The ability to accurately compare medical images is essential for tracking disease, planning care, and studying how the body changes over time. But images taken at different times—or from different people—often don’t line up well.

Iman Aganj, PhD

A simple alignment technique called affine registration can fix overall position and orientation (for example, by applying transformations like rotation or scaling so that everything in the image shifts in a uniform way), but often misses local anatomical shape changes. Deformable registration can handle those details, but may converge slowly or produce inaccurate results when the initial mismatch is large.

To bridge this gap, a new study by Iman Aganj PhD, of the Athinoula A. Martinos Center for Biomedical Imaging at Mass General Brigham, explores a practical middle step. 

Aganj introduces Intermediate Deformable Image Registration (IDIR), a method that gently reshapes one image to better match another before fine-tuning begins.

Unlike affine registration, IDIR allows different parts of the image to move independently—capturing large, smooth anatomical deformations—while providing a strong starting point for standard deformable registration.  

Importantly, IDIR works for many types of scans and does not rely on AI training. When tested on X-rays, brain MRIs, and abdominal CT scans, the approach improved alignment within only a few iterations. As such, it holds promise for improving the accuracy and reliability of medical image analysis, with benefits for both research studies and clinical workflows. 

Published in Scientific Reports on March 13, 2026 | Read the paper: “Efficient cosine-windowed cross-correlation for intermediate deformable image registration” 

Summary reviewed by: Iman Aganj, PhD