Abstract
Malocclusion is a orthodontic condition of teeth misalign- ment, between the upper and lower jaws, affecting roughly half of the global population. Each orthodontic case must be documented in a struc- tured diagnostic report describing occlusal relationships and dental find- ings. Automatic generation of orthodontic reports from intraoral data may reduce documentation burden and improve consistency. Orthodon- tic assessment is inherently multimodal: occlusal relationships are best derived from 3D scans, whereas soft-tissue and restorative findings are primarily visible in intraoral photographs. Here, we introduce Ortho- Scribe, a training-free computational approach that retrieves a similar case report using self-supervised image features, removes unverifiable patient-specific claims, and corrects occlusal descriptors using measure- ments from the 3D meshes. The method was developed on the ODIN 2026 Bite2Text dataset, comprising 997 cases, of which 869 provided usable paired scans, photographs, and English reports, split into 781 training and 88 validation cases. On internal validation, OrthoScribe achieved a RadFact logical F1 of 0.363, outperforming a constant-report baseline (0.326), and a fine-tuned vision–language model (0.274). On the official held-out multicenter test phase, OrthoScribe achieved BLEU-4 and ME- TEOR scores of 0.173 and 0.388, respectively. Code and trained weights are available on GitHub and Zenodo.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_challenges_025.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=vjKJLvG8Od
BibTex
@InProceedings{KacSan_OrthoScribe_MICCAISAT2026,
author = { Kachole, Sanket AND Bakas, Spyridon},
title = { { OrthoScribe: Training-Free Multimodal Orthodontic Report Generation via Photographic Retrieval and 3D Geometry Correction } },
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}
}
