Incorporating artificial intelligence and clinical informatics for curve progression risk evaluation in adolescent idiopathic scoliosis to facilitate population screening

Abstract

Adolescent idiopathic scoliosis (AIS) is a complex three-dimensional spinal deformity affecting 2-5% of the general population. AIS is diagnosed when Cobb angles exceed 10°, and deterioration of curve magnitude during puberty occurs in two-thirds of patients. Whilst curves <25° are considered mild, those of 25°-45° are of moderate severity, and those >45° are severe and indicated for surgical correction. A student screening system for AIS has been adopted in Hong Kong since the 1990's. Patients are commonly diagnosed via such screening in their early teens when the curvature is mild, yet it remains uncertain which curves will continue to deteriorate upon the remaining period of growth. Prediction of curve progression risk in AIS remains elusive. Prior studies have revealed the potential for three-dimensional (3D) morphological parameters to prognosticate progression, but these require specialized biplanar imaging equipment and labor-intensive software reconstruction. In addition, patient demographics, vertebral morphology, skeletal maturity, and bone quality represent individual risk factors for progression but have yet to be integrated towards accurate automated prognostication. The objective of this study was to integrate composite clinical information into deep learning model to accurately predict AIS curves at-risk of progression.

In this study, three-dimensional reconstruction was performed upon X-rays at first spinal clinic visit for 138 patients comprising of 65 progressive (P) and 73 non-progressive (NP) curves, and statistical analyses identified that apical rotation of the major curve and torsion were significantly different between P and NP curve trajectories. This formed the basis for deep learning upon a region of interest centered upon the major curve apex. Next, a capsule neural network with self-attention dynamic routing was constructed to differentiate between curves manifesting P and NP trajectories. Following optimization, the predictive model utilizing an ROI centered on the major curve apex achieved an accuracy of 76.6%, a sensitivity of 75.2% and a specificity of 80.2% upon independent testing (n=110). Cross-platform (n=52) performance upon standard standing PA X-rays yielded an accuracy of 77.1%, a sensitivity of 73.5% and a specificity of 81.0%.

A composite deep learning model was then proposed to combine clinical data, global/regional spine X-rays, and hand X-rays for binary classification of P and NP curves. The hand X-ray module performed automated image classification and segmentation tasks towards estimation of skeletal maturity and bone mineral density. A late fusion strategy integrated these domains towards the prediction of progressive curves at first clinic visit. Composite model performance was assessed on a validation cohort and achieved an accuracy of 83.2%, sensitivity of 80.9%, specificity of 83.6% and an AUC of 0.8, outperforming single and bimodal prediction platforms. This study provides an automated means to predict curve progression upon diagnosis of AIS in a cost-effective and noninvasive manner that does not require specialized imaging equipment. This promises to facilitate point-of-care prognostication to guide management. The potential for pre-emptive treatment and a personalized surveillance plan represents a substantial advancement in disease management, especially where screening platforms are already in place.