Cardiovascular Disease Classification System With ECG-Gating PCG Algorithm and Programmable AI Accelerator Design.

Area of Science:

• Biomedical Engineering
• Cardiology
• Computer Engineering

Background:

• Cardiovascular diseases (CVDs) are a leading cause of mortality, necessitating improved diagnostic tools.
• Current methods for diagnosing cardiac conditions often require clinical settings, limiting timely detection.
• Wearable devices with edge-computing offer potential for real-time, accessible cardiovascular health monitoring.

Purpose of the Study:

• To develop a wearable system for accurate, real-time diagnosis of cardiovascular diseases, specifically arrhythmia and heart valve diseases (HVDs).
• To address the challenges of implementing complex classification models on resource-constrained wearable devices.
• To create an efficient hardware accelerator for multiple diagnostic models.

Main Methods:

• An ECG-gating algorithm was developed to improve phonocardiogram (PCG) signal analysis.
• Advanced classification algorithms were implemented for arrhythmia and HVD detection.
• A systolic array-based accelerator with an application-specific instruction-set processor (ASIP) was designed and fabricated.

Main Results:

• The algorithms achieved high accuracy: 97.8% for arrhythmia and 99.3% for HVD, with minimal hardware quantization error (<0.5%).
• The fabricated accelerator demonstrated low power consumption (414 μW at 1 MHz) and fast classification times (7.2 ms for arrhythmia, 21 ms for HVD).
• Exceptional energy efficiency was achieved (395.3 GOPS/W normalized to 40 nm).

Conclusions:

• The developed system effectively classifies arrhythmia and heart valve diseases using wearable technology.
• The combination of advanced algorithms and a specialized hardware accelerator enables efficient on-device cardiac diagnosis.
• This technology holds significant promise for improving early detection and management of cardiovascular conditions.