A novel method of anatomical landmark selection for rib cage 3D reconstruction from biplanar radiography

Abstract

Methods to reconstruct anatomical structures in 3D are gaining interest in medicine because they give access to quantitative information on the patient's geometry. However, these methods are user-dependent and require a trained operator, which is time consuming and a source of error and unreliability. The aim of this work was to validate a novel method of landmark selection to perform the 3D reconstruction of the rib cage from biplanar calibrated radiographies. The method uses digital painting for digitization of anatomical landmarks (eight ribs midlines, posterior extrema, sternum) to build a first estimate of the 3D ribcage geometry. Twenty scoliotic patients were included (Cobb angle: 43° ± 11°) and their ribcage was reconstructed twice with the proposed method by four trained operators. Measurement reproducibility was similar to previously validated methods. Uncertainty (95% CI) was 2.3° for the rib hump measurement, 9.7 mm and 3.8 mm for maximal antero-posterior and lateral diameter, 395 cm 3 for ribcage volume. The method was qualitatively considered more user-friendly than previous versions, although it still requires a trained operator, and it took approximately 2 minutes of manual digitization. The new method should facilitate diffusion of 3D quantitative analysis of ribcage in clinical routine.

A convolutional neural network to detect scoliosis treatment in radiographs

“…This is the most adopted position in radiological examination of scoliosis because the arms do not hide the spine neither in frontal radiographs nor in lateral ones, and the ribcage remains fairly visible in both views. Indeed, for a three-dimensional assessment of scoliosis, biplanar radiographs are acquired when possible, since they allow 3D reconstruction of the bony structures [21,22]. It is possible that feeding radiographs with patients in different position (such as in lateral bending [23]) could alter the results.…”

Towards a predictive simulation of brace action in adolescent idiopathic scoliosis

“…The spine (T1-L5), ribcage and pelvis of each patient was reconstructed in 3 D with previously validated methods (Humbert et al 2009;Ghostine et al 2017;Vergari et al 2020), ; both before and in-brace (Figure 1). This reconstruction allowed automatic calculation of 3 D geometrical spinopelvic parameters, but it also provided 3 D models of all vertebrae, pelvis and rib cage, that were used to build a subject-specific finite element model of the trunk (Figure 1).…”

“…Patient balance was predicted through the analysis of balance compensation mechanisms between prebrace and in brace geometrical parameters (Vergari et al 2020). It was observed that the variation induced by the brace on the sagittal and axial orientation of T1 vertebra and on the sagittal orientation of the pelvis (i.e., pelvic tilt) could be predicted from pre-brace values (Figure 2).…”

“…However, it is still unclear how to take into account the compensation mechanisms deployed by the patient to keep balance in-brace (Vergari et al 2020). This limitation was underlined in Vergari et al (2016), where the in-brace pelvic tilt and the position of T1 relative to the pelvis were imposed in the model as boundary conditions.…”

Trunk Growth in Early-Onset Idiopathic Scoliosis Measured With Biplanar Radiography

“…The recent development and diffusion of biplanar stereoradiography gave access to high-quality radiographs without distortion while being 8e10 times less irradiating than conventional radiographs and 800 times less than computed tomography. 3D reconstruction methods have been developed, which can provide quantitative assessment of the trunk in the standing position [16][17][18][19]. Nevertheless, these methods were validated for adolescent idiopathic scoliosis patients as young as 10 years.…”