ECG Denoising: From Wavelet Transform to Artificial Intelligence-A Comparative Study

Electrocardiogram (ECG) signals are essential for diagnosing heart conditions but a range of noises often corrupt ECG recordings, thus accurate interpretation becomes difficult. Among these noises are baseline drift, electrical interference, muscle contraction, and electrode movement. This work evaluates five reliable conventional denoising techniques—wavelet transform, adaptive filtering, and empirical mode decomposition (EMD)—against 15 advanced deep learning techniques including CNNs, autoencoders, GANs, and transformers. Using metrics such as signal-to-noise ratio (SNR), root mean squared error (RMSE), and percentage root difference (PRD), MIT-BIH ECG records (100–103) were assessed and evaluated. Additionally compared in a controlled experiment using synthetic ECG data are CNN's and EMD's noise reduction features. Visual comparisons show signal clarity and preservation of shape. The study highlights the benefits of deep learning models in ECG denoising as well as their possibilities in real-time healthcare applications. The work provides important insights for researchers and engineers developing next-generation biomedical signal processing systems.