Impact of De-noising and Contrast Enhancement on Segmenting Small PET Features with Low Contrast

摘要

We describe the impact of an advanced PET image reconstruction method, which achieves model/prior-free denoising while enhancing image contrast, on segmenting small PET unique features (< 5 mm in diameter) with low contrast (< 1.7: 1 contrast ratio). Results from brain imaging simulations showed that the model/prior-free de-noised reconstruction improved the segmentation accuracy by 3–4 times as compared to standard reconstruction with post filter. A further 3–4 times improvement in segmentation accuracy was achieved by contrast enhancement with model/prior-free de-noising as compared to model/prior-free de-noising alone. Contrast enhancement with de-noising was also observed to capture small PET features with low contrast more reproducibly than de-noising alone. Similar trends were observed for data obtained from human studies as well.