Classification of Markerless 3D Dorsal Shapes in Adolescent Idiopathic Scoliosis Patients Using Machine Learning Approach

Abstract

Adolescent Idiopathic Scoliosis (AIS) is a musculoskeletal condition commonly seen in pediatric children that causes deformity of the spine. The study aims for early detection and diagnosis as these are the possible options to delimit the progression of the disorder. The work has explored the development of an algorithm that could detect the landmarks and extract the shape-based features from the markerless 3D surface data in AIS patients. An approach to classifying these extracted features using the machine learning algorithm, Support Vector Machine (SVM), has been investigated. The objectives of the work were divided into three frameworks. Framework-1 is aimed at classifying the data based on the asymmetry pattern observed in the spinal surface of the patients. The data corresponding to normal posture were considered as 'without deformity' and data with an asymmetry spinal curve were considered as 'with deformity' based on indicators extracted using the ScolioSIM tool. Framework-2 is aimed at classifying the AIS patients' data based on the three deformity levels namely, mild, moderate and severe. Framework-3 is aimed at classifying the shape orientation of the AIS condition as right or left based on the extracted shape features. The SVM algorithm was able to classify the asymmetry spinal surface pattern and the three deformity levels with accuracy values of 72.4% and 80%, respectively. Furthermore, an accuracy of 94.9% was obtained to classify the shape orientation either as right-or left-oriented. Hence, this noninvasive diagnosis and assessment study paves a new way of approach for the 2D and 3D shape classifications of AIS and expedites the treatment planning process.

Integrated Evolving Spiking Neural Network and Feature Extraction Methods for Scoliosis Classification

“…This further increased to 85.7% using the ReliefF features selection. Classification of the markerless 3D dorsal has been proposed by [13]. In this research, nine shape features and SVM were used to classify the spine with and without deformity, achieving an accuracy of 72.4% and 80%, respectively.…”