Updated · USC Viterbi School of Engineering · Aug 11
USC Scientists Build Qu-Net, Beating Classical Cancer Imaging AI by 7%
Updated
Updated · USC Viterbi School of Engineering · Aug 11
USC Scientists Build Qu-Net, Beating Classical Cancer Imaging AI by 7%
1 articles · Updated · USC Viterbi School of Engineering · Aug 11
Summary
Qu-Net, a hybrid quantum-classical AI system from USC researchers, improved medical image segmentation by about 7% over a leading classical model in tests tied to cancer imaging.
250,000 parameters powered Qu-Net versus 1.5 million for standard U-Net, suggesting quantum feature extraction could deliver sharper tumor boundaries with lower computing demands and shorter training times.
Amir Kalev and Naman Jain built the system by adding their QuFeX quantum module to U-Net, aiming to help AI work better on the small, low-quality datasets common in medicine.
Keck School of Medicine physicians are now working with the team to test the approach on simulated scans and real patient images for radiation planning, where faster, more precise tumor outlining could cut treatment-plan preparation from days to a single visit.
The researchers say the early-stage work could extend beyond cancer to brain, cardiovascular and surgical imaging, and eventually to nonmedical computer vision such as autonomous vehicles and satellite analysis.
Why is a new quantum-enhanced AI outperforming massive classical models in outlining cancer tumors while requiring drastically less computing power?
Can integrating quantum computing into medical imaging safely shrink radiation planning from several days down to a single appointment without risking accuracy?