LSU Shreveport students shine with AI model for spine analysis

SHREVEPORT, La. (KTAL/KMSS) — The Artificial Intelligence and Machine Learning Lab at LSU Shreveport (LSUS) won a first-place award at the national AIM-AHEAD annual meeting for its research poster on automatic spine segmentation earlier this month.

“SpineSegGAN: Automatic Segmentation of Lumbar Spine” explained how their algorithms and model accurately segmented 93 percent of MRIs, an improvement over other models. Competing models range from 75 to 80 percent.

A team of student co-authors from the Master of Science in Computer Systems Technology program, including Devesh Sarda, Udaysinh Rathod, and Mridula Mavuri, was guided by lab director Dr. Subhajit Chakrabarty.

“This award is significant because of the national scale of this conference. These are master’s students competing with and beating doctoral and postdoctoral researchers along with medical doctors and other faculty members who work in established medical research labs full-time,” said Dr. Chakrabarty of the Dallas-based conference that draws more than 7,000 artificial intelligence researchers and experts.

AIM-AHEAD is a grant program established to help the development of AI/ML models to perform medical research, starting with electronic health record data.

The LSUS team won in the hub-specific project category, one of 16 research categories that featured 220 total research posters.

Rathod said, “Everyone is coming up to our posters asking if we’re doing our Ph.D.’s on this – but we’re only master’s students. We really have to thank Dr. Chakrabarty because of his direction and guidance.”

Spine segmentation is the process of using medical images, such as MRI and CT scans, and precisely outlining anatomical structures, specifically vertebrae, intervertebral discs, and the spinal canal. The task is performed by radiologists and assists the patient’s medical team in diagnosing conditions, identifying abnormalities, and planning surgeries.

Sarda explained that the model identifies and highlights spinal structures. This allows a doctor to see the presence of spinal constriction, disc movement, or hernias. Because the AI does that part of the work, the doctor only has to confirm what the AI concluded rather than do a full diagnosis.

“Doctors can validate the diagnosis, or if they notice something that is out of place or they don’t concur with the model’s diagnosis, they can take a deeper look at that particular case,” Sarda said.

The first-place poster is the fourth for the lab. The first two came at the Louisiana Biomedical Research Network, the third came at a previous AIM-AHEAD conference.