AI chatbot model generates complete reports on ACL tears

Thursday, November 30 | 9:10 a.m.-9:20 a.m. | R3-SSIN07-2 | Room N227B

In this session, an AI chatbot called MedVisGPT will be introduced that can help diagnose anterior cruciate ligament (ACL) tears on knee x-rays and produce detailed medical reports in seconds based on conversations with users.

Presenter Xingxin He, PhD, a postdoc research fellow at Massachusetts General Hospital in Boston, will discuss MedVisGPT (Medical Visual Generative Pre-Trained Transformer), a model based on data from 10,000 individuals with confirmed isolated ACL tears and no prior ACL reconstruction. All x-rays included anteroposterior and lateral views and were collected from patients with an average age of 14.

To demonstrate MedVisGPT’s capabilities, the group gave the chatbot knee x-rays of different views from two patients (a control case and a case with ACL) and asked it to describe the x-rays briefly,  provide its impression of the x-rays, provide the rationale behind the task, and provide a report in a formal format. Additionally, for the ACL case, they asked questions that required suggestions: namely, “Is it an ACL?” and “So does it need further MRI to confirm?”

MedVisGPT properly answered each question regarding the given images, gave an MRI recommendation for the suspicious case, and summarized the context during chatting to generate a complete report in a professional format, according to the results.

“The entire process takes less than one minute, which tremendously reduces the workload of human radiologists in diagnosing medical images,” He noted.

Check out the session to learn more about MedVisGPT.