Abstract
Retrieving clinically similar historical cases from dental imag- ing archives supports diagnosis, treatment planning, and education, yet remains largely unaddressed for cone-beam computed tomography (CBCT). We propose a diagnosis-aware multimodal retrieval framework that fine- tunes BioMedCLIP with a cross-modal supervised contrastive objective, learning a shared embedding space for CBCT volumes and clinical notes. CBCT volumes are represented via informativeness-weighted axial slice pooling, emphasizing anatomically informative regions over background- heavy slices, while clinical notes are encoded with a PubMedBERT-based text encoder. During fine-tuning, patients sharing a diagnosis are treated as additional positive pairs weighted by Jaccard diagnosis similarity across both modalities, encouraging clinically related cases to cluster be- yond same-patient correspondence alone. We evaluate on MMDental, the first public multimodal dental dataset pairing 403 CBCT volumes with expert clinical records, across scan-to-scan, scan-to-note, note-to-scan, and note-to-note retrieval. Combined with diagnosis-overlap reranking, our approach achieves 98.0% and 97.0% Recall@1 for note-to-note and note-to-scan retrieval respectively, improves cross-modal alignment more than five-fold (0.175 →0.928), and yields 77.0% clinically meaningful top-1 retrievals overall, substantially outperforming lexical, sparse, and zero-shot vision-language baselines.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_016.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ODIN_016_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=jVl756HoqF
BibTex
@InProceedings{BinNeh_DiagnosisAware_MICCAISAT2026,
author = { Binoy, Neha Ann AND Sankar J., Nandana AND Nair, Neena S. AND George, Ajo Babu AND Abraham, Ashish Anton},
title = { { Diagnosis-Aware Multimodal Retrieval for Paired CBCT Scans and Clinical Notes } },
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}
}
