Examining lumbar spine MRI images of humans using finite element method and transfer learning approaches

Low back pain (LBP) is one of the most prevalent causes of global disability, and lumbar degenerative disc disease (LDDD) stands out as its clinical presentation. We present an automated pipeline involving U-Net with Transfer Learning (TL) and Finite Element Method (FEM) simulation to analyze the biomechanical functioning of the lumbar spine from Magnetic Resonance Imaging (MRI). We first segment the lumbar region using U-Net with TL and then generate patient-specific finite element (FE) models analyzing their biomechanics under various angular displacements (30°, 60°, and 90°) and loading conditions (482-699N). The fluid properties were tuned using Grey Wolf Optimization (GWO) algorithm. The U-Net with TL was able to segment the lumbar region with DSC value of 95.73%. We found the highest von Mises stresses for all the cases at L5/S1 level ranging from 1·0781e+008 Pa to 1·444e+008 Pa. Our results demonstrate the potential of combining machine learning and biomechanical modeling approaches for clinical evaluation of the human lumbar spine. Notably, the study makes following three key contributions: (1) it is the first work to employ U-Net with GWO for FE analysis of the lumbar spine, (2) it validates the model performance under multiple angular displacements, and (3) provides an end-to-end solution for the biomechanical analysis of the spine from MRI. We present an automated pipeline involving U-Net with Transfer Learning (TL) and Finite Element Method (FEM) simulation to analyze the biomechanical functioning of the lumbar spine from Magnetic Resonance Imaging (MRI).