Auditory Fiber Bundle Parcellation using Text-Prompted Decomposition Networks

We propose a novel text-prompted bundle decomposition network (TP-BDN) to separately identify fiber bundles in the auditory pathway. TP-BDN consists of two core modules, i.e., 1) multi-modality fusion module (MMFM) and 2) bundle decomposition module (BDM). MMFM integrates complementary anatomical information from multi-modality MRI data. And, BDM decomposes bundle-specific features from the obtained integrative features that are driven by text-prompt features derived by a pretrained Biomed-CLIP. Besides, a bundle shape loss is further presented to guide segmentation from coarse to fine. Extensive experiments are conducted, demonstrating the superiority of our proposed method. The codes will be publicly available at https://github.com/CH2-Carbene/TP-BDN.