Traditional school scoliosis screening approaches remains debatable due to unnecessary referal and excessive cost. Deep learning algorithms have proven to be powerful tools for the detection of multiple diseases; however, the application of such methods in scoliosis screening requires further assessment and validation. Here, the investigators develop an artificial system for the automated screening of scoliosis using disrobed back images, and conduct clinical trial to validate if the diagnostic system can offsetting the shortcomings of human doctors.

Inclusion Criteria:

• 1.Patients included both pretreatment back photos and whole spine (C7-S1) standing X-ray or ultrasound images (for healthy population); 2. All the documents are clear to be recognized by naked eyes; 3. Back photos and are taken at the same time (not >1month); 4.Patients were consider as idiopathic scoliosis according to clinical photos.

Exclusion Criteria:

• 1. Patients were considered as non-idiopathic scoliosis for obvious abnormal features of trunck,such as Cafe-au-Lait spots for neurofibromatosis, Spider finger, Abnormal hair spot of back, pelvic tilt, lower limb discrepancy and so on; 2.The taken time between back photo and X-ray or ultrasound was more than 1month; 3.The clinical photos and images were not clear; 4. The X-ray film or ultrasound images not including whole spine (C7-S1).