KecNet: A Light Neural Network for Arrhythmia Classification Based on Knowledge Reinforcement.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiology

Background:

• Mobile electrocardiographic (ECG) signal acquisition offers efficiency but deep learning models demand significant resources.
• Developing resource-efficient deep learning algorithms is crucial for mobile health applications.

Purpose of the Study:

• To propose KecNet, a lightweight deep learning scheme for arrhythmia classification tailored to resource-constrained mobile devices.
• To leverage domain knowledge, signal analysis, and medical expertise in designing the KecNet model.

Main Methods:

• KecNet, a novel lightweight neural network, was developed using signal analysis and medical knowledge.
• Performance evaluation utilized the Association for the Advancement of Medical Instrumentation (AAMI) protocol and the MIT-BIH arrhythmia database.
• The model was trained to classify five distinct arrhythmia categories.

Main Results:

• KecNet achieved high classification performance with Accuracy (ACC) of 99.31%, Sensitivity (SEN) of 99.45%, and Precision (PRE) of 98.78%.
• The model demonstrated robustness in noisy environments, low memory footprint, and physical interpretability.
• KecNet proved effective in classifying five common arrhythmia types.

Conclusions:

• KecNet is a highly accurate and efficient deep learning model for arrhythmia classification on mobile devices.
• Its lightweight design and robustness make it ideal for practical implementation in wearable and mobile health technologies.
• The model's performance supports its application in real-time cardiac monitoring and diagnosis.