S. Parent

A hierarchical statistical modeling approach for the unsupervised 3D reconstruction of the scoliotic spine

In this paper, we propose a new and accurate 3D reconstruction technique for the scoliotic spine from a pair planar and conventional radiographic images (postero-anterior and lateral). The proposed model uses a priori hierarchical global knowledge, both on the geometric structure of the whole spine and of each vertebra. More precisely, it relies on the specification of two 3D templates. The first, a rough geometric template on which rigid admissible deformations are defined, is used to ensure a crude registration of the whole spine. 3D reconstruction is then refined for each vertebra, by a template on which nonlinear admissible global deformations are modeled, with statistical modal analysis of the pathological deformations observed on a representative scoliotic vertebra population. This unsupervised coarse-to-fine 3D reconstruction procedure is stated as a double energy function minimization problems efficiently solved with a stochastic optimization algorithm. The proposed method, tested on several pairs of biplanar radiographic images with scoliotic deformities, is comparable in terms of accuracy with the classical CT-scan technique while being unsupervised and requiring a lower amount of radiation for the patient.

3D biplanar reconstruction of scoliotic vertebrae using statistical models

This paper presents a new 3D reconstruction method of the scoliotic vertebrae of a spine, using two conventional radiographic views (postero-anterior and lateral), and global prior knowledge on the geometrical structure of each vertebra. This geometrical knowledge is efficiently captured by a statistical deformable template integrating a set of admissible deformations, expressed by the first modes of variation in the Karhunen-Loeve expansion of the pathological deformations observed on a representative scoliotic vertebra population. The proposed reconstruction method consists in fitting the projections of this deformable template with the segmented contours of the corresponding vertebra on the two radiographic views. The 3D reconstruction problem is stated as the minimization of a cost function for each vertebra and solved with a gradient descent technique. The reconstruction of the spine is then made vertebra by vertebra. This 3D reconstruction method has been successfully tested on several biplanar radiographic images, yielding very promising results.

A computer-based classifier of three-dimensional spinal scoliosis severity

ObjectiveThis article describes a computer-based method for the classification of spine scoliosis severity. This is a first step toward an effective computerized tool to assist general practitioners diagnose spine scoliosis. The method progresses away from Cobb angles toward pattern and magnitude categorization based upon 3D configurations.Materials and methodsThe purpose is to classify spine shapes reconstructed from a pair of calibrated X-ray images into one of three categories, namely, normal spine, moderate scoliosis, and severe scoliosis. The spine shape is represented by the three-dimensional coordinates of a sequence of equidistant points sampled by interpolation on the reconstructed spine shape. Classification is carried out using a self- organizing Kohonen neural network trained using this representation.ResultsThe tests were performed using a database of 174 spine biplane X-rays. The classification accuracy was 97%.ConclusionThe results demonstrate that classification of 3D spine descriptions by a Kohonen neural network affords a solid basis for an effective tool to assist clinicians in assessing scoliosis severity.

3D correction of AIS in braces designed using CAD/CAM and FEM: a randomized controlled trial

BackgroundRecent studies showed that finite element model (FEM) combined to CAD/CAM improves the design of braces for the conservative treatment of adolescent idiopathic scoliosis (AIS), using 2D measurements from in-brace radiographs. We aim to assess the immediate effectiveness on curve correction in all three planes of braces designed using CAD/CAM and numerical simulation compared to braces designed with CAD/CAM only.MethodsSRS standardized criteria for bracing were followed to recruit 48 AIS patients who were randomized into two groups. For both groups, 3D reconstructions of the spine and patient’s torso, respectively built from bi-planar radiographs and surface topography, were obtained and braces were designed using the CAD/CAM approach. For the test group, 3D reconstructions of the spine and patient’s torso were additionally used to generate a personalized FEM to simulate and iteratively improve the brace design with the objective of curve correction maximization in three planes and brace material minimization.ResultsFor the control group (CtrlBraces), average Cobb angle prior to bracing was 29° (thoracic, T) and 25° (lumbar, L) with the planes of maximal curvature (PMC) respectively oriented at 63° and 57° on average with respect to the sagittal plane. Average apical axial rotation prior to bracing was 7° (T) and 9° (L). For the test group (FEMBraces), initial Cobb angles were 33° (T) and 28° (L) with the PMC at 68° (T) and 56° (L) and average apical axial rotation prior to bracing at 9° (T and L). On average, FEMBraces were 50% thinner and had 20% less covering surface than CtrlBraces while reducing T and L curves by 47 and 48%, respectively, compared to 25 and 26% for CtrlBraces. FEMBraces corrected apical axial rotation by 46% compared to 30% for CtrlBraces.ConclusionThe combination of numerical simulation and CAD/CAM approach allowed designing more efficient braces in all three planes, with the advantages of being lighter than standard CAD/CAM braces. Bracing in AIS may be improved in 3D by the use of this simulation platform. This study is ongoing to recruit more cases and to analyze the long-term effect of bracing.Trial registrationClinicalTrials.gov, NCT02285621

Pelvic Osteotomy for Spinal Deformities

Much like lower limb axes are important in the planning and correction of limb deformity, sagittal spinal alignment parameters are key in the assessment and treatment of spinal deformity. They include the standard sagittal spinal alignment parameters (thoracic kyphosis and lumbar lordosis), sagittal sacropelvic alignment (pelvic incidence, pelvic tilt, and sacral slope), and global sagittal balance parameters such as the C7 plumb line, spinal sacral angle, and spinal tilt. In patients with an abnormal sagittal spinal alignment, vertebral osteotomies such as pedicle subtraction osteotomies and fusion can be employed to correct the alignment. However, this can be technically very demanding, be prone to significant neurologic complications, and decrease spinal mobility. Alternatively, pelvic osteotomies can also be used to restore sagittal alignment, thus preserving lumbar mobility and positively impacting the patient’s quality of life and function.

Convolutional Neural Network and In-Painting Techniques for the Automatic Assessment of Scoliotic Spine Surgery from Biplanar Radiographs

Assessing the effectiveness of scoliosis surgery requires the quantification of 3D spinal deformities from pre- and post-operative radiographs. This can be achieved from 3D reconstructed models of the spine but a fast-automatic method to recover this model from pre- and post-operative radiographs remains a challenge. For example, the vertebrae’s visibility varies considerably and large metallic objects occlude important landmarks in postoperative radiographs. This paper presents a method for automatic 3D spine reconstruction from pre- and post-operative calibrated biplanar radiographs. We fitted a statistical shape model of the spine to images by using a 3D/2D registration based on convolutional neural networks. The metallic structures in postoperative radiographs were detected and removed using an image in-painting method to improve the performance of vertebrae registration. We applied the method to a set of 38 operated patients and clinical parameters were computed (such as the Cobb and kyphosis/lordosis angles, and vertebral axial rotations) from the pre- and post-operative 3D reconstructions. Compared to manual annotations, the proposed automatic method provided values with a mean absolute error <5.6° and <6.8° for clinical angles; <1.5 mm and <2.3 mm for vertebra locations; and <4.5° and <3.7° for vertebra orientations, respectively for pre- and post-operative times. The fast-automatic 3D reconstruction from pre- and post in-painted images provided a relevant set of parameters to assess the spine surgery without any human intervention.

Torso Shape Extraction from 3D Body Scanning Data Using Automatic Segmentation Tool

The automatic and standardized extraction of the torso shape from 3D body scanning data has an important role in biomedical applications. In scoliosis clinics, the asymmetry analysis of the 3D scoliotic trunk shape relies on a prior cropping of the regions corresponding to the arms and neck. At Sainte-Justine Hospital, a system of four optical digitizers (Capturor II LF, Creaform Inc.) is used to scan the body of scoliosis patients. At present, the cropping of the trunk shapes is a manual process and is therefore operator-dependent, time-consuming and can affect the reliability of subsequent trunk asymmetry analysis. In addition, the inferior portion of the trunk (pelvic region) includes noisy geometric features that are due to the patient’s lower body clothing and are irrelevant to the study of scoliotic trunk shape deformations. In this paper, we present a robust and efficient tool to extract the meaningful torso regions based on automatic segmentation. The 3D body scanning system provides a 3D triangulated mesh of the shape accompanied by an RGB color map of the texture. An anatomical landmark placed at the midpoint of the posterior-anterior iliac spines (MPSIS) prior to the acquisition determines the separation level between the pelvic region and the rest of the torso (i.e. the lumbar and thoracic regions). We propose a two-phase segmentation algorithm. In the first phase, a skin-color model is used to separate the pelvic region from the other portions of the torso. The second phase separates the arms and neck regions using relevant geometric features captured by a spectral representation of the shape. We tested our algorithm on a dataset composed of 56 scoliotic body shapes scanned in neutral standing and lateral bending postures by comparing the torsos cropped automatically versus manually by an operator. The results show that our algorithm achieves a 0.95 (± 0.04) degree of overlap, in terms of the average Dice similarity measure, between the extracted torso shapes and their ground truth counterparts. The proposed automatic segmentation method thus constitutes a useful tool to include in the 3D body surface scanning systems used in scoliosis clinics.