Electrocardiogram signal denoising and heart disease classification

Abstract

Electrocardiogram (ECG) recordings are often contaminated by baseline wander (BLW), power-line interference, and motion or muscular noise, reducing the reliability of both manual and automated diagnosis. The paper presents a light and reproducible MATLAB pipeline applying finite-impulse-response (FIR) filters designed using Kaiser and Hamming windows for ECG denoising, which after R-peak detection follows an RR-interval analysis for classification of heart rate as tachycardia, bradycardia, or normal. In the experiments, 15 MIT-BIH records with added Gaussian noise at several SNR levels were used for benchmarking the performance of denoising. FIR band-pass and low-pass windowed filters improved the clarity of the waveform and supported robust R-peak detection; RR-interval-based classification reached a mean accuracy of ~98.7% on the study set. The approach is computationally lightweight and quite suitable for embedded real-time deployment but is restricted to the small sample of records and synthetic noise modeling. Future work will compare the efficacy of windowed FIR filtering against modern deep-learning denoisers (CNN/RNN/GAN architectures) and assess the pipeline in larger clinical datasets.