Abstract
Predicting a structured dental record from cone beam com- puted tomography (CBCT) is difficult because paired records are scarce and several clinical fields are only partly observable from the image. We present GQ-TSR, a Guarded Query Teacher Student Retrieval frame- work that predicts seven clinical record fields from CBCT alone while using unpaired scans during training. Training follows three stages. First, image only pretraining uses all training CBCT volumes to initialize the three dimensional encoder. Then, supervised learning uses paired vol- umes and records to learn global and local visual representations to- gether with structured field semantics. Finally, a guarded EMA teacher and student stage incorporates unpaired scans while retaining the su- pervised solution when semi supervised training is not beneficial. The selected models retrieve each field from labeled reference records, and their predictions are combined across folds. On the STS 2026 Challenge public validation set, GQ-TSR achieved a Weighted Score of 0.26, a field completion rate of 1.00, and no missing predictions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_challenges_009.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=paI9BPN0hv
BibTex
@InProceedings{ZenWei_GQTSR_MICCAISAT2026,
author = { Zeng, Wei},
title = { { GQ-TSR: Guarded Query Teacher–Student Retrieval for CBCT-to-Clinical Record Prediction } },
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}
}