Lumbar Scoliosis Analysis Using Deep Learning Based Technique

Abstract

Medical image interpretation automation saves physicians time and boosts diagnostic confidence. Most medical imaging diagnosis is done manually or semi-automatically. These methods vary when performed by several physicians. This thesis includes mid- sagittal lumbar spine magnetic resonance imaging (MRI) images with labeling and spinal metrics. Two pieces were marked. Professional radiologists created pixel-wise masks to detect vertebral bodies (VBs) in each picture, which a panel of spine surgeons subsequently examined. An enhanced method (VBSeg) compares the segmentation work of traditional and deep-learning architectural techniques. A novel computerized spinal misalignment evaluation method may help spine surgeons make objective decisions about critical surgeries. Angular deviation classifies spondylolisthesis 89% accurately, whereas the area inside the enclosed lumbar curve zone classifies LL adequacy/inadequacy 93%   accurately.