Pinched nerve or misalignment: AI finds the cause of lower back pain AI can assist radiologists in detecting abnormalities in the spine

A trained computer can detect various causes of lower back pain on an MRI scan, even more effectively than a radiologist. This technology could help in the future when deciding whether surgery for back pain is necessary. These results are described in the dissertation of Jasper van der Graaf from Radboud university medical center.

Nearly half of the Dutch population will experience lower back pain at some point in their lives. This can have various causes: problems with the intervertebral discs, nerves, bones, joints, or muscles. A doctor uses an MRI scan to map the spine, look for abnormalities, and then determine whether surgery is advisable. PhD candidate Jasper van der Graaf wondered if AI, a tireless pattern recognizer, could make this assessment more accurately.

To test this, Van der Graaf first trained a computer to recognize the structures in the spine from MRI scans. 'We input nearly five hundred scans from four hospitals, where all the structures were manually labeled. This taught the AI what the spine looks like, and then the algorithm could color in the parts itself and detect any abnormalities.'

Like a bicycle tire

Next, he searched the literature for abnormalities in the spine that have been shown to cause back pain. He focused on the three main culprits for back pain. In scoliosis, there is a curve in the back. In central canal stenosis, narrowing in the nerve canal causes a pinched nerve. And intervertebral discs can also cause pain, either through collapse or by bulging like a bicycle tire and pressing on nerves.

Van der Graaf developed an AI system that accurately and reliably measures these three causes of back pain. For scoliosis and stenosis, he compared the AI with at least three radiologists and showed that the AI performed even better. 'In scoliosis, a radiologist measures the curvature of the spine, looking for the angle with the largest deviation', he explains. 'Our research showed that in only nine of fifty patients did the radiologists all select the same angle. AI does this much better, as it calculates all possible angles and automatically selects the largest deviation.'

Usefulness of surgery  

AI is very good at detecting abnormalities in the spine and making various measurements. But does this help in better determining whether surgery is needed? Van der Graaf: 'We haven’t investigated that yet. The Sint Maartenskliniek has a large database of patients, with known characteristics such as age, weight, and type of complaint, but also whether surgery was helpful. This database supports the decision-making about surgery, as the doctor can compare it to similar past patients. It would be great to integrate our AI tools into that.'

'This research was the first step in that direction', says Professor of Medical Image Analysis Bram van Ginneken. 'This AI system automatically extracts relevant information from the MRI and presents it in a clear dashboard to the orthopedic surgeon. We can now begin integrating this information into the decision model of the Sint Maartenskliniek, which assesses whether surgery is appropriate.'

Two examples of an MRI scan of the spine, with the original image on the left and the same image on the right where AI has colored the different parts to detect abnormalities.

About the thesis defense

PhD defense on February 7, 2025, at 12:30 PM by Jasper van der Graaf. Dissertation title: AI-driven MRI analysis for low back pain management (available online after February 7). Supervisors: Prof. Dr. M. de Kleuver and Prof. Dr. B. van Ginneken. Co-supervisors: Dr. M.L. van Hooff and Dr. M.J.C.M. Rutten. The defense can be followed via this livestream.

PhD candidate Jasper van der Graaf

• Want to know more about these subjects? Click on the buttons below for more news.

  News homepage news & mediaArtificial intelligence