Prompt-Guided Refinement: A Novel Technique for Improving Intervertebral Disc Semantic Labeling

Accurate detection and semantic labeling of intervertebral discs (IVDs) in magnetic resonance imaging (MRI) are crucial for evaluating and treating spinal-related disorders. Conventional approaches typically utilize convolutional neural networks (CNNs) to extract contextual features from MRI images, but they often overlook the inherent geometric structure of the vertebral column, leading to inaccuracies in IVD localization and segmentation. Addressing this limitation, we propose a novel prompt encoder method that incorporates geometric information to enhance the semantic labeling of IVDs in MRI images. Our approach effectively learns the skeleton structure of the spinal column and adaptively adjusts its predictions to conform to this anatomical framework. Extensive evaluations on multi-center spine datasets demonstrate that our method outperforms existing state-of-the-art techniques, consistently achieving superior performance in both T1w and T2w MRI modalities. The incorporation of geometric information significantly improves the accuracy and robustness of IVD semantic labeling, paving the way for more precise and reliable assessments of spine health and disease progression.