Using Deep Learning Networks to Improve Ultra-low Dose Tau PET/MR Image Quality

Abstract

Purpose: We investigated whether deep learning techniques can be used in enhancing ultra-low-dose 18 F-PI-2620 tau PET/MR images, acquired in a healthy aging population as well as those with neurodegenerative diseases, to produce diagnostic quality images. Methods: 44 total participants were recruited for this study and 18 F-PI-2620 tau PET/MR data were simultaneously acquired. A generative adversarial network was trained to enhance ultra- low-dose tau images, which were reconstructed from a random sampling of 1/20 (approximately 5% of original count level) of the original full-dose data; MR images were also utilized as additional input channels. Region-based analyses as well as a reader study were conducted to assess the image quality of the enhanced images compared to their full-dose counterparts. Results: The enhanced tau images showed significant noise reduction compared to the ultra-low-dose images. The regional standard uptake value ratios showed that while there is an underestimation for both image types, especially in regions with higher uptake, when focusing on the healthy but amyloid positive population, this bias was present in the unenhanced ultra- low-dose images. The radiotracer uptake patterns in the enhanced images were read accurately compared to their full-dose counterparts. Conclusion: The clinical readings of deep learning-enhanced ultra-low-dose tau PET images were consistent with those performed with full-dose imaging, suggesting the possibility to reduce dose and enable more frequent examinations in patients with dementia.