Aims:
Adolescent Idiopathic Scoliosis (AIS) is a three-dimensional spinal deformity that affects children’s health. Traditional screening and diagnosis require patients to undergo X-ray examinations, which is detrimental to adolescents' health. This study aims to propose a novel solution using deep learning and depth sensing techniques to generate the 3D spine model for
Methods:
From October 9, 2019, to May 21, 2022, a total of 2238 AIS patient data were collected at Queen Mary Hospital and Duchess of Kent Children's Hospital at Sandy Bay in Hong Kong. Among these, data from 1936 patients were used for training and validating the deep learning model, and data from 302 patients were used for prospective independent testing. The collected patient data included demographic data, colour and depth (RGBD) images of patients' nude backs captured using a depth camera and the whole spine X-ray images. Using the developed deep learning algorithm, a precise 3D spine model can be generated for the patients. The severity of the condition was assessed based on the generated spine model and compared with the gold standard obtained from X-ray results to analyze the feasibility and clinical significance of the proposed method.
Results:
The generated spine models were used to evaluate the severity of the condition, with prediction accuracy rates of 83.5% for 85 normal or mild patients, 93.5% for 184 moderate patients, and 90.9% for 33 severe patients. Visualizations of the generated spine model revealed a high prediction accuracy, fitting the geometric morphology of the patients' unclothed backs.
Conclusions:
This study explores the technical feasibility of accurately generating patients' spine models using their back geometric morphology data combined with deep learning algorithms. The obtained spine model can aid doctors in diagnosing the severity of the condition. This finding provides new research directions and practical support for non-radiation AIS assessment methods.