Researchers at the Massachusetts Institute of Technology (MIT) and collaborating institutions have developed a patient-specific artificial intelligence technique that could improve the safety and precision of minimally invasive surgeries by helping doctors better navigate surgical instruments inside the body.
Called X-ray volume registration (xvr), the technique rapidly matches real-time X-ray images taken during surgery with a patient’s preoperative 3D medical scans, such as CT or MRI images. This can help surgeons determine the exact position and orientation of instruments during procedures performed through small incisions.
Minimally invasive procedures often rely on X-rays for navigation. However, X-rays produce two-dimensional images, making it difficult to understand the precise location of instruments relative to organs, bones and other tissues. Doctors traditionally have to manually align these images with 3D scans, a process that can be slow and demanding.
The MIT team designed xvr to adapt specifically to each patient. The system uses a patient’s existing CT or MRI scan to generate thousands of realistic, synthetic X-rays through a physics-based simulation. An AI model then uses this information to align the actual surgical X-rays with the patient’s 3D scan.
A major advantage is speed. While training a completely new model for each patient could take around 12 hours, the researchers’ pretrained foundation model can adapt to a new patient in approximately five minutes and perform the registration within seconds, with sub-millimetre precision.
Researchers tested the technology using data from five hospitals, covering various body parts, organs and both adult and paediatric patients. The system outperformed existing AI-based approaches in accuracy and robustness.