Continuous ECG Signal Pathological Detection via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for electrocardiographic (ECG) signal analysis primarily focus on classifying single heartbeats and lack effective solutions for pathological identification of continuous ECG signals.
Innovation Solution
A method and apparatus utilizing convolutional neural networks to segment and analyze continuous ECG signals, determining feature data for each heartbeat, and identifying pathological categories, with the ability to locate abnormal heartbeats by segmenting the signal and inputting data to separate neural networks for detailed analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single heartbeat classification methods are used, then the analysis process is simple, but pathological identification of continuous ECG signals cannot be performed
Solution Approach 1:
The continuous ECG signal is segmented into multiple single heartbeat signals with overlapping time windows. Each segmented heartbeat is processed independently by the classification model, and the results are aggregated to achieve continuous signal pathological identification. This segmentation approach allows the simple single-heartbeat classifier to be applied to continuous signals while capturing temporal relationships through the overlapping windows.
2Measurement precision
If continuous ECG signals are analyzed directly, then pathological diagnosis accuracy is improved, but the analysis complexity increases
Solution Approach 1:
The continuous ECG signal is divided into multiple overlapping single heartbeat segments. Each segment is processed by a pre-trained single-heartbeat classification model, converting a complex continuous signal analysis problem into multiple simpler single-heartbeat classification tasks. The final diagnosis is obtained by aggregating results from all segments, thus improving accuracy while managing complexity through division.
Solution Approach 2:
A single-heartbeat classification model is pre-trained beforehand using training data. This preliminary training allows the model to learn pathological patterns in advance, so that when analyzing continuous signals, the pre-trained model can be directly applied to segmented data without requiring complex real-time processing, thereby improving efficiency and accuracy.
3Measurement precision
If multiple neural networks are used for detailed analysis, then abnormal heartbeat location precision is improved, but device complexity increases
Solution Approach 1:
The analysis system is segmented into two independent neural networks: a first convolutional neural network for extracting features from single heartbeat signals, and a second convolutional neural network for classifying pathological categories. This segmentation allows each network to specialize in a specific task, improving location precision and classification accuracy while managing overall system complexity through modular architecture.
Solution Approach 2:
Feature data extracted by the first convolutional neural network serves as an intermediary between the raw ECG signal and the final pathological classification. This intermediate feature representation captures essential characteristics of the heartbeat signal, enabling the second neural network to perform accurate classification without directly processing raw signals, thus improving precision while maintaining manageable complexity.
Data Source
AI summary
The present application provides a method and apparatus for detecting an ECG signal and an electronic device. According to an example of the method, an ECG signal with a set time length is segmented to obtain a first set number of single heartbeats; feature data corresponding to each of the first set number of single heartbeats is determined to obtain a first set number of feature data; and a pathological category of the ECG signal with the set time length is determined based on the ECG signal with the set time length and the first set number of feature data.