AI Accelerator With Ultralightweight Time-Period CNN-Based Model for Arrhythmia Classification.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Arrhythmia diagnosis relies heavily on electrocardiography (ECG) interpretation.
• Efficient and accurate arrhythmia classification is crucial for timely patient management.
• Existing methods may lack the speed or accuracy required for real-time clinical application.

Purpose of the Study:

• To develop an efficient and accurate AI-based classification system for arrhythmias.
• To create an ultralightweight convolutional neural network model incorporating R-peak interval features.
• To design a high-performance, low-power AI accelerator for hardware implementation of the classification algorithm.

Main Methods:

• A naive preprocessing procedure for ECG data was developed.
• An ultralightweight convolutional neural network model was designed using R-peak interval features.
• The model was trained and tested on the MIT-BIH and NCKU-CBIC databases adhering to AAMI standards.
• A customized integrated circuit with a parallelized processing element array architecture and hybrid stationary techniques was designed for AI acceleration.
• The accelerator was implemented using the TSMC 180 nm CMOS process.

Main Results:

• The AI model achieved high classification accuracies of 98.32% and 97.1% on the MIT-BIH and NCKU-CBIC databases, respectively.
• The web-based system provides instant ECG wave condition viewing for cardiologists and patients.
• The AI accelerator demonstrated low power consumption (122 µW), low classification latency (6.8 ms), and high energy efficiency (0.83 µJ/classification).

Conclusions:

• The proposed AI system significantly enhances arrhythmia classification accuracy and diagnostic efficiency.
• The ultralightweight model and custom AI accelerator offer a practical solution for real-time, low-power medical applications.
• This integrated approach improves the quality of medical examinations and patient care through instant AI-driven insights.