LSTM-Based ECG Classification for Continuous Monitoring on Personal Wearable Devices.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Continuous cardiac monitoring is crucial for managing heart conditions.
• Wearable devices offer a platform for continuous monitoring but have limited processing power.
• Existing deep learning algorithms for ECG classification are often too computationally intensive for wearables.

Purpose of the Study:

• To develop a novel, lightweight ECG classification algorithm for continuous monitoring on resource-constrained wearable devices.
• To achieve accurate cardiac monitoring without compromising device performance.

Main Methods:

• The proposed algorithm utilizes a combination of wavelet transform and multiple long short-term memory (LSTM) recurrent neural networks.
• This architecture is designed to be computationally efficient.

Main Results:

• The algorithm demonstrated superior ECG classification performance compared to existing methods.
• It meets the timing requirements for continuous, real-time execution on various wearable hardware platforms.

Conclusions:

• The developed algorithm is accurate and lightweight, making it suitable for continuous ECG classification on wearable devices.
• It overcomes the limitations of compute-intensive deep learning approaches for wearable applications.