Abstract
2D/3D registration is a key enabling algorithm in image-guided surgery, yet even state-of-the-art algorithms occasionally produce clinically significant misalignments. Because even millimeter-scale errors can lead to surgical complications, robust quality assurance is essential. However, current strategies are insufficient for reliable human verification, and existing uncertainty quantification methods do not robustly correlate with registration error. To address these gaps, we present a learning-based framework for 2D/3D registration quality assessment, augmented with explainability mechanisms to support human oversight. Our approach employs early fusion of fluoroscopic images and digitally reconstructed radiographs (DRRs), processed through a convolutional backbone with spatial cross-attention to capture both local alignment features and global misalignment patterns. The model predicts registration success, defined as mean target registration error (mTRE) below 2 mm. To provide interpretable rationale, we integrate Grad-CAM for spatial attribution and conformal prediction for calibrated confidence estimates. Evaluated using leave-one-specimen-out cross-validation on a pelvic fluoroscopy dataset, our proposed approach achieves an accuracy of 0.801 ± 0.030, outperforming a baseline CNN without cross-attention while exhibiting reduced variance across specimens. This framework establishes a foundation for explainable, AI-assisted quality assurance in 2D/3D registration. Our code is available at: https://github.com/jasminecho1008/XAI-2D3D-RegQuality.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SafeSurg_016.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=zCXGwNMw1y
BibTex
@InProceedings{ChoSue_Explainable_MICCAISAT2026,
author = { Cho, Sue Min AND Do, Alexander AND Taylor, Russell H. AND Unberath, Mathias},
title = { { Explainable AI for Image-based 2D/3D Registration Quality Assessment } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS pending},
month = {pending},
page = {pending}
}