A Multi-views Models Ensemble Method for Thumb Trapeziometacarpal Joint Segmentation from Computed Tomography Images

This paper presents a cost-effective methodology for automatic segmentation of trapeziometacarpal (TMC) joint morphology from computed tomography (CT) images. We utilize the properties of CT imaging by incorporating multiple views into our approach to enhance segmentation accuracy. Instead of a single model, our pipeline utilizes an ensemble of multiple segmentation models, each trained to recognize distinct features from different perspectives of the TMC joint. The results from these multiple models are then consolidated using a bagging technique to generate the final segmentation output. Our framework provides a promising alternative to resource-intensive 3D segmentation networks by delivering comparable performance at significantly lower computational costs. This study serves as an important step towards more efficient and effective segmentation techniques in medical imaging, facilitating better diagnostics and treatment plans.