# Two-stream Network for ECG Signal Classification

**Authors:** Xinyao Hou, Shengmei Qin, Jianbo Su  
**arXiv:** [2210.06293](https://arxiv.org/abs/2210.06293)  
**Date:** 2022-10-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.06293v1/

## Key points

- Deep learning algorithm for automatic classification of multi-classes of heartbeat types based on ECG data
- Two-stream architecture incorporating identified and temporal stream networks
- Identified network extracts features of individual heartbeats
- Temporal network captures temporal variations in ECG signals
- Achieves 99.38% accuracy on MIT-BIH Arrhythmia Database and 88.07% positive accuracy on real-life data
- Importance of quick, accurate, and robust analysis of ECG data due to variety of heart diseases and volume of data
- Need for automated analysis methods to reduce errors and resource consumption
- Various feature extraction techniques used in ECG signal analysis
- Proposed algorithm improves health surveillance capability by accurately categorizing heart diseases using computer analysis
- Feasibility of determining diseased organs through heartbeats to enhance medical treatment

## Summary

This paper presents a deep learning algorithm for automatic classification of multi-classes of heartbeat types based on electrocardiogram (ECG) data. The authors address the challenge of analyzing the increasing quantity of ECG data, which consumes a lot of medical resources. They propose a two-stream architecture that incorporates identified and temporal stream networks to extract features and temporal correlations between heartbeats. The identified network focuses on extracting features of individual heartbeats, while the temporal network aims to capture the temporal variations in ECG signals. The proposed algorithm achieves an accuracy of 99.38% on the MIT-BIH Arrhythmia Database and an 88.07% positive accuracy on massive real-life data, demonstrating its high diagnostic performance in categorizing different classes of heartbeats. The authors highlight the importance of quick, accurate, and robust analysis of ECG data due to the wide variety of heart diseases and the sheer volume of ECG data. Manual diagnosis can introduce errors and consume significant medical resources; therefore, there is an urgent need for automated analysis methods. Machine learning techniques have been widely used in ECG signal analysis with feature extraction being a crucial step in diagnosing heart disease. Various feature extraction techniques are employed such as detection algorithms for the P-QRS-T complex and statistical attributes or frequency-domain-based techniques. The proposed algorithm contributes to improving health surveillance capability by accurately categorizing heart diseases using computer analysis. It also suggests that determining diseased organs through heartbeats is feasible which can further enhance medical treatment. In conclusion, this paper presents a deep ECG signal classification model with a two-stream architecture that effectively categorizes different classes of heartbeat types. The model extracts identity features and static features from individual heartbeats while monitoring temporal variations in ECG signals.