Prediction of spinal curve progression in Adolescent Idiopathic Scoliosis using Random Forest regression

Highlights

• Morphologic descriptors could help characterize changes in the spine in 3D space.

• Using independent components, we predicted the evolution of the spine's shape.

• The predictor models the changes in the spine up to 18 months from first visit.

• Prediction of the shape of the spine could help devise patient-specific treatment.

• Future research may detect curve patterns signaling progressive idiopathic scoliosis.

Abstract

Background

The progression of the spinal curve represents one of the major concerns in the assessment of Adolescent Idiopathic Scoliosis (AIS). The prediction of the shape of the spine from the first visit could guide the management of AIS and provide the right treatment to prevent curve progression.

Method

In this work, we propose a novel approach based on a statistical generative model to predict the shape variation of the spinal curve from the first visit. A spinal curve progression approach is learned using 3D spine models generated from retrospective biplanar X-rays. The prediction is performed every three months from the first visit, for a time lapse of one year and a half. An Independent Component Analysis (ICA) was computed to obtain Independent Components (ICs), which are used to describe the main directions of shape variations. A dataset of 3D shapes of 150 patients with AIS was employed to extract the ICs, which were used to train our approach.

Results

The approach generated an estimation of the shape of the spine through time. The estimated shape differs from the real curvature by 1.83, 5.18, and 4.79° of Cobb angles in the proximal thoracic, main thoracic, and thoraco-lumbar lumbar sections, respectively.

Conclusions

The results obtained from our approach indicate that predictions based on ICs are very promising. ICA offers the means to identify the variation in the 3D space of the evolution of the shape of the spine. Another advantage of using ICs is that they can be visualized for interpretation.

Introduction

Adolescent idiopathic scoliosis (AIS) is a complex 3D deformation of the spine which looks like an “S” or “C” shape from the posteroanterior plane. It is called idiopathic because its cause is unknown. It is the most common type of scoliosis, with a high prevalence in adolescents between 10 and 18 years of age. AIS affects between 1 and 4% of adolescents, mainly females [1]. In a meta-analysis, Cheng et al. [2] showed that the global prevalence of AIS with the main curvature ≥ 10° was 1.34%. Currently, the evaluation of the spine relies mainly on the observation of conventional posteroanterior and sagittal X-rays, which constitute the most common imaging modalities for observing the spine in a standing position in clinical practice.

The Cobb angle represents the gold standard method for measuring the curvature of the spine. Its measurement is based on the most tilted vertebrae, at the top (upper vertebra) and at the bottom (lower vertebra) of the curve. The angle is formed by the line parallel to the superior endplate of the upper vertebra and the inferior endplate of the lower vertebra. It should however be noted that the Cobb angle has certain limitations. First and foremost, it is a measurement of a 3D spinal deformity from 2D radiographs. This is noteworthy because two spines with radically different 3D morphologies could yield similar Cobb angle estimations [3]. Furthermore, it is known that Cobb angle measurements could vary by up to 10° [4]. This is relevant since two spines with similar curves may render different recommendations for treatment [3].

Predictions of the progression of a spinal curve should provide valuable insights into how the deformation is going to evolve and should greatly assist in guiding treatment strategies. Maturity (chronological, skeletal, and menarcheal age), curve magnitude, and curve location [1] have traditionally been the main clinical indices used to assess spinal curve progression, with treatment decision based mainly on the curve magnitude: because the Cobb angle is normally used to assess the curve magnitude, this therefore means that the treatment depends on high-variability measurements.

Other clinical indices, such as different body length dimensions (sitting height, subischial leg length, and foot length or shoe size), secondary sexual characteristics, skeletal age in different areas, the Risser index, status of the triradiate cartilage, and electromyography ratios of the paraspinal muscle activity, have also been considered as predictors of curve progression [[5], [6], [7], [8], [9], [10], [11]]. Additionally, the relationship between a rapid growth of the patient and the evolution of the spinal deformity has been widely studied [[5], [6], [7], [8],[12], [13], [14]]. Noshchenko et al. [15] carried out a systematic review of 25 studies presenting clinical parameters that are statistically significantly associated with the progression of AIS. However, the parameters presented a limited or little evidence as predictors of the final deformation.

Studying the analysis of the spine in 3D is of vital importance, since it can lead to a more relevant and reliable 3D classification method for assessing and treating AIS [3]. In this respect, computerized clinical indices [16] and geometric descriptors [[17], [18], [19], [20], [21], [22]] have been proposed to capture the complexity of the spinal deformity. However, characterizing the spine in 3D space with meaningful descriptors is still challenging. This characterization must be capable of retaining the most significant information, not only in order to achieve the highest classification performance, but also to be clinically relevant.

In statistical shape analysis, methods such as Active Shape Models or Active Appearance Models have been used to study the main directions of shape variations [[23], [24], [25]] with the objective of mapping high-dimensional feature vectors onto lower-dimensional representations, while maintaining most of the variability of the original dataset. Usually, these models use Principal Component Analysis (PCA) to derive the low-dimensional representation of the data. The eigenvectors with the highest variance are used as modes of shape variations. The main disadvantage of PCA is the assumption of a Gaussian distribution of data, which could lead to incorrect descriptions.

Using support vector machines, Assi et al. [26] analyzed several dimensional reduction techniques, which were used before surgery to predict the postoperative appearance of a patient's trunk. Recently, a supervised model based on discriminant manifolds was proposed to study the 3D morphology of the curve progression [22]. The samples in the dataset of the latter were labeled as progressive and non-progressive, based on the Cobb angle. However, since there are many forces acting simultaneously in the curve progression, the prediction could fail if only patterns related to the Cobb angle are considered, which may not necessarily characterize the progression in a 3D space in sufficient detail.

Independent Component Analysis (ICA) is another technique that has been used in shape analysis to obtain the modes of shape variations [[27], [28], [29], [30]]. Unlike PCA, ICA generates independent non-Gaussian components. It also takes into account higher-order moments of data distribution, instead of variance maximization, as in PCA. Hence, ICA could obtain more representative modes of variation from the dataset.

In this work, we propose an approach for predicting the progression of spinal curvatures using ICA to capture the modes of shape variation of 3D models of the spine from a cohort of patients with AIS. We compared the performance of shape variation modes obtained with ICA against a low-dimensional representation of 3D models of the spine, generated from Stacked Denoising Autoencoders (SDAE).