Differentiating between structural and functional spinal deformities in children and adolescents using machine learning

Abstract: Severe skeletal disorders of children are often mistaken for poor posture, leading to late admittance to treatment and thus limiting treatment outcome and patients’ quality of life. This work proposes a machine learning based methodology to differentiate between structural and functional spinal deformities. Four machine learning algorithms combined with feature engineering were used to train a dataset of 262 rasterstereographic measurements, containing the classes “scoliosis”, “other structural deformities” and “functional deformities”. The overall achieved F1-score was 0.876 with support vector machine. Based on these results, we propose rasterstereography as a novel method for determining the type of spinal deformities.