Patient information integration processing system based on artificial intelligence
By integrating multimodal patient data through an AI-based patient information integration and processing system, the problems of delayed epilepsy diagnosis and lack of individualized treatment in traditional methods have been solved, achieving high-precision diagnosis and personalized treatment, and improving the efficiency and accuracy of diagnosis and treatment.
Patent Information
- Application Number
- CN202511874815.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-03-31
AI Technical Summary
Traditional methods struggle to effectively integrate multimodal patient data, leading to delays in the diagnosis and lack of individualized treatment for brain diseases such as epilepsy. Existing systems are unable to accurately link amygdala abnormalities with the diagnosis and treatment of epilepsy.
An AI-based patient information integration and processing system is adopted, including modules such as multi-source information acquisition, information preprocessing and feature extraction, brain structure analysis, EEG abnormality detection, multimodal information fusion and disease diagnosis, personalized treatment plan recommendation, knowledge base and knowledge graph, user interaction and visualization, etc., and deep learning and reinforcement learning technologies are used for data integration and analysis.
It improves the diagnostic accuracy of brain diseases such as epilepsy, enables personalized treatment, increases diagnostic and treatment efficiency, reduces the risk of misdiagnosis, and promotes brain science research.
Smart Images
Figure CN121768558A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical information technology, specifically to a patient information integration and processing system based on artificial intelligence. Background Technology
[0002] The human brain, as the most complex and intricate organ in the human body, is susceptible to various neurological diseases due to even subtle changes in its structure and function. With the rapid development of modern medical technology, medical imaging techniques (such as MRI, CT, and PET), gene sequencing, physiological signal monitoring, and electronic medical record systems have been widely applied, generating massive amounts of patient data. This data contains a wealth of information about patients' brain structure, function, and disease-related aspects.
[0003] However, traditional methods for manual diagnosis and treatment planning have significant limitations. On the one hand, these heterogeneous, multimodal data sources are widespread and diverse in format, making them difficult for humans to effectively integrate and analyze, easily leading to the omission of crucial information. Taking the analysis of epilepsy patients' conditions as an example, doctors need to consider multiple aspects of information simultaneously, including brain imaging, electroencephalograms (EEGs), genetic testing results, and clinical symptoms. Manual processing is not only time-consuming and labor-intensive but may also lead to errors in judgment due to subjective factors. On the other hand, traditional methods struggle to discover potential correlations between data, failing to fully exploit the data's value, thus resulting in diagnostic delays or a lack of individualized treatment plans.
[0004] Epilepsy, in particular, is a brain disorder with complex and diverse etiologies, pathogenesis, and clinical manifestations. Epileptic seizures involve multiple brain regions and functional pathways, with the amygdala playing a crucial role in the occurrence and spread of epilepsy. The amygdala is an important component of the limbic system and is closely related to functions such as emotion, memory, and neural regulation. Studies have shown that structural and functional abnormalities of the amygdala are closely associated with various types of epileptic seizures, such as temporal lobe epilepsy. However, existing systems are insufficient in processing this complex correlation, failing to accurately link amygdala abnormalities to the diagnosis and treatment of epilepsy. Summary of the Invention
[0005] Purpose of the invention: To provide an artificial intelligence-based patient information integration and processing system to solve the above-mentioned problems existing in the prior art.
[0006] Technical solution: A patient information integration and processing system based on artificial intelligence, comprising eight parts: a multi-source information acquisition module, an information preprocessing and feature extraction module, a brain structure analysis submodule, an electroencephalogram abnormality detection submodule, a multimodal information fusion and disease diagnosis submodule, a personalized treatment plan recommendation submodule, a knowledge base and knowledge graph module, and a user interaction and visualization module.
[0007] The multi-source information acquisition module is used to collect patient information related to brain diseases from different data sources, including medical imaging data, electrophysiological signal data, genomic data, clinical information data, and other physiological signal data. The medical imaging data includes MRI, CT, and PET; the electrophysiological signal data includes EEG and MEG; the genomic data is a gene sequencing report; the clinical information data includes medical history, physical signs, neurological examination results, laboratory test results, medication history, and lifestyle habits; and the other physiological signal data includes electrocardiogram, blood oxygen saturation, and temperature.
[0008] The information preprocessing and feature extraction module cleans, denoises, and standardizes the collected heterogeneous data to improve data quality and consistency. It also uses deep learning, image processing, and signal analysis techniques to extract key features related to brain structure, brain function, and disease characteristics from the raw data.
[0009] The brain structure analysis submodule analyzes medical imaging data based on convolutional neural networks, quantifies the morphological parameters of various structures in the human brain, and identifies structural lesions.
[0010] The EEG abnormality detection submodule uses recurrent neural networks, long short-term memory networks, or Transformer models to analyze EEG and magnetoencephalogram (MEG) signals, automatically identify epileptiform discharges, analyze EEG asymmetry, and perform preliminary localization of epileptic seizure types and origins.
[0011] The multimodal information fusion and disease diagnosis submodule uses multimodal deep learning methods to fuse extracted features from imaging, electrophysiology, genes, and clinical aspects to build a disease prediction model for high-precision diagnosis of brain diseases, including disease type, stage, severity, and prediction of disease progression and recurrence risk.
[0012] The personalized treatment plan recommendation submodule utilizes reinforcement learning and graph neural network technologies to recommend individualized treatment plans based on the patient's specific condition, including drug selection, surgical indication determination, brain stimulation therapy parameter settings, and rehabilitation training plans.
[0013] The knowledge base and knowledge graph module stores and manages information about the structure, function, neurotransmitters, pathophysiological mechanisms of common brain diseases, diagnostic criteria, treatment guidelines, drug information, and clinical trial data of the human brain; it constructs a brain disease knowledge graph to connect different information entities and their relationships, providing background knowledge and reasoning basis for artificial intelligence models.
[0014] The user interaction and visualization module provides an intuitive user interface, allowing doctors to input patient information and view AI analysis results; through 3D visualization technology, it intuitively displays the structure of the human brain, lesion areas, and the occurrence and propagation path of epileptic discharges.
[0015] In a further embodiment, in the information preprocessing and feature extraction module, data cleaning removes noise, missing values, and outliers from the data; noise reduction employs wavelet filtering and Kalman filtering algorithms; and standardization processes data from different sources to make them have the same scale and range.
[0016] In a further embodiment, the information preprocessing and feature extraction module includes: medical image feature extraction, which uses convolutional neural networks to analyze MRI and CT image data and extract morphological parameters, texture features, and spatial features of brain structures; electrophysiological signal feature extraction, which uses time-frequency analysis, statistical analysis, and deep learning models to extract brain rhythms, abnormal discharge patterns, and source localization information from electroencephalogram (EEG) and magnetoencephalogram (MEG) signals; genomics feature extraction, which extracts gene variation information related to neural development, ion channels, and neurotransmitters; and clinical information feature extraction, which directly extracts features from structured clinical information and uses natural language processing technology to mine text from unstructured clinical information, extracting key information and converting it into numerical features.
[0017] In a further embodiment, in the multimodal information fusion and disease diagnosis submodule of the artificial intelligence analysis and diagnosis module, feature fusion adopts feature concatenation and feature weighted summation methods; model training adopts support vector machine, random forest and deep learning model for training; disease diagnosis inputs the fused features of the patient to be diagnosed into the trained model and outputs the diagnosis result and predicted probability of the disease.
[0018] In a further embodiment, in the personalized treatment plan recommendation submodule of the artificial intelligence analysis and diagnosis module, data modeling uses the patient's detailed information and diagnostic results as input to construct a personalized model of the patient; treatment plan search uses reinforcement learning algorithms to search in the corresponding treatment plan space to find the optimal treatment strategy; and plan recommendation recommends personalized treatment plans based on the output of the reinforcement learning model, and provides the expected effect and risk assessment of each treatment plan.
[0019] In a further embodiment, in the knowledge base and knowledge graph module, knowledge collection gathers relevant knowledge about brain diseases from multiple channels such as medical literature, clinical guidelines, and expert experience; knowledge representation adopts ontology and semantic network methods to represent the collected knowledge in a structured form; knowledge graph construction uses different knowledge entities as nodes and the relationships between them as edges to construct a brain disease knowledge graph.
[0020] In a further embodiment, the user interaction and visualization module employs a graphical user interface design method, including a patient information input box, an analysis result display area, and a treatment plan recommendation area. The 3D visualization technology utilizes 3D modeling software to convert medical imaging data into a 3D model, displaying the structure and lesion areas of the human brain. Combined with the source localization information of EEG and MEG signals, it displays the occurrence and propagation path of epileptic discharges. The interactive function allows doctors to rotate, scale, and translate the 3D model, and click on specific areas within the model to view detailed information and analysis results for that area.
[0021] In a further embodiment, during the multi-source information acquisition and preprocessing:
[0022] Image data acquisition and processing adopts the DICOM standard interface to import MRI and CT image data, process them using the corresponding formats, segment brain structures using FreeSurfer and SPM tools, extract quantitative parameters of key regions, and perform denoising, normalization, and registration preprocessing on the image data.
[0023] Electrophysiological data acquisition and processing adopted EDF and BIDS standard formats, and preprocessed them by filtering, segmentation, and artifact removal.
[0024] Genomics data acquisition and processing uses VCF and FASTQ formats to collect gene sequencing data, and performs quality control, alignment, and variant detection preprocessing on the gene sequencing data to extract relevant gene variant information.
[0025] Clinical information collection and processing involves importing structured electronic medical records and extracting key information from unstructured text using natural language processing.
[0026] In a further embodiment, when analyzing medical imaging data, the brain structure analysis submodule focuses on the amygdala and hippocampus, regions closely related to epilepsy, and detects whether their volume, shape, and texture are abnormal.
[0027] In a further embodiment, the multimodal information fusion and disease diagnosis submodule combines the correlation between abnormal features of the amygdala and EEG or MEG signals to improve the diagnostic accuracy of specific types of epilepsy.
[0028] Beneficial Effects: This invention relates to an artificial intelligence-based patient information integration and processing system. The invention has the following beneficial effects:
[0029] Improving diagnostic accuracy: By integrating multimodal information, particularly by closely combining the structure (such as changes in the amygdala) and function (such as abnormal EEG activity) of the human brain, and using advanced AI models for analysis, the diagnostic accuracy for brain diseases such as epilepsy has been significantly improved. For example, by jointly analyzing the imaging features of the amygdala with abnormal discharge patterns in EEG signals, amygdala-related epilepsy can be diagnosed more accurately.
[0030] Achieving personalized treatment: Based on AI's in-depth analysis of patients' individual characteristics, the most suitable treatment plan can be recommended, avoiding a "one-size-fits-all" approach and improving the effectiveness and safety of treatment. For example, based on a patient's genetic background and drug response history, the most suitable drug type and dosage can be recommended, reducing drug side effects.
[0031] Improving diagnostic and treatment efficiency: Automated information integration, analysis, and diagnostic processes significantly reduce doctors' working hours, allowing them to focus on more complex clinical decisions and patient communication. For example, the system can complete the analysis and diagnosis of large amounts of patient data in a short time, providing doctors with rapid and accurate diagnostic results and treatment recommendations.
[0032] Supporting Clinical Decision Making: The system provides powerful support for doctors' decision-making through visualized results and analytical reports, reducing the risk of misdiagnosis and missed diagnosis. Doctors can gain a deeper understanding of patients' conditions and make more rational clinical decisions through intuitive 3D models and detailed analysis reports.
[0033] Advancing brain science research: The vast amount of data and analysis results accumulated systematically can provide valuable data support for the study of the mechanisms of brain diseases, the development of new drugs, and the exploration of new treatment methods. For example, through the analysis of data from a large number of epilepsy patients, new epilepsy pathogenesis mechanisms and therapeutic targets can be discovered, providing new ideas and methods for the treatment of epilepsy. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the overall process of the present invention.
[0035] Figure 2 This is a schematic diagram of the overall composition framework of the present invention. Detailed Implementation
[0036] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.
[0037] The patient information integration and processing system based on artificial intelligence that relates to this invention mainly includes the following modules:
[0038] 1. Multi-source information acquisition module
[0039] This module is responsible for collecting patient information related to brain diseases from different data sources, specifically covering:
[0040] Medical imaging data includes MRI (Magnetic Resonance Imaging), CT (Computed Tomography), and PET (Positron Emission Tomography). MRI provides high-resolution images of brain soft tissue, clearly showing the macroscopic and microscopic structures of the human brain, and can be used to identify structural abnormalities, such as changes in the size, shape, and texture of the amygdala. CT is advantageous in detecting lesions such as brain calcification and hemorrhage. PET can reflect the metabolic activity of the brain and helps to identify areas of functional abnormalities.
[0041] Electrophysiological signal data, such as EEG (electroencephalography) and MEG (magnetoencephalography). EEG records electrical activity on the scalp surface to detect abnormal electrical discharges in the brain, which can be used to identify patterns and origins of epileptic seizures. MEG is a non-invasive brain function testing technique that can more accurately locate the sources of neural electrical activity in the brain, and is of great significance for the precise localization of epileptic foci.
[0042] Genomic data, such as gene sequencing reports, are used to analyze genetic susceptibility associated with brain diseases, including some types of epilepsy. Certain gene mutations can lead to abnormal development of neurons or dysfunction of ion channels in the brain, thereby increasing the risk of developing epilepsy. Analyzing genomic data can provide important genetic evidence for the diagnosis and treatment of epilepsy.
[0043] Clinical information data includes medical history, physical examination, neurological examination results, laboratory test results, medication history, and lifestyle habits. Medical history helps doctors understand the progression of the patient's disease; physical examination and neurological examination results directly reflect the patient's neurological function; laboratory test results (such as blood and cerebrospinal fluid tests) can rule out other diseases that may cause brain symptoms; and medication history and lifestyle habits are important references for assessing the patient's treatment effectiveness and the risk of disease recurrence.
[0044] Other physiological signals, such as electrocardiogram (ECG), blood oxygen saturation, and temperature, are used to assess the patient's overall health and rule out other factors that may affect brain activity. For example, heart disease can lead to insufficient blood supply to the brain, thereby triggering seizures; hypoxemia or fever may also induce seizures.
[0045] 2. Information Preprocessing and Feature Extraction Module
[0046] This module performs preprocessing on the collected heterogeneous data, including cleaning, noise reduction, and standardization, to improve data quality and consistency. The specific processing methods are as follows:
[0047] Data cleaning: Removing noise, missing values, and outliers from data. For example, EEG data may contain noise signals caused by poor electrode contact or external interference, which need to be removed through filtering and other methods.
[0048] Denoising: Use appropriate filtering algorithms (such as wavelet filtering, Kalman filtering, etc.) to denoise the data in order to improve the signal-to-noise ratio of the data.
[0049] Standardization: Standardizing data from different sources to give them the same scale and scope, making it easier for subsequent analysis and comparison.
[0050] Based on preprocessing, deep learning, image processing, and signal analysis techniques are used to extract key features related to brain structure (such as the amygdala, hippocampus, and cortex), brain function (such as EEG rhythms and event-related potentials), and disease characteristics (such as epileptic focus localization and degree of brain atrophy) from the raw data. The specific feature extraction methods are as follows:
[0051] Medical image feature extraction: Deep learning models such as convolutional neural networks (CNNs) are used to analyze MRI, CT, and other image data to extract morphological parameters (such as volume, area, and perimeter), texture features (such as gray-level co-occurrence matrix features and local binary pattern features), and spatial features (such as location and orientation) of brain structures. For specific brain regions such as the amygdala, specialized CNN models can be trained for feature extraction to improve the accuracy and specificity of the features.
[0052] Electrophysiological signal feature extraction: Utilizing time-frequency analysis (such as wavelet transform, short-time Fourier transform, etc.), statistical analysis (such as mean, variance, skewness, etc.), and deep learning models (such as RNN, LSTM, etc.), features such as brainwave rhythms, abnormal discharge patterns, and source localization information are extracted from EEG and MEG signals. For example, wavelet transform can decompose EEG signals into sub-signals of different frequencies, thereby analyzing brainwave activity in different frequency bands.
[0053] Genomics feature extraction: Extracting gene variation information related to neural development, ion channels, and neurotransmitters, such as single nucleotide polymorphisms (SNPs) and insertion / deletion variations (Indels). Bioinformatics methods can be used to analyze gene sequencing data and screen for gene features associated with brain diseases.
[0054] Clinical information feature extraction: For structured clinical information (such as laboratory test results, medication history, etc.), feature extraction can be performed directly; for unstructured clinical information (such as medical history, symptom description, etc.), natural language processing (NLP) technology can be used for text mining to extract key information and transform it into numerical features.
[0055] 3. Artificial Intelligence Analysis and Diagnosis Module
[0056] This module is the core of the system and includes the following sub-modules:
[0057] Brain structure analysis submodule
[0058] Based on deep learning models such as convolutional neural networks (CNNs), medical imaging data is analyzed to quantify the morphological parameters of various brain structures (especially the amygdala and hippocampus, regions closely related to epilepsy) and identify structural lesions. The specific implementation steps are as follows:
[0059] Data input: Input the preprocessed medical image data into the CNN model.
[0060] Feature extraction: CNN models automatically extract features from image data through operations such as convolutional layers and pooling layers.
[0061] Lesion identification: Fully connected layers and classifiers are used to classify extracted features to identify whether there are lesions in brain structures, as well as the type and location of the lesions. For example, for the amygdala, it can detect whether its volume is abnormal, its shape is distorted, or its texture is altered.
[0062] Electroencephalological Abnormality Detection Submodule
[0063] Using models such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), or Transformers, EEG / MEG signals are analyzed to automatically identify epileptiform discharges, analyze brain electrical asymmetry, and preliminarily locate the type and origin of epileptic seizures. The specific implementation steps are as follows:
[0064] Data input: Input the preprocessed EEG / MEG signals into the model.
[0065] Feature learning: RNN, LSTM or Transformer models can learn the time series features of a signal and capture the dynamic changes in the signal.
[0066] Anomaly detection: The trained model classifies the input signals to determine the presence of epileptiform discharges and analyzes the discharge patterns and characteristics. Simultaneously, source localization algorithms (such as dipole localization and distributed source localization) can be used to preliminarily locate the origin of epileptic seizures.
[0067] Multimodal information fusion and disease diagnosis submodule
[0068] A multimodal deep learning approach is employed to fuse extracted features from imaging, electrophysiology, genetics, and clinical data to construct a disease prediction model. This model can provide high-precision diagnosis of brain diseases such as epilepsy based on integrated patient information, including disease type, stage, severity, and prediction of disease progression and recurrence risk. In particular, by combining the correlation between amygdala abnormalities and EEG / MEG signals, the diagnostic accuracy for specific types of epilepsy (such as temporal lobe epilepsy) can be improved. The specific implementation steps are as follows:
[0069] Feature fusion: This involves fusing features from different modalities using methods such as feature concatenation and weighted summation. For example, medical imaging features, electrophysiological signal features, genomic features, and clinical information features can be concatenated to form a high-dimensional feature vector.
[0070] Model training: The disease prediction model is trained using the fused feature vectors. Support vector machines (SVM), random forests, and deep learning models (such as multilayer perceptrons, combinations of convolutional neural networks and recurrent neural networks) can be used for training.
[0071] Disease diagnosis: Input the fused features of the patient to be diagnosed into the trained model, and output the diagnosis result and predicted probability of the disease.
[0072] Personalized treatment plan recommendation submodule
[0073] Based on a confirmed diagnosis, this submodule utilizes techniques such as reinforcement learning and graph neural networks to recommend individualized treatment plans according to the patient's specific circumstances (including disease characteristics, genetic background, drug response history, and sensitivity prediction to different treatment methods). The plans may include drug selection (dosage and type), surgical indication assessment, parameter settings for brain stimulation therapy (such as DBS and VNS), and rehabilitation training plans, aiming to optimize treatment outcomes and reduce side effects. The specific implementation steps are as follows:
[0074] Data modeling: Using detailed patient information and diagnostic results as input, a personalized model of the patient is constructed. Techniques such as graph neural networks can be used to represent different patient characteristics (such as brain structure, electrophysiological activity, genetic information, etc.) as graph structures to better capture the relationships between features.
[0075] Treatment Plan Search: Utilizing reinforcement learning algorithms, the system searches the entire space of possible treatment plans to find the optimal treatment strategy. The reinforcement learning model learns the reward values of different treatment plans by continuously interacting with the environment, thereby selecting the best treatment plan.
[0076] Recommended Treatment Plan: Based on the output of the reinforcement learning model, personalized treatment plans are recommended. The expected effects and risk assessments for each treatment plan are also provided for physicians' reference.
[0077] 4. Knowledge Base and Knowledge Graph Module
[0078] This module stores and manages information on the human brain's structure, function, neurotransmitters, pathophysiological mechanisms of common brain diseases (including different types of epilepsy), diagnostic criteria, treatment guidelines, drug information, and clinical trial data. By constructing a brain disease knowledge graph, it connects different information entities (such as brain regions, diseases, genes, and drugs) and their relationships, providing richer background knowledge and reasoning basis for artificial intelligence models, thus enhancing the models' interpretability and reliability. The specific construction method is as follows:
[0079] Knowledge gathering: Collect relevant knowledge about brain diseases from multiple sources, including medical literature, clinical guidelines, and expert experience.
[0080] Knowledge representation: Using methods such as ontology and semantic networks, the collected knowledge is represented in a structured form, which facilitates computer processing and reasoning.
[0081] Knowledge graph construction: Different knowledge entities (such as brain regions, diseases, genes, drugs, etc.) are used as nodes, and the relationships between them (such as causal relationships, treatment relationships, association relationships, etc.) are used as edges to construct a knowledge graph of brain diseases. For example, the knowledge graph can represent the association between "amygdala abnormality" and "temporal lobe epilepsy", as well as the treatment relationship between "a certain drug" and "treatment of a certain type of epilepsy".
[0082] 5. User Interaction and Visualization Module
[0083] This module provides an intuitive user interface, allowing doctors to input patient information and view AI analysis results (such as brain structure maps, abnormal EEG areas, diagnostic reports, and treatment recommendations). Through 3D visualization technology, it visually displays the human brain structure, lesion areas, and the occurrence and propagation pathways of epileptic discharges, helping doctors to gain a deeper understanding of the condition. The specific implementation method is as follows:
[0084] User Interface Design: Employ a graphical user interface (GUI) design approach to create a simple and user-friendly interface. The interface should include patient information input fields, an analysis results display area, and a treatment plan recommendation area.
[0085] 3D visualization technology: Using 3D modeling software (such as VTK, ITK-SNAP, etc.), medical imaging data is converted into 3D models, intuitively displaying the structure and lesion areas of the human brain. Simultaneously, by combining source localization information from EEG and MEG signals, the occurrence and propagation paths of epileptic discharges can be displayed on the 3D model, helping doctors to more intuitively understand the pathogenesis of epilepsy.
[0086] Interactive functionality: The system provides interactive features, allowing doctors to rotate, scale, and translate the 3D model to observe the brain structure and lesion areas from different angles. Doctors can also click on specific areas in the model to view detailed information and analysis results for that area.
[0087] This system can be deployed on a medical institution's server or provided as a cloud service. The following are the specific implementation steps:
[0088] 1. Multi-source information acquisition and preprocessing
[0089] Image data acquisition and processing: MRI, CT, and other imaging data are imported using the DICOM standard interface. NIfTI and other formats are used for processing. Tools such as FreeSurfer and SPM are used for brain structure segmentation, extracting quantitative parameters such as volume, shape, and cortical thickness of key regions like the amygdala and hippocampus. Simultaneously, image data preprocessing is performed, including denoising, normalization, and registration, to improve data quality and consistency.
[0090] Electrophysiological data acquisition and processing: EEG / MEG data are processed in standard formats such as EDF and BIDS. Preprocessing includes filtering, segmentation, and artifact removal. For example, bandpass filtering is used to remove power frequency interference and low-frequency drift from EEG signals, and independent component analysis (ICA) is used to remove artifacts from electrooculography (EOG) and electromyography (EMG).
[0091] Genomics data acquisition and processing: Gene sequencing data were acquired using formats such as VCF and FASTQ. Preprocessing of the gene sequencing data included quality control, alignment, and variant detection to extract gene variation information related to neural development, ion channels, and neurotransmitters.
[0092] Clinical information collection and processing: Importing structured electronic medical records and extracting key information from unstructured text using natural language processing (NLP). For example, techniques such as named entity recognition and relation extraction are used to extract key information, such as disease name, symptoms, and medication history, from medical history and symptom descriptions.
[0093] 2. Feature Extraction
[0094] Medical image feature extraction: Deep learning models such as CNNs are used to extract features from preprocessed image data. Models can be specifically trained to target features such as the amygdala.
[0095] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.
Claims
1. An artificial intelligence-based patient information integration processing system, characterized by Comprise: Multi-source information collection module, for collecting patient information related to brain diseases from different data sources, including medical image data, electrophysiological signal data, genomics data, clinical information data and other physiological signal data; Information preprocessing and feature extraction module, the collected heterogeneous data is cleaned, denoised, standardized and pretreated to improve data quality and consistency; and deep learning, image processing and signal analysis are used to extract key features related to brain structure, brain function and disease characteristics from raw data; Brain structure analysis sub-module, based on convolutional neural network, medical image data is analyzed, morphological parameters of each structure of human brain are quantified, and structural lesions are identified; Electrophysiological abnormality detection sub-module, using recurrent neural network, long short-term memory network or Transformer model, electroencephalogram and magnetoencephalogram signals are analyzed, epileptiform discharges are automatically identified, electroencephalogram asymmetry is analyzed, and seizure type and origin are preliminarily located; Multi-modal information fusion and disease diagnosis sub-module, using multi-modal deep learning method, the extracted features from image, electrophysiology, gene and clinical aspects are fused to construct disease prediction model for high-precision diagnosis of brain diseases, including disease type, stage, severity, and prediction of disease progression and recurrence risk; Personalized treatment recommendation sub-module, using reinforcement learning and graph neural network technology, individualized treatment plan is recommended according to patient's specific condition, including drug selection, surgical indication judgment, brain stimulation treatment parameter setting and rehabilitation training plan; Knowledge base and knowledge graph module, store and manage the information about brain structure, function, neurotransmitter, common brain disease pathophysiological mechanism, diagnostic criteria, treatment guidelines, drug information and clinical trial data; construct brain disease knowledge graph, connect different information entities and their relationships, and provide background knowledge and reasoning basis for artificial intelligence model; User interaction and visualization module, provide intuitive user interface, allow doctors to input patient information and view AI analysis results; through three-dimensional visualization technology, brain structure, lesion area and epileptiform discharge occurrence and propagation path are displayed intuitively.
2. The patient information integration processing system based on artificial intelligence according to claim 1, wherein: In the information preprocessing and feature extraction module, data cleaning removes noise, missing values and outliers in the data; wavelet filtering and Kalman filtering algorithm are used for denoising; and standardization is performed on data from different sources to make them have the same scale and range.
3. The patient information integration processing system based on artificial intelligence according to claim 1, wherein: In the information preprocessing and feature extraction module, medical image feature extraction uses convolutional neural network to analyze MRI and CT image data, extracts morphological parameters, texture features and spatial features of brain structure; The electro-physiological signal feature extraction extracts the electroencephalogram rhythm, abnormal discharge mode, and source positioning information from electroencephalogram and magnetoencephalogram signals by using time-frequency analysis, statistical analysis, and deep learning models; the genomics feature extraction extracts gene variation information related to neural development, ion channels, and neurotransmitters; and the clinical information feature extraction directly extracts features from structured clinical information, and extracts key information and converts the key information into numerical features by using natural language processing technology for text mining on unstructured clinical information.
4. The patient information integration processing system based on artificial intelligence according to claim 1, characterized in that: In the multi-modal information fusion and disease diagnosis sub-module of the artificial intelligence analysis and diagnosis module, feature fusion adopts feature splicing and feature weighted summation method; model training adopts support vector machine, random forest, and deep learning model for training; and disease diagnosis inputs the fusion features of the patient to be diagnosed into the trained model to output the diagnosis result and prediction probability of the disease.
5. The patient information integration processing system based on artificial intelligence according to claim 1, characterized in that: In the personalized treatment plan recommendation sub-module of the artificial intelligence analysis and diagnosis module, data modeling takes the detailed information and diagnosis result of the patient as input to construct a personalized model of the patient; treatment plan search uses reinforcement learning algorithm to search in the corresponding treatment plan space to find the corresponding optimal treatment strategy; and plan recommendation recommends a personalized treatment plan according to the output of the reinforcement learning model, and gives the expected effect and risk assessment of each treatment plan.
6. The patient information integration processing system based on artificial intelligence according to claim 1, characterized in that: In the knowledge base and knowledge graph module, knowledge collection collects relevant knowledge about brain diseases from multiple channels such as medical literature, clinical guidelines, and expert experience; knowledge representation adopts ontology and semantic network method to represent the collected knowledge in a structured form; and knowledge graph construction constructs a brain disease knowledge graph by taking different knowledge entities as nodes and their relationships as edges.
7. The patient information integration processing system based on artificial intelligence according to claim 1, characterized in that: In the user interaction and visualization module, the user interface design adopts graphical user interface design method, and the interface includes patient information input box, analysis result display area, and treatment plan recommendation area; The three-dimensional visualization technology converts medical image data into a three-dimensional model by using three-dimensional modeling software to display the structure of the human brain and the lesion area, and combines the source positioning information of electroencephalogram and magnetoencephalogram signals to display the occurrence and propagation path of epileptic discharge; and the interactive function allows doctors to rotate, zoom, and translate the three-dimensional model, and click on a specific area in the model to view detailed information and analysis results of the area.
8. The patient information integration processing system based on artificial intelligence according to claim 1, characterized in that: In the multi-source information collection and preprocessing, The image data acquisition and processing adopts the DICOM standard interface to import MRI and CT image data, processes them in corresponding formats, uses FreeSurfer and SPM tools to segment brain structures, extracts quantitative parameters of key regions, and performs denoising, normalization, and registration preprocessing on the image data. The electrophysiological data acquisition and processing adopts EDF and BIDS standard formats for filtering, segmentation, and artifact rejection preprocessing. The genomics data acquisition and processing adopts VCF and FASTQ formats for gene sequencing data acquisition, quality control, alignment, and variant detection preprocessing, and extracts relevant gene variant information. The clinical information acquisition and processing imports structured electronic medical records and extracts key information from unstructured text through natural language processing.
9. The patient information integration and processing system based on artificial intelligence according to claim 1, characterized in that: In the brain structure analysis submodule, when analyzing medical image data, the amygdala, hippocampus, and regions closely related to epilepsy are focused on, and the volume, shape, and texture of these regions are detected for abnormalities.
10. The patient information integration and processing system based on artificial intelligence according to claim 1, characterized in that: In the multi-modal information fusion and disease diagnosis submodule, the abnormal features of the amygdala and the relevance of electroencephalogram or magnetoencephalogram signals are combined to improve the diagnostic accuracy of specific types of epilepsy.