Learning Spatial Configuration Feature for Landmark Localization in Hand X-rays

Medical landmark localization is crucial for treatment planning. Although FCN-based heatmap regression methods have made significant progress, there a lack of research focused on features that can learn spatial configuration between medical landmarks, notwithstanding the well-structured patterns these landmarks. In this paper, we propose novel spatial-configuration-feature-based network effectively learns anatomical correlation Specifically, focus regularization method and loss capture relationship Each heatmap, generated using U-Net, transformed into an embedded feature vector soft-argmax maps, here, Cartesian Polar coordinates. A map landmarks based used to calculate loss, along with output. This approach adopts end-to-end learning approach, requiring only single feedforward execution during test phase localize all The proposed computationally efficient, differentiable, highly parallelizable. experimental results show our global contextual achieve state-of-the-art performance. Our expected significantly improve accuracy when applied healthcare systems require accurate localization.