In Europe, hundreds of thousands of breast cancer cases are diagnosed per annum. Of these patients, two-thirds will develop metastatic bone disease (MBD), the most common target being the spine. Spinal MBD results in the degradation of the bone in the spine and can lead to vertebra fracture. The treatment for the most severe cases consists of major surgery with recuperation time of 6-8 months, a significant amount of time for patients with short life expectancy [1]. There is an outstanding need for diagnostic tools which evaluate patient-specific fracture risk. Here, we propose a novel computational framework where patient-specific simulations ascertain the loading distribution (and thus fracture risk) in a vertebra with a growing metastasis. Our approach can, from a CT, create patient-specific geometries through a deep-learning architecture. The architecture is broken down into three sections: (1) a U-Net places a bounding box around the spine, (2) a Spatial Configuration Network (SC-NET) identifies the centroid of each vertebra, and (3) a Statistical Shape Model (SSM) segments the vertebra. It was fitted against hundreds of CT scans (none with MBD). From the segmentations, volumetric meshes are created and imported into Firedrake, a finite element solver in Python. Simulations of different loading regimes can then be performed, where it is assumed that the bone and tumours behave, individually, as homogeneous, isotropic and elastic-plastic materials with properties defined based on data from the literature. As a proof of concept, the framework was applied to simulate the lumbar portion of the spine. Our ambition is to extend the framework to encompass the whole spine and validate it using spine MBD data. With such a framework, we hope to provide a quantitative tool to empower clinicians to make decisions for patients at risk of vertebra fracture due to spinal MBD.