Enhancing Low-Dose Pet Imaging: A Novel Contrastive Learning Method for Perceptual Loss and an Organ-Aware Loss

Xinrui Zhan, Aimei Li, Jian He, Lei Xiang

Abstract: Conventional pixel-wise loss functions used in medical image enhancement and denoising often result in over-smoothing. While perceptual loss has shown effectiveness in natural image tasks, its performance diminishes in medical imaging due to the incompatibility of nature image-trained backbone models and the challenge of assembling a medical dataset comparable to ImageNet. Addressing this, our paper introduces a novel self-supervised-learning approach, Medical Volume framework for Contrastive Learning of visual Representation (MedCLR), to train a backbone model better suited for perceptual loss in medical contexts. We apply this loss to enhance low-dose Positron Emission Tomography (LPET) images. Additionally, we propose an Organ Loss that focuses on enhancing Standard Uptake Value (SUV) quantification accuracy in key anatomical regions. The integration of these loss functions significantly improves the realism and diagnostic value of LPET images, marking a significant advancement in medical imaging AI and its potential extension to various imaging modalities.