A deep-learning model called a convolutional encoder-decoder residual network, developed by Stanford University researchers, was used to process ultralow-dose PET image data and produced synthetic images similar in quality to full-dose images at 99% reduced radiotracer dose for PET and at 99.5% reduced tracer dose for FDG-PET/MRI. The approach, presented at a radiology society meeting, also yielded more than 50% lower root-mean-square error and 7 dB higher peak signal-to-noise ratio than the original ultralow-dose images.
Study evaluates use of deep-learning model to reduce PET radiotracer dose
Sign up for SNMMI SmartBrief
Nuclear medicine and molecular imaging news
Get the intelligence you need: news and information that is changing your industry today, hand-curated by our professional editors from thousands of sources and delivered straight to your inbox.