Fetal Cerebellum Landmark Detection Based on 3D MRI: Method and Benchmark

Haifan Gong, Huixian Liu, Yitao Wang, Xiaoling Liu, Xiang Wan, Qiao Shi, Haofeng Li

Abstract:

Fetal cerebellum landmark detection is cru-cial for assessing fetal brain development. Although deeplearning has become the standard for automatic landmarkdetection, most previous methods have focused on us-ing 2D ultrasound or thick Magnetic Resonance Imaging(MRI). To improve accuracy, landmarks should be locatedon thin 3D MRIs. However, abnormal development, highnoise, and fuzzy boundaries in 3D fetal brain images maketraditional methods less effective for cerebellum landmarkdetection. To address this, we introduce the AnatomicalPseudo-label Guided Attention (APGA) network alongside a3D MRI-based benchmark for fetal cerebellum landmark de-tection. During training, we use a shared encoder to extractimage features and two decoders for landmark regressionand anatomical pseudo-label segmentation. We design aFeature Decoupling Transformer (FDT) and embed it intothe encoder to better calibrate the features for the twotasks. We only need the encoder, the FDT, and the landmarkdecoder during the inference phase. Extensive experimentson our proposed benchmark and out-of-domain test sethave shown the effectiveness of our method. Our simula-tions also demonstrated that 3D biometrics are better than2D biometrics.

(PDF) Fetal Cerebellum Landmark Detection Based on 3D MRI: Method and Benchmark. Available from: https://www.researchgate.net/publication/390672491_Fetal_Cerebellum_Landmark_Detection_Based_on_3D_MRI_Method_and_Benchmark [accessed Nov 12 2025].