To design a quasi-automated three-dimensional reconstruction method of the spine from biplanar X-rays as the daily used method in clinical routine is based on manual adjustments of a trained operator and the reconstruction time is more than 10 minutes per patient. Methods: The proposed method of 3D reconstruction of the spine (C3-L5) relies first on a new manual input strategy designed to fit clinicians' skills. Then, a parametric model of the spine is computed using statistical inferences, image analysis techniques and fast manual rigid registration. Results: An agreement study with the clinically used method on a cohort of 57 adolescent scoliotic subjects has shown that both methods have similar performance on vertebral body position and axial rotation (null bias in both cases and standard deviation of signed differences of 1mm and 3.5° around respectively). In average, the solution could be computed in less than 5 minutes of operator time, even for severe scoliosis. Conclusions: The proposed method allows fast and accurate 3D reconstruction of the spine for wide clinical applications and represents a significant step toward full automatization of 3D reconstruction of the spine. Moreover, it is to the best of our knowledge the first method including also the cervical spine.

3D reconstruction of the spine using biplanar X-rays remains approximate and requires human-machine interactions to adjust the position of important features such as vertebral corners and endplates. The purpose of this study is to develop a method to extract automatically the accurate position of lumbar vertebrae posterior corners. In the proposed method we select corner point candidates from an initial edge map. A dedicated pipeline is designed to discard unwanted candidates, involving polyline simplification, curvature thresholding and multiscale Haar filtering. Ultimately, we use a priori knowledge derived from an initial 3D spine model to define search areas and select the final corner points. The framework was tested on 21 biplanar X-rays from scoliotic children. Corner positions are compared with manual selections by two experts. The results report a localization accuracy between 0.7 and 1.6 mm, comparable to manual expert variability.

Quantitative measurements of spine shape parameters on planar X-ray images is critical for clinical applications but remains tedious and with no fully-automated solution demonstrated on the whole spine. This study aims to limit manual input, while demonstrating precise vertebrae corners positioning and shape parameter measurements from sagittal radiographs of the cervical and lumbar regions, exploiting novel dedicated visual features and specialized classifiers. A database of manually annotated X-ray images is used to train specialized Random Forest classifiers for each spine regions and corner types. An original combination of local gradient characteristics, Haarlike features, and contextual features based on patch intensity and contrast is used as visual features. The proposed method is evaluated on 49 sagittal X-rays of asymptomatic and pathological subjects, from multiple imaging sites, and with a large age range (6-69 years old). Performance is first evaluated for positioning a 2D spine shape model, where precisely detected corners enable to adjust the model via an original multilinear statistical regression. Root-mean square errors (RMSE) of corners localization and vertebra orientations are reported, demonstrating state-of-the-art precision compared to existing methods, but with minimal manual input. The method is then evaluated for the extraction of additional vertebrae shape characteristics, such as centre positioning, endplate centres positioning and endplate length measures, rarely studied in previous literature. The proposed method enables, with minimal initialization, fast and precise individual vertebrae delineations on sagittal radiographs on normal and pathological cases, with a level of precision and robustness required for objective support for diagnosis and therapy decision making.