Physics-Inspired Diffusion Probabilistic Models for Improved Denoising in Intracardiac Time Series.

Area of Science:

• Biomedical Engineering
• Computational Electrophysiology
• Artificial Intelligence in Medicine

Background:

• Intracardiac electrophysiological (EP) signals are vital for diagnosing arrhythmias but are often corrupted by noise.
• Existing noise reduction techniques may compromise essential signal morphology, hindering accurate diagnosis.

Purpose of the Study:

• To develop and evaluate a physics-inspired conditional denoising diffusion probabilistic model (cDDPM) for reconstructing intracardiac EP signals.
• To compare the cDDPM's performance against classical filters and variational autoencoders in preserving signal fidelity.

Main Methods:

• Trained a cDDPM using 5706 monophasic action potentials from 42 patients with simulated and real EP noise.
• Employed an iterative training process inspired by Brownian motion.
• Evaluated performance using Root Mean Square Error (RMSE), Pearson Correlation Coefficient (PCC), and Peak Signal-to-Noise Ratio (PSNR) metrics.

Main Results:

• The cDDPM demonstrated superior performance across RMSE, PCC, and PSNR metrics compared to traditional methods.
• The model proved robust against complex noise while preserving critical signal morphological features.
• Achieved significant enhancement in signal fidelity and noise reduction.

Conclusions:

• Physics-inspired diffusion models offer a powerful approach for denoising intracardiac EP signals.
• The proposed cDDPM can significantly improve the accuracy of arrhythmia diagnosis and guide clinical interventions.
• This technology holds potential to enhance the clinical utility of EP signal analysis.