Multi-scanner domain shift has been a persistent barrier to quantitative and qualitative CT imaging and robust medical AI. I worked on this problem during my PhD because I found images varied across scanners and models shouldn’t fail simply because a patient was scanned on a different detector, scanner, or even image acquisition protocol. By combining physics-based simulation with deep learning harmonization, we can make spectral CT consistent across scanners and conditions. Ultimately, this enables reliable imaging and the development of medical AI that can be trusted in clinical practice.
Derived from my LinkedIn Post.
Mridul Bhattarai, PhD
Building physics-informed AI frameworks for reliable clinical imaging.
