Statistical Modeling of Spine MRI: GMM and AGMM-Based Segmentation for Lumbar Angle Classification

The precise provision of medical services with the goal of maintaining and improving health via prevention, diagnosis, treatment, and rehabilitation is referred to as healthcare. Spinal problems greatly impair mobility and quality of life, thus diagnosing and treating them is essential. The lumbar spine is composed of five vertebrae. (L1 to L5) are essential for maintaining body weight and safeguarding the spinal cord and nerves. Accurate diagnostic techniques are crucial for the prevention and treatment of diseases including spinal stenosis and degenerative disc degeneration. In order to diagnose spinal problems, medical imaging, MRI in particular, is frequently utilized. Segmenting the lumbar spine from MRI scans is a crucial part of this process, which entails locating important anatomical features such the vertebrae, inter vertebral discs, and spinal canal. The constraints of manual operations have made automated segmentation methods-especially machine learning techniques-essential because they offer higher levels of efficiency and precision. This study focuses on AGMM (Adaptive Gaussian Mixture Model) and GMM (Gaussian Mixture Model) statistical modeling techniques for lumbar spine segmentation. These models offer flexible approaches for handling the unpredictability of MRI data. In order to evaluate the Lumbar Lordotic Angle (LLA) or Lumbosacral Angle (LSA), which is essential for identifying aberrant curvatures such scoliosis and lordosis, the study intends to categorize segmented lumbar spine areas. Regarding robustness, computing efficiency, and segmentation accuracy against noise and artifacts, the efficacy of GMM and AGMM are evaluated. The results could improve the clinical use of MRI in spinal healthcare by offering new perspectives on how to best use machine learning algorithms for better imaging results.