Evaluation of Level-Crossing ADCs for Event-Driven ECG Classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate and efficient analog-to-digital conversion (ADC) is crucial for biomedical signal processing.
• Existing ADCs face challenges in balancing sampling accuracy, data rate, and computational load for real-time applications like arrhythmia detection.
• Event-driven ADCs offer potential for reduced data rates but require optimized design methodologies.

Purpose of the Study:

• To introduce a new methodology for designing level-crossing analog-to-digital converters (LC-ADCs) to improve sampling accuracy and reduce data stream rate.
• To evaluate the performance of LC-ADC models using the MIT-BIH Arrhythmia dataset.
• To develop and assess a 1D-CNN classifier for event-driven data from LC-ADCs for cardiac arrhythmia classification.

Main Methods:

• Design and simulation of several LC-ADC models with optimized parameters.
• Evaluation of LC-ADC models based on data compression and signal-to-distortion ratio (SDR).
• Development of a 1D-CNN classifier to analyze event-driven data from LC-ADCs and compare performance against uniformly sampled data.

Main Results:

• The 1D-CNN classifier achieved high accuracy (99.49% overall, 92.4% sensitivity, 94.78% specificity) with uniformly sampled data.
• A specific 7-bit LC-ADC model demonstrated comparable performance (99.2% overall accuracy, 89.98% sensitivity, 91.64% specificity) with significantly reduced data rates (3x compression) and lower computational cost (49% FLOPS).
• The LC-ADC approach maintained a good signal-to-distortion ratio (21.19 dB).

Conclusions:

• The proposed LC-ADC design methodology effectively improves sampling accuracy and reduces data rates for biomedical signal processing.
• The 1D-CNN classifier is well-suited for analyzing event-driven data from LC-ADCs, enabling efficient cardiac arrhythmia detection.
• The study presents an open-source event-driven arrhythmia database, facilitating further research and development in this field.