Biomechanical Parameters

The procedure for replicating the subject-specific spinal alignment with the AnyBody musculoskeletal model (Figures 2C,D), including the rearrangement of ribs and sternum, positioning of the vertebral centers of mass, preservation of the abdominal muscle structure, setting of the trunk muscle parameters, simulation of the load of the raised arms, and muscle co-activation in maintaining the upright posture, is reported in detail in (Barba et al., 2021). In brief, the pelvis is constrained to the ground and rigidly connected to the sacrum. The spinal alignment is replicated by setting the orientation of the sacrum in the sagittal plane and the rotation of the intervertebral spherical joints from T1 to L5, according to the vertebral orientations obtained from the geometrical reconstruction. Joint moments, representing the stiffness-related contribution of passive elements such as ligaments and facet joints, are assumed as zero to replicate neutral upright position. The physiological cross-section area of the trunk muscles is scaled according to reference values acquired in adolescent subjects and depending on age (Been et al., 2018). As regards the scaling of the body model, weight and height were predicted by exploiting linear regression models taking into account anthropometrical and geometrical parameters manually measured on the radiographic images (see Appendix section), since real data were not recorded together with the images in the PACS. These models were trained by another available dataset of 85 AIS subjects with comparable age range and scoliosis severity and known weight and height data, evaluated by our group in a previous study (Bassani et al., 2019). The predicted values were exploited to scale the body model by default length-mass-fat approach. Inverse static analysis was run to calculate muscle activation and intervertebral reaction force (F) in the assigned standing posture. The activity of each muscle fascicle ranged between 0 and 1, obtained by dividing the muscle force by the maximum force generating capacity (set as the product of the cross-section area and the assumed uniform muscle stress, 90 N/cm 2). The asymmetry of erector spinae (ES) and multifidus (MF) muscle activity, between the convex and concave side of the scoliotic curve, was calculated by the normalized activity ratio (nES, and nMF) at each vertebral level inside the curve. As explained in detail in (Barba et al., 2021), this parameter is calculated by accounting for the sum of the activations of the individual fascicles crossing the respective vertebral mid-plane. It measures the (convex − concave)/(convex + concave) activity at specific vertebral level, providing values near zero in correspondence of balanced activation, and positive and negative values (ranging from 0 to ±1) in case of larger activation in the convex and concave side, respectively. As regards F, the absolute value of the intervertebral lateral shear (F_lat), expressed in the local coordinate system of the vertebra (Figure 2E), was taken into account since expected as the most affected by lateral deviations of the spine in the coronal plane which characterize scoliosis. The following eleven biomechanical parameters were accounted for: F_lat, nES, and nMF calculated at apex, upper and lower end levels of the scoliotic curve (Figure 2E), and nES and nMF along the whole curve, obtained by summing the contributions at all levels (from upper to lower end) in the convex and concave side. The setting steps and the simulations were run in batch process using custom routines written in MATLAB (MathWorks Inc., Natick, MA, United States), as well as the procedures for predictive modelling and statistical analysis reported in the next sections.