BA-MSAT-Net: A Multi-Scale Attention and Transformer for 3D Tooth Segmentation in CBCT Images

Résumé fourni par la source

BA-MSAT-Net, a boundary-aware three-dimensional (3D) encoder–decoder framework for segmentation of tooth and alveolar structures. Specifically, residual volumetric encoding and multi-scale spatial modeling were used to extract local anatomical details and heterogeneous contextual information. Transformer-based self-attention was incorporated at the bottleneck to model long-range crown–root–alveolar-bone relationships, attention-gated skip fusion suppressed irrelevant background responses during feature reconstruction. An auxiliary boundary-prediction branch with boundary-aware supervision was further introduced to refine anatomically ambiguous interfaces. BA-MSAT-Net improves volumetric accuracy and boundary delineation in multi-structure CBCT segmentation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.