TriAtt-HRNet: Attention-Enhanced High-Resolution Network for Spine Landmark Detection

Accurate identification of anatomical landmarks in spinal X-ray images plays a vital role in the quantitative diagnosis and clinical management of spinal disorders. In this study, we introduce TriAtt-HRNet, a novel high-resolution network designed for vertebral landmark detection, which incorporates a tri-branch attention mechanism. Built upon the HRNet backbone, our architecture integrates spatial, channel, and combined attention modules to enhance feature representations by capturing both global structural context and fine-grained local details. The proposed method is evaluated on the public BUU-LSPINE dataset using standard metrics, including MAE, MRE, and SDR. Experimental results demonstrate that TriAtt-HRNet consistently outperforms existing state-of-the-art models in terms of accuracy and robustness. These improvements underline the potential of our method to serve as a reliable tool for automated spinal assessment and may contribute to improved clinical workflows in spine diagnosis and treatment planning.