Prior Guided 3D Medical Image Landmark Localization

Accurate detection of 3D medical landmarks is critical for evaluating and characterizing anatomical features and performing preoperative planning. However, detecting 3D landmarks can be challenging due to the local structural homogeneity of medical images. In this study, we present a prior guided coarse-to-fine framework for efficient and accurate 3D medical landmark detection. Specifically, we utilize the prior knowledge that in specific settings, physicians often annotate multiple landmarks on a same slice. In the coarse stage, we perform coordinate regression on downsampled 3D images to maintain the structural relationships across different landmarks. In the fine stage, we categorize landmarks as independent and correlated landmarks based on their annotation prior. For each independent landmark, we train a single localization model to capture local features and deliver reliable local predictions. For correlated landmarks, we mimic the manual annotation process and propose a correlated landmark detection model that fuses information from various patches to query key slices and identify correlated landmarks. The proposed method is extensively evaluated on two datasets, exhibiting superior performance with an average detection error of respective 3.29 mm and 2.13 mm.