MIT's xvr Matches Live X-rays to 3D Scans. Surgery Gets Simpler. The Registration Problem Is Older Than You Think.
Researchers at MIT CSAIL developed a technique called xvr that matches 2D X-rays captured during surgery with a patient's preoperative 3D scan. The work, led by postdoc Vivek Gopalakrishnan with advisor Polina Golland, appears in Nature. The stated goal is making minimally invasive procedures more accessible to broader populations.
This illustrates the registration problem in computer vision. You have two representations of the same object, one captured before surgery in 3D and one captured during surgery in 2D, and you must align them in real time. The principle generalizes beyond medicine. Any time you fuse data from different sensors or viewpoints, you are solving this same mathematical alignment problem.
MIT CSAIL's Vivek Gopalakrishnan, a graduate of the Harvard-MIT Program in Health Sciences and Technology, led the work. His advisor Polina Golland, the Sunlin and Priscilla Chou Professor of EECS, co-authored the paper.
- Open a free account on Physiopedia or similar anatomy education site and find a labeled 3D skeletal model. Study how bones connect from multiple angles. This is your preoperative scan equivalent.
- Search for a flat 2D X-ray image of the same anatomical region on a public medical image database like Radiopaedia. Try to mentally identify which 3D structures correspond to the flat image. You are now doing registration by hand.
- Compare your mental alignment to the labeled 3D model. Notice how ambiguous 2D becomes without spatial context. That ambiguity is exactly what xvr's AI model resolves automatically.