Automated Framework for Vertebral Structure Assessment from Medical Imaging under Metastatic Conditions
Please login to view abstract download link
Spinal bone metastases are a severe complication that can cause vertebral fractures and other skeletal-related events (SREs), markedly reducing patient’s quality of life. Selecting the most appropriate treatment is therefore crucial, highlighting the need for predictive tools that anticipate metastatic progression and its structural consequences [1]. Here, we propose a fully automated, patient-specific pipeline for vertebral structural analysis from computed tomography (CT) images. First, machine-learning models perform semantic segmentation of CT scans, removing the need for manual, case-by-case segmentation [2]. Next, patient-specific boundary conditions are derived to capture physiological variability. Vertebral loads are predicted using a machine-learning algorithm trained on musculoskeletal models and then transferred to the target vertebrae via Coherent Point Drift (CPD) point-set registration, enabling automated definition of loading conditions [3]. Finally, fracture risk is quantified across multiple tumour scenarios using structural simulations based on the Cartesian Grid Finite Element Method (cgFEM) [4]. Overall, this framework combines deep learning segmentation, musculoskeletal load prediction, CPD registration, and cgFEM simulation to provide an automated, personalized assessment of vertebral structural behaviour in the presence of metastases, with potential to enhance clinical evaluation and treatment planning.