Vertebrae Landmark Detection and Scoliosis Assessment Using Deep Learning

Deep learning is transforming a number of industries including finance, healthcare, and autonomous cars. It has shown extraordinary success in many areas. We propose a novel deep-learning approach for vertebrae landmark detection and scoliosis assessment. Scoliosis, a spinal deformity which is characterized by the deformation of the spine structure resulting in ‘S’ or ‘C’ shaped curves. The U-Net is employed to determine whether the given image exhibits scoliosis or not. Additionally, the U-Net method determines the cobb angle by identifying the apex, bottom, and top vertebrae. By considering the measured angle values from the U-Net output, we can determine the scoliosis stage using OpenCV. The suggested technique uses an artificial deep convolution neural network to identify spinal landmarks in X-ray images. Trained on a large dataset of annotated images, the network learns complex patterns and relationships between different vertebrae. The detected landmarks are then used to compute various clinical parameters related to scoliosis, such as the Cobb angle and vertebral rotation. Scoliosis is classified into three stages: Mild scoliosis (<25), Moderate scoliosis (25-45), and Severe scoliosis (>45).