Denoising method for image quality improvement in photoacoustic microscopy using deep learning

Abstract

Photoacoustic microscopy (PAM) is an imaging technology developed rapidly in recent years. The technology has the advantages of high resolution, rich contrast of optical imaging and high penetration depth of acoustic imaging. It is widely used in biomedical field, such as tumor detection. Photoacoustic images can not only reflect the structural characteristics of tissues, but also reflect the metabolic state, disease characteristics and even nerve activity of tissues, so as to realize functional imaging. Photoacoustic (PA) signals are inherently recorded in noisy environments and are also exposed to the noise of system components. The presence of noise has a great negative impact on image quality and interferes with image details. Therefore, it is necessary to reduce the noise in PA signals to reconstruct images with less interference information. Because deep learning can process image information quickly and efficiently, deep learning has become the preferred method for photoacoustic image denoising in recent years. In this study, the photoacoustic blood vessel image obtained was added with a certain intensity of Gaussian noise, and the denoising generative adversarial network based on Wasserstein distance (WGAN) was used to denoise the photoacoustic blood image. For the purpose of evaluation, the Peak Signal-to-Noise Ratio (RSNR), Structural Similarity Index Metric (SSIM), Universal Quality Index (UQI) and Image Enhancement Factor (IEF) were calculated. According to the calculation results, this study effectively improves the image quality, proves the effectiveness of the neural network, and has good clinical significance and broad application prospects.