiating a subsequent case tracking process;

[0084] The normal processing branch includes a continuous monitoring module for continuously monitoring data determined to be normal to ensure timely discovery of possible abnormal changes in future data, a data archiving learning module for archiving and storing normal data as a reference for future model training and optimization, and a user feedback optimization module for collecting user feedback to evaluate the detection effect of the system and user satisfaction;

[0085] The anomaly detection algorithm combination formula is as follows:

[0086] Formula 1:

[0087]

[0088] x: original input data vector;

[0089] g(x): encoder function, which maps the input data x to a low-dimensional latent representation space;

[0090] f(g(x)): decoder function, which recovers a reconstructed version of the original data from the latent representation z

[0091] The reconstruction error represents the squared Euclidean distance between the original input data and the reconstructed data;

[0092] Formula 2:

[0093]

[0094] x: input data point to be detected;

[0095] h(x): the average path length of the data point x in formula 2. The path length refers to the number of nodes passed from the root node to the leaf node in the random tree;

[0096] c(n): The average path length of the tree. For a balanced binary tree with n samples, the formula is:

[0097]

[0098] Where H(i) is the i-th harmonic number, which is approximately ln(i)+γ, and γ is the Euler constant;

[0099] Normalization factor, used to adjust the path length to be in the range [0,1];

[0100] P(x): normalized path length score, reflecting the degree of abnormality of data point x. The shorter the path length, the more likely the data point is an abnormal point.

[0101] Formula 3:

[0102] A(x)=α·Normalize(E(x))-β·Normalize(P(x))

[0103] A(x): combined anomaly score, which comprehensively considers the reconstruction error of Formula 1 and the path length of Formula 2 to obtain the final anomaly score;

[0104] α and β: weight parameters used to balance the impact of reconstruction error and path length in the final score, usually satisfying α+β=1;

[0105] Normalize(E(x)): Normalize the reconstruction error E(x) to standardize its value range, usually between [0,1];

[0106] Normalize(P(x)): Normalize the path length score P(x) to standardize its value range, usually between [0,1];

[0107] In the anomaly detection algorithm module, formula 1 is used to learn the features of normal data to reduce the reconstruction error of normal data, but the abnormal data is less or does not exist in the training set, so that the data cannot be effectively reconstructed, resulting in a large reconstruction error; formula 2 in the anomaly detection algorithm module calculates the isolated value of the data by identifying the data path length to identify normal or abnormal data, and formula 3 in the anomaly detection algorithm module combines the scores of formula 1 and formula 2 and adjusts the weight parameters α and β, so as to optimize the model performance according to the specific application scenario;

[0108] The anomaly detection algorithm module can be used to train Formula 1 with a large amount of normal data on a daily basis, optimize the parameters of g(x) and f(g(x)) in Formula 1 to minimize the reconstruction error E(x), train Formula 2 with the same data, and construct multiple random trees to calculate the path length score P(x); when the anomaly detection algorithm module is used, the reconstruction error E(x) and the path length score P(x) are calculated for the new input data x, and then E(x) and P(x) are normalized to calculate the combined anomaly score A(x), and compared with the set threshold θ:

[0109] If A(x)>θ, then x is considered abnormal;

[0110] Otherwise, it is judged as normal.

[0111] In the above embodiment, the abnormal pattern detection module cleans and normalizes the input data through the data preprocessing module, processes missing values, noise and abnormal values, the feature extraction module extracts meaningful features from the preprocessed data, and the feature selection module selects the most relevant and informative features to reduce the data dimension, the abnormal detection algorithm module combines the reconstruction error of formula 1 and the path length score of formula 2, calculates the comprehensive abnormal score through formula 3, and the binary classification determination module determines whether the data has an abnormal pattern. For data determined to be abnormal, the system further verifies it through the abnormal confirmation analysis module, then automatically sends an alarm to medical personnel, and records the results in the electronic health record to start follow-up tracking; for normal data, it is monitored in real time by the continuous monitoring module, and the data archiving learning module stores and archives it for future model optimization, and improves system performance through the user feedback optimization module. This working principle effectively improves the accuracy and timeliness of abnormal detection, reduces the risk of misdiagnosis and missed diagnosis, and improves the efficiency and reliability of early screening for lung cancer, thereby helping to achieve early detection and intervention, and improve the treatment effect and survival rate of patients.

[0112] Among them, the data preprocessing module cleans and normalizes the input data, processes missing values, noise and outliers, and then passes the sorted data to the feature extraction module to extract meaningful features; the feature extraction module extracts useful features from the preprocessed data, and then passes these features to the feature selection module to select the most relevant and informative features to reduce the data dimension; the important features screened out by the feature selection module are input into the anomaly detection algorithm module, which uses an unsupervised learning algorithm to analyze these feature data and identify potential abnormal patterns; the abnormal pattern recognition results generated by the anomaly detection algorithm module are passed to the binary classification determination module, which determines whether there is an abnormal pattern in the data.

[0113] Among them, when the binary classification determination module identifies an abnormal state, the abnormal confirmation analysis module further verifies and analyzes these detected abnormalities; when the binary classification determination module determines that the data is in a normal state, the continuous monitoring module continuously monitors these normal data so as to promptly discover possible abnormal changes in the future.

[0114] Working principle: When the present invention is used, the intelligent analysis and diagnosis module uses technologies such as multimodal image analysis, molecular image recognition, abnormal pattern detection, dynamic risk assessment and lesion tracking prediction to deeply explore the potential risks and early signs of lung cancer in the data. The intelligent decision support module combines personalized treatment recommendations, clinical pathway optimization, multidisciplinary collaboration and simulation analysis and prediction to provide doctors with accurate decision support and assist in formulating scientific and effective treatment plans. The user interaction management module provides a user-friendly interface and comprehensive health management tools through intelligent health assistants, virtual reality education, real-time monitoring feedback and community support to enhance user experience and participation.

[0115] In actual use, the abnormal pattern detection module plays a key role. Its workflow includes data preprocessing, feature extraction and selection, application of anomaly detection algorithm and binary classification judgment. The data preprocessing module cleans and normalizes the input health data, handles missing values, noise and outliers, and ensures data quality. The feature extraction module extracts meaningful features from the preprocessed data, and screens out the most relevant and informative features through the feature selection module, reduces the data dimension, and improves computing efficiency. The anomaly detection algorithm module combines the reconstruction error of formula 1 and the path length score of formula 2, calculates the comprehensive anomaly score through formula 3, and identifies potential abnormal patterns. The binary classification judgment module determines whether there is an abnormal pattern in the data based on the set threshold. If an abnormality is detected, the system will promptly issue an alarm to relevant medical personnel and start the subsequent processing process through steps such as abnormal confirmation analysis, alarm notification and case record tracking; for normal data, continuous monitoring and data archiving are carried out to facilitate future model optimization and system performance improvement.

[0116] This system significantly improves the accuracy and efficiency of early screening for lung cancer through deep learning and advanced data processing technology. By fully integrating and intelligently analyzing multi-source data, it can identify potential lung cancer risks at an early stage, reduce the probability of misdiagnosis and missed diagnosis, and thus achieve early intervention and improve patients' treatment effects and survival rates. The application of blockchain technology ensures the transparency and traceability of data and enhances the security and reliability of data sharing. The personalized recommendations and multidisciplinary collaboration of the intelligent decision support module further optimize the clinical decision-making process and improve the scientificity and effectiveness of medical services. The user interaction management module improves users' health management experience and participation through functions such as intelligent health assistants and virtual reality education, promotes good interaction between patients and medical institutions, and improves the overall effect of lung cancer screening and management, providing strong technical support for public health.

[0117] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A lung cancer early screening system based on deep learning, characterized in that: It includes intelligent data ecology module, intelligent analysis and diagnosis module, intelligent decision support module and user interaction management module; The intelligent data ecosystem module includes a multi-dimensional health data integration module for integrating health data from different channels to build a comprehensive health ecosystem, an environmental lifestyle monitoring module for collecting users' environmental factors and lifestyle data to assess their impact on lung cancer risk, a data semantic understanding module for parsing unstructured data using doctors' diagnostic reports, a data federation sharing module for achieving cross-institutional and cross-regional data sharing and collaborative training through federated learning technology, and a block data traceability module for using blockchain technology to record the data collection, transmission and use process; The intelligent analysis and diagnosis module includes a multimodal image analysis module for comprehensive analysis using a deep learning model in combination with multiple medical images such as CT, X-ray, and PET, a molecular image recognition module for analyzing molecular image data, identifying the molecular characteristics of cancer cells, and assisting in early and accurate diagnosis, an abnormal pattern detection module for discovering abnormal patterns in images and data through unsupervised learning algorithms and identifying potential early signs of lung cancer, a dynamic risk assessment module for dynamically assessing an individual's lung cancer risk based on real-time data and historical data and generating a personalized risk report, and a module for long-term tracking of detected suspicious lesions and predicting their development trends. Lesion tracking and prediction module to assist clinical decision making.

2. According to claim 1, a deep learning-based lung cancer early screening system is characterized in that: The intelligent decision support module includes a personalized treatment suggestion module for generating personalized treatment plan suggestions according to the specific conditions of the patient, a clinical pathway optimization module for analyzing a large amount of clinical data to optimize the clinical pathway for lung cancer screening and treatment, a multidisciplinary collaboration support module for integrating expert knowledge from multiple disciplines such as oncology, radiology, and pathology to provide comprehensive decision support, a simulation analysis and prediction module for using simulation models to predict the potential results of different treatment plans, and an intelligent knowledge graph module for building a knowledge graph related to lung cancer and integrating the latest research results and clinical data; The user interaction management module includes an intelligent health assistant module for providing 24 / 7 intelligent health consulting services through natural language processing technology to answer users' questions about lung cancer screening and health management, a virtual reality education module for using VR technology to provide users and medical staff with immersive educational experiences in lung cancer prevention, screening and treatment, a real-time monitoring feedback module for real-time monitoring of users' health data and screening progress and providing instant feedback and health advice, a community support interaction module for building a support community for patients and their families to provide a communication platform, and an automatic report generation and distribution module for automatically generating detailed screening reports and distributing them to users and relevant medical staff through multiple channels.

3. The deep learning-based lung cancer early screening system according to claim 1, characterized in that: The abnormal pattern detection module includes a data preprocessing module for cleaning and normalizing input data, processing missing values, noise and abnormal value operations, a feature extraction module for extracting meaningful features from the preprocessed data, a feature selection module for selecting the most relevant and informative features, reducing data dimensions and reducing computational complexity, an abnormality detection algorithm module for applying an unsupervised learning algorithm to analyze feature data and identify potential abnormal patterns, and a binary classification determination module for determining whether an abnormal pattern exists based on the output of the abnormality detection algorithm; The binary classification determination module divides the condition into an abnormal processing branch for an abnormal state and a normal processing branch for a normal state; The exception handling branch includes an exception confirmation analysis module for further verifying and analyzing detected exceptions, an alarm notification module for automatically sending alarms and notifications to relevant medical personnel upon confirmation of the existence of an exception, and a case record tracking module for recording the abnormality detection results in the patient's electronic health record and initiating a subsequent case tracking process; The normal processing branch includes a continuous monitoring module for continuously monitoring data determined to be normal to ensure timely detection of possible abnormal changes in future data, a data archiving learning module for archiving and storing normal data as a reference for future model training and optimization, and a user feedback optimization module for collecting user feedback to evaluate the detection effect and user satisfaction of the system.

4. The deep learning-based lung cancer early screening system according to claim 3, characterized in that: The anomaly detection algorithm combination formula is as follows: Formula 1: x: original input data vector; g(x): encoder function, which maps the input data x to a low-dimensional latent representation space; f(g(x)): decoder function, which recovers a reconstructed version of the original data from the latent representation z The reconstruction error represents the squared Euclidean distance between the original input data and the reconstructed data; Formula 2: x: input data point to be detected; h(x): the average path length of the data point x in formula 2. The path length refers to the number of nodes passed from the root node to the leaf node in the random tree; c(n): The average path length of the tree. For a balanced binary tree with n samples, the formula is: Where H(i) is the i-th harmonic number, which is approximately ln(i)+γ, and γ is the Euler constant; Normalization factor, used to adjust the path length to be in the range [0,1]; P(x): normalized path length score, reflecting the degree of abnormality of data point x. The shorter the path length, the more likely the data point is an abnormal point. Formula 3: A(x)=α·Normalize(E(x))-β·Normalize(P(x)) A(x): combined anomaly score, which comprehensively considers the reconstruction error of Formula 1 and the path length of Formula 2 to obtain the final anomaly score; α and β: weight parameters used to balance the impact of reconstruction error and path length in the final score, usually satisfying α+β=1; Normalize(E(x)): Normalize the reconstruction error E(x) to standardize its value range, usually between [0,1]; Normalize(P(x)): Normalizes the path length score P(x) to a standard range, usually between [0,1].

5. The deep learning-based lung cancer early screening system according to claim 4, characterized in that: In the anomaly detection algorithm module, formula 1 is used to learn the normal data features to reduce the reconstruction error of normal data, but the abnormal data is less or does not exist in the training set, making it impossible to effectively reconstruct these data, resulting in a large reconstruction error; in the anomaly detection algorithm module, formula 2 calculates the isolated value of the data by identifying the data path length to identify normal or abnormal data; in the anomaly detection algorithm module, formula 3 combines the scores of formulas 1 and 2 and adjusts the weight parameters α and β, so as to optimize the model performance according to the specific application scenario.

6. The deep learning-based lung cancer early screening system according to claim 5, characterized in that: The anomaly detection algorithm module can be used to train Formula 1 with a large amount of normal data on a daily basis, optimize the parameters of g(x) and f(g(x)) in Formula 1 to minimize the reconstruction error E(x), train Formula 2 with the same data, and construct multiple random trees to calculate the path length score P(x); when the anomaly detection algorithm module is used, the reconstruction error E(x) and the path length score P(x) are calculated for the new input data x, and then E(x) and P(x) are normalized to calculate the combined anomaly score A(x), and compared with the set threshold θ: If A(x)>θ, then x is considered abnormal; Otherwise, it is judged as normal.

7. The deep learning-based lung cancer early screening system according to claim 1, characterized in that: The environmental lifestyle monitoring module evaluates the impact of environmental and lifestyle factors on lung cancer risk in the health data from different channels summarized by the multidimensional health data integration module; the data semantic understanding module parses the environmental and lifestyle data collected by the environmental lifestyle monitoring module; the data federation sharing module uses the structured data processed by the data semantic understanding module to realize cross-institutional and cross-regional data sharing and collaborative training; the block data traceability module records the collection, transmission and use process of the data shared by the data federation sharing module to ensure the transparency and traceability of the data.

8. The deep learning-based lung cancer early screening system according to claim 1, characterized in that: The molecular image recognition module uses the multimodal image analysis module to comprehensively analyze various medical images such as CT, X-ray, PET, etc. to identify the molecular characteristics of cancer cells; the abnormal pattern detection module analyzes the cancer cell feature data identified by the molecular image recognition module and uses an unsupervised learning algorithm to discover potential early signs of lung cancer; the dynamic risk assessment module evaluates the individual's lung cancer risk in real time and generates a personalized risk report based on the abnormal data detected by the abnormal pattern detection module; The lesion tracking and prediction module uses the risk assessment results generated by the dynamic risk assessment module to conduct long-term tracking and development trend prediction of detected suspicious lesions to assist clinical decision-making.

9. The deep learning-based lung cancer early screening system according to claim 2, characterized in that: The clinical pathway optimization module analyzes the treatment plan recommendations for specific patient conditions generated by the personalized treatment recommendation module to optimize the clinical pathway for lung cancer screening and treatment; the multidisciplinary collaboration support module integrates the clinical pathway optimized by the clinical pathway optimization module, combines expert knowledge from multiple disciplines such as oncology, radiology, and pathology, and provides comprehensive decision support; the simulation analysis and prediction module uses the expert knowledge integrated by the multidisciplinary collaboration support module to predict the potential outcomes of different treatment plans; the intelligent knowledge graph module integrates the prediction results of the simulation analysis and prediction module, combines the latest research results and clinical data, and further enhances decision support.

10. The deep learning-based lung cancer early screening system according to claim 2, characterized in that: The virtual reality education module uses the 24 / 7 health consultation service provided by the intelligent health assistant module through natural language processing technology to provide an immersive education experience for users and medical staff; the real-time monitoring feedback module monitors the user's health data and screening progress in real time based on the educational content provided by the virtual reality education module, and provides instant feedback and health advice; the community support interaction module communicates and supports patients and their families through the health advice generated by the real-time monitoring feedback module; the automatic report generation and distribution module uses the communication platform constructed by the community support interaction module to automatically generate detailed screening reports, and distributes them to users and relevant medical staff through multiple channels.

Images