Self-Supervised Pre-Training for Deep Image Prior-Based Robust PET Image Denoising

Deep image prior (DIP) has been successfully applied to positron emission tomography (PET) restoration, enabling represent implicit using only convolutional neural network architecture without training dataset, whereas the general supervised approach requires massive low-and high-quality PET pairs. To answer increased need for imaging with DIP, it is indispensable improve performance of underlying DIP itself. Here, we propose a self-supervised pre-training model DIP-based denoising performance. Our proposed acquires transferable and generalizable visual representations from unlabeled images by restoring various degraded in approach. We evaluated method clinical brain data radioactive tracers (18F-florbetapir, 11C-Pittsburgh compound-B, 18F-fluoro-2-deoxy-D-glucose, 15O-CO2) acquired different scanners. The achieved robust state-of-the-art while retaining spatial details quantification accuracy compared other unsupervised methods model. These results highlight potential that particularly effective against rare diseases probes helps reduce scan time or radiotracer dose affecting patients.