Scoliosis Detection using the XBox Kinect

The current clinical standard for scoliosis detection is manual assessment of the Cobb Angle from radiographic images. This method is time-consuming, unaffordable, invasive (subjects the patient to high levels of radiation when done frequently), and not viable for mass screenings. Manually solving for the Cobb Angle is also prone to intra- and inter-observer errors. Using the depth sensing applications of the Microsoft Xbox Kinect and the data processing capabilities of machine learning algorithms, the research aims to develop a scoliosis detection program that is fast, inexpensive, non-invasive, and user-friendly. The detection program made use of four alternative scoliosis assessment algorithms called 'trunk surface metrics' developed in earlier studies, namely the 1) Shoulder Angle Alignment, 2) Body Alignment, 3) Posterior Trunk Symmetry Index, and 4) Deformity in the Axial Plane Index. A decision tree, based on the C4.5 algorithm, for data analysis was developed. A preliminary testing (n = 26) was implemented to calibrate the program and train the decision tree. The final testing on the program (n = 60) then produced a 52.38% mean sensitivity and a 61.11% mean specificity. In conclusion, the results of the study show the potential of using depth-sensing technologies and machine learning algorithms for scoliosis screening and other biomechanical applications.