Enhancing signal-to-noise ratio in LED-based photoacoustic imaging using Conditional Denoising Diffusion Probabilistic Model

Photoacoustic imaging (PAI) is an emerging medical imaging technique with applications in blood oxygen imaging and tumor detection. LED-based PAI offers a cost-effective and accessible alternative but faces challenges in high-frame-rate applications due to significant noise, necessitating extensive signal averaging. In this work, we investigate the use of deep learning techniques, specifically a conditional Denoising Diffusion Probabilistic Model (cDDPM), for denoising photoacoustic images obtained from an LED-based system. Our study evaluates the effectiveness of cDDPM in reconstructing image features and explores optimization through scheduler modifications. We implement a cosine scheduler to reduce redundant denoising steps, significantly improving inference efficiency while maintaining high image quality. These results demonstrate the potential of diffusion models for enhancing low-frame-averaged photoacoustic images.