Back pain is a significant and growing problem in Europe. In severe cases, surgery may be required to stabilise the spine. However, knowing when and how to perform this surgery remains highly subjective. The EU-funded iBack project aimed to build tools that can automatically characterise the biomechanical properties of the spine from medical scans, giving researchers a better understanding of why interventions were successful or not. Clinical radiologist Jan Kirschke acted as project coordinator. “Often, all the surgeon is doing is changing the biomechanical set-up of the spine, and hoping the pain will resolve,” he says. “This is mainly based on clinical experience, as we are not able to determine the exact biomechanical situation in an individual patient currently.”
Kirschke and his team at the Technical University of Munich (website in German) looked to several image-based biomarkers that are known to correspond to biomechanical properties of the skeleton, such as bone mineral density (BMD). This is a crucial factor in the success of spinal surgery, as it affects how well bolts and screws will adhere to the bone. “We found that many patients indeed suffer from low BMD: cases in which this would never have been suspected without dedicated measurements. To select the appropriate surgical approach in these patients, and prevent complications, we need to identify osteoporotic patients reliably, using quantitative measurements,” explains Kirschke. Through the iBack project, Kirschke and his team developed machine learning software to assess BMD using CT data from any scanner. The system is currently open for other researchers to use, for example, to investigate surgical outcomes. Kirschke has been awarded a proof of concept grant to develop the software toward a clinically validated product for pre-surgery use.
The second major contribution of the iBack project was to establish a biomechanical model of the spine. “This model integrates information from bones, muscles, ligaments, body weight, all the important anthropometric parameters of the patient, and allows us to simulate different actions such as walking or carrying heavy loads, to calculate the real forces for every element of the spine,” says Kirschke. The team is now investigating how these local-loading conditions correlate with back pain and surgical failure, in order to predict what mechanical implant is best suited for the patient. “Previously some studies annotated these scans manually, usually fewer than 10 scans, and that was the end of the line in what was possible,” adds Kirschke. “With our algorithms, we can process several thousand patients in a couple of days. That really scales up our possibilities.” The work was supported by the European Research Council. “When I started this project 7 years ago, AI wasn’t a big thing. It really came along in the meantime.” Kirschke is now examining the possibility of exploring large data sets such as the German National Cohort, which contains high-resolution MRI scans of 30 000 participants. “It would make perfect sense to apply these techniques to that kind of big-scale epidemiological data, to reveal more insights into our basic understanding of back pain,” he concludes.
iBack, spine, back, pain, surgery, scan, MRI, biomechanical, osteoporotic, stabilise, algorithms, anthropometric, loading