Abstract
The MICCAI STSR 2026 Challenge Task 3 aims to derive clinically relevant information and structured dental reports from three- dimensional cone-beam computed tomography (CBCT). To address the limited labeled data and the complex spatial relationships among mul- tiple teeth, we propose a segmentation-guided and FDI-aligned three- dimensional dental ROI Transformer. An nnU-Net-based tooth segmen- tation model is first employed to localize individual teeth. Instead of applying the segmentation as a hard mask, the proposed method crops the complete dentition together with surrounding periapical tissues, pe- riodontal structures, neighboring teeth, and alveolar bone. A 3D residual convolutional network extracts a global dentition representation and in- termediate tooth-level ROI features. Each tooth token integrates its local visual feature, FDI positional embedding, tooth-presence information, and segmentation-derived reliability descriptors. A Tooth Transformer and attention-pooling module are then used to model inter-tooth re- lationships and obtain a dentition-level representation. The global and tooth-level features are fused for multi-task prediction of tooth nota- tions, diagnosis categories, treatment actions, medications, and auxiliary clinical information. The predicted entities are converted into the seven required report fields using a deterministic template-based module. On the validation set, our method achieved a total score of 10.08 and ranked second in Task 3, demonstrating the effectiveness of segmentation-guided volumetric and tooth-level representation learning for structured dental CBCT reporting. Our code are available at https://github.com/duola- wa/MICCAI-2026-STS-Task-3.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_challenges_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=UDcQNP7rBH
BibTex
@InProceedings{JiCha_SegmentationGuided_MICCAISAT2026,
author = { Ji, Changkai AND Liu, Yusheng AND Jiang, Yuxian AND Wang, Lisheng},
title = { { Segmentation-Guided 3D Dental ROI Transformer for Structured Clinical Report Generation from CBCT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17270},
month = {pending},
page = {pending}
}
