AI Model Enhances Precision in Minimally Invasive Surgeries

How X-Ray Volume Registration Works in Surgery

Clinicians perform minimally invasive procedures such as angioplasty by inserting instruments through tiny incisions while using mobile X-ray scanners to visualize the anatomy from various angles. According to Vivek Gopalakrishnan, a postdoc in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and lead author of the study published in Nature, standard X-rays provide flat 2D images that make it difficult to determine exact tool orientation without decades of specialized training.

To solve this localization challenge, clinicians traditionally perform manual registration by guessing tool positions or punching coordinates into a computer screen. While prior artificial intelligence tools attempted to automate this alignment, they often failed to generalize across diverse patient anatomies. The xvr framework addresses this limitation by tailoring a patient-specific model using the individual’s own preoperative MRI or CT scan.

Patient-Specific AI Training and Physics-Based Simulation

Instead of relying on generic generative models, xvr uses physics-based simulation to generate approximately 1,000 synthetic X-rays per second from a patient’s preoperative CT or MRI scan. According to Gopalakrishnan, this purely physics-based approach leaves no room for hallucinations because the data stems directly from the patient’s actual medical imaging.

Training a model from scratch for each patient typically takes about 12 hours, which is impractical for emergency settings. To overcome this hurdle, the research team used xvr to pretrain a versatile foundation model. By leveraging whole-body 3D scans from more than 2,000 patients covering diverse ages and modalities, the system acquired the ability to adapt to a new patient in roughly five minutes while maintaining high accuracy.

Hospital Testing and Future Medical Applications

The research team tested xvr on the largest available dataset of real 2D/3D registrations, incorporating clinical data from five hospitals that spanned dozens of bones and organ systems in both adult and pediatric patients. Co-authors on the study include Polina Golland and Neel Dey as co-senior authors, alongside clinicians and researchers from Harvard Medical School, Massachusetts General Hospital, St. Luke’s Marion Bloch Neuroscience Institute, Shriners Children’s Hospital, Brigham and Women’s Hospital, and Boston Children’s Hospital.

According to the study findings, xvr outperformed existing artificial intelligence registration methods by an order of magnitude. The development team is now collaborating with surgical robotics companies and clinical groups to integrate the algorithm into navigation tools for surgeries and expand its capabilities to handle complex, moving body parts.

Frequently Asked Questions

What does xvr stand for?

Xvr stands for X-ray volume registration, a technique that matches 2D surgical X-rays with 3D preoperative patient scans.

How long does it take the xvr model to adapt to a patient?

According to MIT researchers, the pretrained foundation model adapts to a new patient in about five minutes and performs the image registration in a matter of seconds.

What medical procedures can benefit from this technology?

The system is designed to assist clinicians during minimally invasive procedures such as emergency stroke interventions and angioplasties by improving the guidance of catheters, endoscopes, and surgical tools.