COMPARISON OF SPINAL LOAD PREDICTIONS FROM MUSCULOSKELETAL AND STATISTICAL SHAPE MODELLING

Successful fracture risk stratification of metastatic vertebrae and subsequent personalized intervention planning require understanding of spinal loads on the tissue (transmission through spinal segment) and full-body scale. Spinal loads predicted with musculoskeletal (MSK) modes can provide realistic boundary conditions for multi-component finite element models, but require expert skill and specialized software. Thus, this study evaluates the feasibility of predicting spinal loads directly from spinal sagittal alignment profiles that could offer a rapid approach to support in-silico model development. The upright-standing sagittal postures of 43 spine surgery patients, described with spino-pelvic parameters, and segmental vertebral positions and orientations, were represented in an established AnyBody MSK model [2] to estimate the associated thoraco-lumbar spinal loads. Spinal postures were described with six main shape mode (SM) coefficients, based on a previously established statistical shape model of a normative population [1], to minimise the number of alignment variables and express shape features as deviation from the normative reference. Segmental Multiple Linear Regression (MLR) models were fitted to predict compressive, and shear loads for each thoracolumbar (T1-S1) spinal segment from characteristic SM coefficients. The MLR models showed moderate to strong predictive power for segmental loads, with goodness-of-fit values ranging from 0.38 = R 2 = 0.74 for compressive loads and 0.19 = R 2 = 0.76 for shear loads across the 17 thoracolumbar segments. Out of 119 alignment predictors, significant associations were observed in 60/119 and 59/119 cases, for compressive and shear loads, respectively. Shape modes 3–6 were the main modulators of the loads, with direct relationships observed for β 4 and β 5 and inverse relationships for β 3 and β 6 , in predicting compressive loads. This study demonstrates the relevance of spinal alignment shape features in modulating segmental loads and demonstrates the feasibility of estimating spinal loads directly from sagittal profiles. Moderate/good accuracy of these estimations highlights the trade-off between computational efficiency and accuracy. Acknowledgements. This work is part of the METASTRA project (European Union HEU topic HLTH-2022-12-01 grant 101080135) and was funded from the Swiss State Secretariat for Education, Research and Innovation (SERI). MLR model coefficients (intercept β 0 ; slope coefficients β i ) for compression and shear loads based on SM coefficients. The values of MLR coefficients β i indicate the influence of i th shape mode on segmental loads. The model intercept β 0 corresponds to load magnitude of the normative population mean posture.