2D Photogrammetry Image of Adolescent Idiopathic Scoliosis Screening Using Deep Learning

Adolescent idiopathic scoliosis (AIS) is a high incidence disease in adolescents, with a long treatment time and difficult to cure. As a consensus, the preliminary AIS screening is of crucial importance to detect the disease at an early stage and allows proactive interventions to prevent the disease from becoming worse and reduce future treatment. Currently, the conventional palpation or Adam forward leaning is the most widely used preliminary screening method considering the Axial Trunk Rotation (ATR) value calculated by scoliosis assessment equipment. However, this method relies heavily on the subject’s standing posture and the doctor’s experience. In this paper, we develop an efficient deep learning-based framework to enable a large-scale scoliosis screening by using only one unclothed two-dimensional (2D) human back image, without any X-radiation equipment. We classify the normal and abnormal scoliosis using ATR value as classification label which calculated from the human back three-dimensional (3D) point cloud. Our accuracy in the task of AIS classification reaches $$81.3\%$$ , far exceeding the accuracy of visual observation by experienced doctor ( $$65\%$$ ), which can be used as a remote preliminary scoliosis screening method.