In today’s healthcare system, advances in technology have significantly improved treatment and diagnosis approaches, especially in the area of medical imaging. A lumbar spine magnetic resonance imaging (MRI) is necessary for the diagnosis of a number of spinal disorders. The lumbar spine, a essential component of the spinal system, comprises five vertebrae, L1 through L5, which provide stability and support the upper body’s weight. These vertebrae have cylindrical bodies, vertebral arch, and muscle attachment processes. Accurate segmentation of these images is essential for effective diagnosis and treatment planning. This study presents a comprehensive analysis of Gaussian Mixture Models (GMM) and Kernel Gaussian Mixture Models (KGMM) for enhancing lumbar spine MRI segmentation. Assuming that each data point originates from a mixture of a finite number of Gaussian distributions with unknown parameters, the GMM is a probabilistic model. By leveraging this approach, GMM effectively handles the variability and complexity of MRI data. However, traditional GMM can be limited by its reliance on linear separability. To address this, the KGMM algorithm is introduced, which extends GMM by incorporating kernel methods to capture nonlinear relationships in the data. This study evaluates the performance of both GMM and KGMM algorithms through extensive experiments on a large dataset of lumbar spine MRI scans. Results demonstrate that KGMM significantly improves segmentation accuracy compared to standard GMM, particularly in regions with complex anatomical structures. The analysis highlights the strengths and limitations of each method, providing valuable insights for selecting appropriate segmentation techniques in clinical practice. The findings underscore the potential of advanced probabilistic models in enhancing the precision of lumbar spine MRI segmentation, ultimately contributing to better patient outcomes.