Scoliosis Detection using Deep Neural Network
It tackles the problem of automating scoliosis detection for medical diagnosis, but appears incremental as it reviews and implements existing models.
This thesis reviews deep learning methods for detecting scoliosis from spinal X-ray images, aiming to improve accuracy and automate Cobb angle prediction to address the time-consuming and variable manual diagnosis process.
Scoliosis is a sideways curvature of the spine that most often is diagnosed among young teenagers. It dramatically affects the quality of life, which can cause complications from heart and lung injuries in severe cases. The current gold standard to detect and estimate scoliosis is to manually examine the spinal anterior-posterior X-ray images. This process is time-consuming, observer-dependent, and has high inter-rater variability. Consequently, there has been increasing interest in automatic scoliosis estimation from spinal X-ray images, and the development of deep learning has shown amazing achievements in automatic spinal curvature estimation. The main target of this thesis is to review the fundamental concepts of deep learning, analyze how deep learning is applied to detect spinal curvature, explore the practical deep learning-based models that have been employed. It aims to improve the accuracy of scoliosis detection and implement the most successful one for automated Cobb angle prediction. Keywords: Scoliosis Detection, Spinal Curvature Estimation, Deep Learning. i