AI drops tracer dose required for PET/MRI by 50%, a potential ‘major’ boost for cancer care
Doctors can use a readily available deep learning tool to perform molecular imaging exams with 50% less radiotracer material, significantly reducing patients’ radiation exposure, experts reported Wednesday.
Clinicians are always looking for ways to decrease radiation dosage, and one technique for doing so involves reducing patients’ 18F-FDG requirements. But this can also degrade image quality and, in turn, affect follow-up management decisions.
Stanford University researchers sought to overcome this problem by using a convolutional neural network to reconstruct PET/MRI scans. The results of their simulated dose study show the tool allows low-dose imaging while correctly assessing treatment response for young patients with lymphoma.
Ashok J. Theruvath, MD, with Stanford’s Molecular Imaging Program, and co-authors say other institutions can immediately use the federally approved algorithm from Subtle Medical across many tumor types to significantly alter patient care.
“Results can have major and broad healthcare impact, as the described deep learning tool to monitor tumor response with minimal radiation exposure will be readily translatable to the clinic, and ultimately will help to minimize the risk of secondary cancer development later in life,” the authors added Oct. 6 in Radiology: Artificial Intelligence.
The findings represent a secondary analysis of prospective data from 20 patients with lymphoma who underwent 18F-FDG PET/MRI scans between July 2015 and August 2019. Scans were performed at baseline and post-chemotherapy, and full-dose data were simulated to lower radiotracer doses.
Theruvath et al. found deep learning-enhanced PET scans with a 50% lower simulated dose accurately assessed patients’ response to cancer treatment compared with 100% dose scans.
The California team noted a 50% reduction is “just a first step” and future artificial intelligence must lower that dose without impeding image quality.
“The applied convolutional neural network for reconstruction of high-dose images from low-dose data are immediately clinically available and could have immediate impact on patient care by enabling us to implement low-dose clinical PET/MRI protocols for pediatric cancer patients,” the authors wrote.
Two authors disclosed holding positions within Subtle Medical. Read the entire study here.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.