Deep Learning Approach via Deep Neural Network to Predict the Curve Progression of Adolescent Idiopathic Scoliosis: A Pilot Study

Abstract

We studied the prognosis of 23 scoliosis adolescents based on gait analysis via machine learning and deep learning approach of prediction models. Our study showed the applicability of the deep learning approach for prognosis prediction. Gait cycles have rich information related to joint movements. However, due to the high complexity of gait cycles, using their information with handcrafted features showed limited applications. Our proposed model was designed to use multi-plane hip and knee gait cycles as input for prognosis prediction and it showed better performance than approaches with handcrafted features from gait cycles. We also visualized an explanation of our model with Gradient-weighted class activation map to interpret our model’s outcome. Mid-stance to pre-swing phase in gait needed to be more precisely evaluated for the prediction of scoliosis prognosis.