Optimizing AR PAM image enhancement: learning & model based approaches with GANs & deep CNNs

Abstract

Photoacoustic Imaging (PAI), an emerging biomedical imaging technology, holds significant promise for medical diagnosis and biological research. This study addresses the challenge of improving image quality in acoustic resolution photoacoustic imaging.Introducing acoustic resolution (AR) and optical resolution ( OR images to train a deep learning network architecture MultiResU Net which is a Fully Connected U shaped Convolutional Network (U Net) that incorporates multiple residual blocks ) enhances the quality of AR PAM images. Subsequently, the Adversarial One Class Deep Transfer Learning Generative Adversarial Network AODTL GAN ) architecture is introduced to overcome domain shift issues, effectively improving perceptual image quality. Quantitative evaluation demonstrates the proposed algorithm's effectiveness, with peak signal to noise ratio (PSNR) increasing from 14.33 dB to 18.47 dB and the structural similarity index (SSIM) increasing from 0.1996 to 0.2975. Furthermore, a novel algorithm combining learning based and model based approaches is explored. Using the generated FFDNet structure as a plug and play (PnP) prior, different levels of additive white Gaussian noise (AWGN) are adaptively eliminated. In vivo experimental results show this method significantly improves image resolution while maintaining enhancement flexibility , opening new possibilities for developing photoacoustic imaging technology.