Management system based on medical electrophysiological examination mass data analysis

Abstract

The invention discloses a medical technology electrophysiological examination mass data analysis-based management system, and relates to the field of data analysis, and the system comprises an equipment access and standardization module which is used for accessing multiple types of electrophysiological examination equipment of different manufacturers into the system; the data acquisition and centralized storage module is used for acquiring original waveform data, physiological parameters and preliminary report content; the big data processing and analysis module is responsible for performing data cleaning, feature extraction, annotation training and model prediction on the acquired data and supporting multi-source heterogeneous fusion and edge-cloud cooperative processing; the intelligent auxiliary diagnosis module is used for generating structured diagnosis suggestions and report contents based on the analysis result, and carrying out waveform comparison and result visual presentation; and the medical technology workflow management module runs through inspection process management, progress synchronization, quality control feedback and reservation optimization. A Reed-Solomon coding algorithm is introduced, redundant backup and automatic error correction are performed on collected data, and data integrity and recoverability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and in particular to a management system based on analysis of massive data from medical electrophysiological examinations. Background Art

[0002] With the development of the medical industry, hospitals have gradually shifted from a model of pursuing scale expansion and extensive growth to one that prioritizes patients and achieves sustainable development through innovative management and cost control. In this transformation, leveraging information technology to optimize resource allocation, reshape hospital business processes, and implement standardization and regularization is essential and a key tool for rapidly improving the patient experience.

[0003] With the growth of medical services and the expansion of examination procedures, electrophysiological examination data has increased annually. Waveform data generated for each patient can range from tens of seconds to several minutes, generating tens of GB of data daily. This data includes a variety of data types, including waveforms, parameters, and text. Waveforms contain subtle characteristic changes that precede disease onset, but existing systems lack the ability to extract them. Currently, most electrophysiological data is used only for immediate diagnosis and is not further utilized for advanced scenarios such as decision support, trend prediction, disease classification, and scientific research modeling. Summary of the Invention

[0004] The present invention proposes a management system based on analysis of massive data from medical electrophysiological examinations to solve the problems mentioned in the above-mentioned prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a management system based on massive data analysis of medical electrophysiological examinations, comprising: The device access and standardization module is used to connect multiple types of electrophysiological examination devices from different manufacturers to the system, automatically identify the device ID, model, and firmware version, match the corresponding communication protocol, and deploy the corresponding protocol parser through a containerized plug-in mechanism to realize data conversion into a unified intermediate data model; The data acquisition and centralized storage module is used to collect raw waveform data, physiological parameters, and preliminary report content. It achieves efficient data writing through a hot and cold data stratification mechanism, Kafka partition consumption, and backpressure control. It also uses Reed-Solomon encoding for data redundancy and error correction, and uses an inverted index and B-Tree structure to enable fast retrieval of high-frequency query fields. The big data processing and analysis module is responsible for performing data cleaning, feature extraction, annotation training and model prediction on the collected data, supporting multi-source heterogeneous fusion and edge-cloud collaborative processing; The intelligent auxiliary diagnosis module generates structured diagnostic suggestions and report content based on the analysis results, and performs waveform comparison and result visualization; The medical technology workflow management module covers examination process management, progress synchronization, quality control feedback and appointment optimization.

[0006] Preferably, the big data processing and analysis module includes: The data quality control submodule is used to perform time series alignment, value range verification, and outlier elimination based on statistical analysis on the collected data; The intelligent labeling and incremental learning submodule performs data labeling based on active learning and semi-supervised algorithms, supporting online learning and consistency constraint mechanisms; The feature extraction and dimensionality reduction submodule combines fast Fourier transform, wavelet analysis and PCA / t-SNE algorithm to extract time domain, frequency domain and time-frequency domain features and reduce the dimensionality.

[0007] Preferably, the big data processing and analysis module further includes: The multi-source heterogeneous data fusion submodule integrates patient examination history, electronic medical records, and imaging data to construct a patient knowledge graph with spatiotemporal dimensions and generate graph embedding vectors for use by downstream models. The model training and adaptation submodule uses transfer learning and domain adaptation strategies to train models, has interpretable output capabilities, and supports cross-center model training based on a federated learning architecture; The resource optimization and edge collaboration submodule combines Kubernetes cluster scheduling strategies with network delay prediction algorithms to achieve dynamic load balancing and flexible deployment of edge nodes and cloud tasks. The resource optimization and edge collaboration submodule adjusts the task allocation strategy in advance based on network delay prediction results, giving priority to executing high-priority tasks on low-latency nodes. The real-time streaming processing submodule builds a streaming computing channel based on Kafka and Flink, supports low-latency event processing, and has the ability to dynamically route messages to different analysis paths based on signal type and priority. The real-time streaming processing submodule dynamically routes the transmitted data to different analysis channels based on signal characteristics and task priority.

[0008] Preferably, the device access and standardization module implements hot-swap updates of the protocol adapter through a containerized plug-in mechanism, supporting subsequent device access without restarting the system.

[0009] Preferably, the data collection and centralized storage module utilizes the Kafka partitioning mechanism to perform high-concurrency writing during the collection process, and uses a back pressure strategy to avoid data loss or processing blockage caused by accumulation during peak periods.

[0010] Preferably, the data acquisition and centralized storage module uses Reed-Solomon coding to perform redundant backup and automatic error correction on data slices.

[0011] Preferably, the intelligent annotation and incremental learning submodule provides an expert annotation tool based on a graphical interface, supports multi-label annotation, region selection and voice annotation, and improves the efficiency and accuracy of manual participation.

[0012] Preferably, the feature extraction and dimensionality reduction submodule further includes a feature supplementation mechanism based on high-order statistics, and uses indicators such as Kurtosis and Skewness to enhance the model's ability to recognize low-frequency abnormal patterns.

[0013] Preferably, the multi-source heterogeneous data fusion submodule introduces a spatiotemporal path analysis method to model the propagation path of the lesion nodes in the knowledge graph.

[0014] Preferably, the model training and adaptation submodules collaboratively train a shared model among multiple medical institutions through a federated learning mechanism, thereby avoiding leakage of patient privacy while improving generalization capabilities.

[0015] Compared with the existing technology, the beneficial effects of the present invention are: (1) The Kafka partition consumption and back pressure mechanism is adopted to effectively alleviate the risk of data accumulation during peak periods and ensure real-time performance and system stability; combined with the hot and cold data stratification mechanism, the original waveforms, structured parameters, and preliminary reports are classified and stored to improve access efficiency and cost control; the Reed-Solomon encoding algorithm is introduced to perform redundant backup and automatic error correction of collected data, thereby improving data integrity and recoverability.

[0016] (2) Construct a complete analysis chain including data cleaning, intelligent labeling, feature extraction, model training, and predictive reasoning to support multi-source data fusion and feature enhancement; introduce strategies such as active learning, incremental learning, and semi-supervised learning to solve the problems of low efficiency of manual labeling and weak model generalization ability; use high-order statistics (such as Kurtosis, Skewness) and graph embedding technology to improve the ability to identify abnormal patterns and model lesion propagation.

[0017] (3) Support federated learning and edge-cloud collaboration to enhance data security and resource scheduling capabilities; support cross-institutional federated learning to improve the diversity and generalization capabilities of model training samples while ensuring data privacy and security; combine edge computing with cloud resource collaboration mechanisms to achieve flexible task deployment and low-latency processing to meet real-time prediction and response needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic block diagram of a management system based on massive data analysis of medical electrophysiological examinations proposed by the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] A management system based on massive data analysis of medical electrophysiological examinations, including: The device access and standardization module is used to connect multiple types of electrophysiological examination equipment from different manufacturers to the system, automatically identify the device ID, model and firmware version, match the corresponding communication protocol, and deploy the corresponding protocol parser through the containerized plug-in mechanism to realize data conversion into a unified intermediate data model; the device access and standardization module realizes hot-swap updates of protocol adapters through the containerized plug-in mechanism, and supports subsequent device access without restarting the system.

[0021] The data acquisition and centralized storage module collects raw waveform data, physiological parameters, and preliminary report content, and achieves efficient data writing through a hot and cold data tiering mechanism, Kafka partition consumption, and backpressure control. Furthermore, it uses Reed–Solomon encoding for data redundancy and error correction, and an inverted index and B-Tree structure for fast retrieval of high-frequency query fields. During the acquisition process, the module leverages the Kafka partitioning mechanism for high-concurrency writing and employs a backpressure strategy to avoid data loss or processing blockage caused by peak accumulation. The module also employs Reed–Solomon encoding for redundant data backup and automatic error correction of data shards.

[0022] The module specifically includes the following technical designs: Structuring of data collection content and format This module is used to collect and receive, in real time, raw waveform data (such as ECG and EEG signals) output by electrophysiological devices, physiological parameter data (such as heart rate, blood pressure, and pulse rate), and the structured or semi-structured examination report content initially generated by the device (such as preliminary diagnostic text and charts). The acquisition interface supports multiple communication protocols (such as TCP / IP, serial ports, and HL7) and converts the data into a unified intermediate data structure for subsequent processing and analysis.

[0023] Hot and cold data tiering mechanism Collected data is automatically categorized into "hot data" and "cold data" based on access frequency and timeliness. Hot data refers to active data that is frequently accessed within seven days or is still within the analysis and processing cycle. It is stored in high-performance SSD storage systems, supporting low-latency reads. Cold data, on the other hand, refers to historical data that has exceeded a certain age limit and is infrequently accessed. It is regularly migrated to high-capacity, low-cost object storage systems (such as distributed HDFS or Ceph clusters). This hot and cold tiering strategy optimizes storage costs and response speeds.

[0024] Kafka partition consumption mechanism The module integrates Apache Kafka as a high-throughput data pipeline, buffering and decoupling collected data streams. Each type of inspection data is divided into multiple topics, each of which is further divided into multiple partitions. Multiple consumers concurrently consume data from each partition, effectively achieving horizontal scalability and supporting parallel data writes to tens of thousands of devices.

[0025] Backpressure control strategy To prevent data volumes from exceeding backend processing capacity during peak periods (such as large screening days), the system employs a backpressure mechanism. When the consumer detects that the processing queue backlog has reached a threshold, it proactively reduces the data write rate to the Kafka producer, preventing system memory bloat or downtime, and ensuring system stability and data integrity.

[0026] Data integrity and redundancy (Reed–Solomon encoding) To ensure data integrity and reliability during transmission and storage, the module shards each batch of data and calculates redundant check blocks based on the Reed–Solomon coding algorithm. This technology, a form of erasure coding, enables accurate recovery even when the original data is partially damaged. Even if some shards are lost due to network jitter or disk failure, the system can reconstruct the original data using the check blocks, thereby improving data availability.

[0027] Index mechanism design (inverted index and B-Tree) To support high-speed queries for subsequent retrieval and analysis tasks, the module constructs two types of index structures for key fields (such as patient ID, examination time, equipment model, waveform type, etc.) during data storage: Inverted index, suitable for fast keyword matching of text fields, such as disease terms mentioned in report content; B-Tree indexes are suitable for range queries and sorting operations, such as retrieving all inspection records generated for a certain device model within the last 30 days.

[0028] The dual-index mechanism ensures that the system can maintain millisecond-level retrieval response when processing tens of millions of data.

[0029] High availability and distributed architecture The data collection and storage modules are deployed in a containerized cluster environment, featuring multi-node redundancy and automatic fault tolerance. Sub-services communicate through a service mesh, enabling faulty node removal and automatic migration. Core data service nodes are configured with a master-slave synchronization mechanism to ensure data continuity and uninterrupted service in the event of a single point of failure.

[0030] The big data processing and analysis module is responsible for performing data cleaning, feature extraction, annotation training and model prediction on the collected data, supporting multi-source heterogeneous fusion and edge-cloud collaborative processing; The big data processing and analysis module includes: The data quality control submodule is used to perform time series alignment, value range verification, and outlier elimination based on statistical analysis on the collected data; This submodule is mainly used to ensure the accuracy, consistency and integrity of collected data. The core technologies include: Timing alignment processing: For data streams from different devices or with inconsistent acquisition cycles, a sliding window mechanism and reference timestamp comparison technology are used to automatically synchronize the waveform data in the time domain to ensure cross-source data comparability; Value range verification: The system has a built-in standardized physiological parameter range table, which dynamically matches the patient's age, gender and other basic information and marks out-of-range data points; Anomaly elimination based on statistical analysis: Calculate statistical features such as mean, standard deviation, and quartiles for each waveform or parameter sequence, and combine the Z-score and IQR (interquartile range) methods to automatically eliminate noise, drift, and sudden changes, preventing false data from entering the training process.

[0031] The intelligent labeling and incremental learning submodule performs data labeling based on active learning and semi-supervised algorithms, and supports online learning and consistency constraint mechanisms. The intelligent labeling and incremental learning submodule provides an expert labeling tool based on a graphical interface, supports multi-label labeling, region selection, and voice annotation, and improves the efficiency and accuracy of manual participation.

[0032] This submodule combines manual annotation with machine learning algorithms to efficiently construct data labels, specifically including: Active learning mechanism: The model automatically selects uncertain samples near the judgment boundary and prompts experts to prioritize labeling, thereby improving sample utilization; Semi-supervised learning algorithm: uses pseudo-label propagation, graph neural network and other methods to self-learn unlabeled data, alleviating the burden of expert labeling; Online learning and consistency constraints: The system dynamically updates model parameters as new samples are received, while maintaining prediction consistency between the old and new models to prevent model drift. Visual expert annotation interface: embedded graphical annotation tools, supporting the following functions: Multi-label annotation: a piece of data can be labeled with multiple clinically significant labels at the same time; Area marking: Use the mouse to select or drag to mark a specific time period or frequency band; Voice annotations: Experts can supplement clinical instructions through voice input, which is automatically transcribed and labeled to improve efficiency and traceability.

[0033] The feature extraction and dimensionality reduction submodule combines fast Fourier transforms, wavelet analysis, and PCA / t-SNE algorithms to extract features from the time, frequency, and time-frequency domains and perform dimensionality reduction. This submodule also includes a feature supplementation mechanism based on high-order statistics, using metrics such as Kurtosis and Skewness to enhance the model's ability to identify low-frequency anomalies.

[0034] This submodule focuses on extracting discriminative electrophysiological signal features and supports the construction of multi-dimensional modeling inputs: Traditional signal processing methods: Time domain characteristics: calculate mean, variance, peak, periodicity and other indicators; Frequency domain features: extract the main frequency distribution and frequency band energy through fast Fourier transform (FFT); Time-frequency domain features: Extract frequency changes within a specific time window based on continuous / discrete wavelet transform.

[0035] Dimensionality reduction: PCA (Principal Component Analysis): compresses feature dimensions and retains the main information; t-SNE: used for high-dimensional feature visualization and outlier clustering.

[0036] High-order statistical feature supplementation mechanism: Use indicators such as Kurtosis and Skewness to capture abnormal patterns of "spikes" and "skewness" in the waveform, effectively improving the ability to identify rare lesions or atypical waveforms.

[0037] The big data processing and analysis module also includes: The multi-source heterogeneous data fusion submodule integrates patient examination history, electronic medical records and imaging data to construct a patient knowledge graph with spatiotemporal dimensions and generate graph embedding vectors for use by downstream models. The multi-source heterogeneous data fusion submodule introduces a spatiotemporal path analysis method to model the propagation path of lesion nodes in the knowledge graph.

[0038] This submodule implements semantic fusion and structured graph modeling of patient-related data: Fusion objects: including patient historical electrophysiological records, electronic medical records (EMR), medical images (such as CT, MRI), etc. Data alignment and standardization: Use the FHIR standard to unify and map data structures; Patient knowledge graph construction: Build a relationship network based on the basic structure of "patient-disease-examination-time node"; Node attributes include examination results, physiological characteristics, imaging manifestations, etc. Graph embedding generation: Vector encoding of structured graphs based on graph neural networks (GNN); The generated embedding vector is used as model input to improve the context-awareness of downstream predictions.

[0039] Space-time path analysis algorithm: Perform path tracing on lesion nodes (such as abnormal heart rhythm points) to identify the propagation pattern from initial onset to spread, providing support for dynamic diagnosis.

[0040] The model training and adaptation submodule uses transfer learning and domain adaptation strategies to train models, has interpretable output capabilities, and supports cross-center model training based on a federated learning architecture. The model training and adaptation submodule uses a federated learning mechanism to collaboratively train shared models among multiple medical institutions, avoiding patient privacy leaks while improving generalization capabilities.

[0041] This submodule completes the training and iterative update of the data-driven model: Transfer learning: Based on the existing model, fine-tuning is used to adapt it to new tasks; Domain adaptation: Addresses distribution shifts caused by differences in equipment and patient populations across hospitals. Explainability Mechanism: Introducing Grad-CAM, SHAP and other technical analysis models to focus on the area; The output can explain the reasons to assist physicians in making decisions.

[0042] Federated learning supports: Achieve joint modeling across medical centers without transmitting original data; Parameters are encrypted and transmitted to protect patient privacy while improving model generalization capabilities.

[0043] The resource optimization and edge collaboration submodule combines Kubernetes cluster scheduling strategies with network delay prediction algorithms to achieve dynamic load balancing and flexible deployment of edge nodes and cloud tasks. The resource optimization and edge collaboration submodule adjusts the task allocation strategy in advance based on network delay prediction results, giving priority to executing high-priority tasks on low-latency nodes. This submodule ensures efficient scheduling and resource utilization of model training and inference tasks: Elastic deployment based on Kubernetes: Dynamically scale the number of training and inference service replicas to save resources; Network delay prediction model: Collect inter-node communication delay data and train prediction models; Adjust computing task allocation strategies in advance based on prediction results; Edge-cloud collaboration mechanism: Prioritize scheduling low-latency, high-priority tasks to edge nodes for execution; The cloud is responsible for training and batch processing tasks, and the edge nodes perform real-time inference.

[0044] The real-time streaming processing submodule builds a streaming computing channel based on Kafka and Flink, supports low-latency event processing, and has the ability to dynamically route messages to different analysis paths based on signal type and priority. The real-time streaming processing submodule dynamically routes the transmitted data to different analysis channels based on signal characteristics and task priority.

[0045] This submodule is responsible for data processing and event feedback with low latency: Kafka-Flink stream processing engine: Kafka is used to receive and buffer real-time data; Flink is used for stateful processing and rule triggering; Dynamic routing mechanism: Automatically select the analysis path based on the signal type (such as ECG, EEG) and task priority of each data packet; High-priority or critical value channels have resource preemption capabilities and receive priority responses.

[0046] The intelligent auxiliary diagnosis module generates structured diagnostic suggestions and report content based on the analysis results, and performs waveform comparison and result visualization; The medical technology workflow management module covers examination process management, progress synchronization, quality control feedback and appointment optimization.

[0047] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A management system based on massive data analysis of medical electrophysiological examinations, characterized by: include: The device access and standardization module is used to connect multiple types of electrophysiological examination devices from different manufacturers to the system, automatically identify the device ID, model, and firmware version, match the corresponding communication protocol, and deploy the corresponding protocol parser through a containerized plug-in mechanism to realize data conversion into a unified intermediate data model; The data acquisition and centralized storage module is used to collect raw waveform data, physiological parameters, and preliminary report content. It achieves efficient data writing through a hot and cold data stratification mechanism, Kafka partition consumption, and backpressure control. It also uses Reed-Solomon encoding for data redundancy and error correction, and uses an inverted index and B-Tree structure to enable fast retrieval of high-frequency query fields. The big data processing and analysis module is responsible for performing data cleaning, feature extraction, annotation training and model prediction on the collected data, supporting multi-source heterogeneous fusion and edge-cloud collaborative processing; The intelligent auxiliary diagnosis module generates structured diagnostic suggestions and report content based on the analysis results, and performs waveform comparison and result visualization; The medical technology workflow management module covers examination process management, progress synchronization, quality control feedback and appointment optimization.

2. A management system based on massive data analysis of medical electrophysiological examinations according to claim 1, characterized in that: The big data processing and analysis module includes: The data quality control submodule is used to perform time series alignment, value range verification, and outlier elimination based on statistical analysis on the collected data; The intelligent labeling and incremental learning submodule performs data labeling based on active learning and semi-supervised algorithms, supporting online learning and consistency constraint mechanisms; The feature extraction and dimensionality reduction submodule combines fast Fourier transform, wavelet analysis and PCA / t-SNE algorithm to extract time domain, frequency domain and time-frequency domain features and reduce the dimensionality.

3. A management system based on massive data analysis of medical electrophysiological examinations according to claim 1, characterized in that: The big data processing and analysis module also includes: The multi-source heterogeneous data fusion submodule integrates patient examination history, electronic medical records, and imaging data to construct a patient knowledge graph with spatiotemporal dimensions and generate graph embedding vectors for use by downstream models. The model training and adaptation submodule uses transfer learning and domain adaptation strategies to train models, has interpretable output capabilities, and supports cross-center model training based on a federated learning architecture; The resource optimization and edge collaboration submodule combines Kubernetes cluster scheduling strategies with network delay prediction algorithms to achieve dynamic load balancing and flexible deployment of edge nodes and cloud tasks. The resource optimization and edge collaboration submodule adjusts the task allocation strategy in advance based on network delay prediction results, giving priority to executing high-priority tasks on low-latency nodes. The real-time streaming processing submodule builds a streaming computing channel based on Kafka and Flink, supports low-latency event processing, and has the ability to dynamically route messages to different analysis paths based on signal type and priority. The real-time streaming processing submodule dynamically routes the transmitted data to different analysis channels based on signal characteristics and task priority.

4. A management system based on massive data analysis of medical electrophysiological examinations according to claim 1, characterized in that: The device access and standardization module implements hot-swap updates of protocol adapters through a containerized plug-in mechanism, supporting subsequent device access without restarting the system.

5. A management system based on massive data analysis of medical electrophysiological examinations according to claim 1, characterized in that: The data collection and centralized storage module uses the Kafka partitioning mechanism to perform high-concurrency writing during the collection process, and uses a back pressure strategy to avoid data loss or processing blockage caused by accumulation during peak periods.

6. A management system based on massive data analysis of medical electrophysiological examinations according to claim 1, characterized in that: The data acquisition and centralized storage module uses Reed-Solomon coding to perform redundant backup and automatic error correction on data slices.

7. A management system based on massive data analysis of medical electrophysiological examinations according to claim 2, characterized in that: The intelligent labeling and incremental learning submodule provides an expert labeling tool based on a graphical interface, supports multi-label labeling, region selection and voice annotation, and improves the efficiency and accuracy of manual participation.

8. A management system based on massive data analysis of medical electrophysiological examinations according to claim 2, characterized in that: The feature extraction and dimensionality reduction submodule also includes a feature supplementation mechanism based on high-order statistics, which uses indicators such as Kurtosis and Skewness to improve the model's ability to recognize low-frequency abnormal patterns.

9. A management system based on massive data analysis of medical electrophysiological examinations according to claim 3, characterized in that: The multi-source heterogeneous data fusion submodule introduces a spatiotemporal path analysis method to model the propagation path of the lesion nodes in the knowledge graph.

10. A management system based on massive data analysis of medical electrophysiological examinations according to claim 3, characterized in that: The model training and adaptation submodules use a federated learning mechanism to collaboratively train shared models among multiple medical institutions, thereby avoiding the leakage of patient privacy while improving generalization capabilities.

Images