Application of machine learning standardized integral area algorithm in measuring the scoliosis

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Scoliosis, a spinal deformity, requires accurate measurement for treatment planning.
• Traditional methods for measuring scoliosis severity, like the Cobb angle, can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate a computer vision-based method for automated scoliosis degree measurement.
• To assess the efficacy of a machine learning algorithm in quantifying scoliosis severity using X-ray images.

Main Methods:

• A computer vision algorithm utilizing histogram equalization and a flipping method was employed to identify spinal curves.
• The degree of scoliosis was quantified using a standardized integral area calculated from fitted cubic splines.
• The algorithm was tested on 204 idiopathic scoliosis patient X-rays, with results compared to manual Cobb angle measurements.

Main Results:

• The standardized integral area algorithm demonstrated a strong positive correlation (Spearman r=0.643, P<0.001) with the manual Cobb angle measurement.
• The algorithm showed good consistency in evaluating scoliosis degree across mild, moderate, and severe cases.
• Receiver Operating Characteristic (ROC) curve analysis indicated high accuracy for classifying moderate (AUC=0.865) and severe (AUC=0.873) scoliosis.

Conclusions:

• The developed machine learning-based computer vision method offers a quick and efficient approach to assess scoliosis severity.
• The standardized integral area algorithm is a valuable tool for screening scoliosis in large datasets and complements traditional Cobb angle measurements.