Abstract
Registering intraoral scans to CBCT is challenging due to modality differences and varying coordinate handedness. We propose a handedness-aware framework for MICCAI STS 2026 Task 2. The method normalizes STL and CBCT surfaces, encodes them via modality-specific PointNet++, and establishes soft correspondences with bidirectional cross- attention. A differentiable Procrustes layer estimates O(3) transforma- tions, enabling both rotations and reflections. A bounded residual branch corrects translation only. We adopt a two-stage semi-supervised strategy. The PointNet++ encoders are first pre-trained on the unlabeled train- ing cases using contrastive self-supervision and are then fine-tuned on the limited labeled cases using ground-truth rigid transformations. Training combines point, rotation, translation, and residual losses; inference uses five-sample test-time aggregation. On a development-only stratified split of six held-out patients (24/6), the upper model achieved 15.402 mm translation and 9.287◦rotation error; the lower model achieved 27.612 mm and 12.640◦. Jaw-averaged errors were 21.507 mm and 10.963◦. These component choices were not independently isolated on this small split. On the official final test set, the method achieved a translation error of 39.3343 mm and a rotation error of 11.6964◦.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_challenges_003.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=gRxV2w8ei3
BibTex
@InProceedings{LinZhi_HandednessAware_MICCAISAT2026,
author = { Lin, Zhihao AND Yan, Tianxu AND Zhang, JinXuan AND Shen, Wu AND Lin, Hui AND Zhang, Dongping},
title = { { Handedness-Aware Cross-Attention SVD for Multimodal Dental Registration } },
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}
}