Abstract
Medical-image validation can be reproducible yet answer a different question from the intended geometric task. We test three interpretations in a dental-CBCT case study using patient-paired falsification. First, a corrected, patient-disjoint two-corpus laterality experiment with equal training counts and 20 refits shows early target-minus-source gaps of 1.39% and 1.98% shrink to 0.00% and 0.22% by 320 epochs; deterministic physical-rank rules achieve 1.000 accuracy without fitting. Second, an affine-compensated storage translation preserves every foreground world point but produces only 1.12% and 0.93% decision flips; the 95-case cohort has 38/95 complete-grid feasibility, all invariant controls have zero flips, and the prespecified promotion gate fails. Third, we prove that equal TP, FP, FN, predicted volume, and their named derived overlap scores do not identify a thresholded tooth–canal clearance proxy. A frozen actual-prediction construction yields 1,757 exact pairs in 64/92 patients; at 2 mm its patient-equal away-minus-toward action difference is 0.481 (95% CI [0.402, 0.559]), while 1,752/1,757 (99.7%) pairs require 3–6-mm shifts. These results reject the reconstructed transfer interpretation and establish grid-bounded metric non-identifiability for one dental cohort and one checkpoint. They do not estimate natural-error prevalence, clinical outcomes, safety, or architecture-general behavior.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_065.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MLMI_065_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=plBVSefrHy
BibTex
@InProceedings{MasBeh_LEEWAY_MICCAISAT2026,
author = { Mashayekhi, Behnood AND Amiramjadi, Arian AND Hosseini, Parsa},
title = { { LEEWAY: Same Anatomy, Same Score, Different Verdict Using Paired Falsification Tests in Dental CBCT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}