Generation of severely scoliotic subject-specific musculoskeletal models

Abstract

Surgical correction of severe scoliosis has been associated with high rates of complications and reoperation rates. Patient outcomes could be improved with the use of personalised scoliotic spine models. To be clinically applicable, these models must be robust to operator variability, must accurately reflect the patient anatomy, and be rapidly generated. This study developed a semi-automatic pipeline for creating subject-specific models, and the inter- and intra-operator variability and model accuracy were assessed. An existing generic spine model was modified and morphed into a subject-specific spine model using manually performed virtual anatomical landmark palpations through a semi-automatic pipeline. The inter- and intra-operator variability of the virtual palpations was assessed, and the model was compared to the ground truth of a radiographic evaluation and segmentation of computed tomography data. The interclass correlation coefficient showed excellent inter- and intra-operator repeatability (>0.9). Although significant differences were found, they were not associated with specific anatomical landmarks. The mean inter- and intra-operator variability of the virtually palpated anatomical landmarks was 2 mm, and the maximum was 10.3 mm. The vertebral centres of the patient-specific model had maximum median errors of 10.3 mm. While the mean curvature of the model reflected the radiographic measurements, there were substantial deviations from the mean. The semi-automatic pipeline successfully created a subject-specific scoliotic spine model that included automatically adjusted muscle paths. The results indicated the process was robust to inter- and intra-operator variability but would benefit from full automation, particularly to improve the definition of the intervertebral joint.