Abstract
In medical image segmentation, reported performance is strongly influenced by the evaluation protocol, especially when datasets such as LIDC-IDRI lack a standard testing procedure. We experimentally study how split granularity, information leakage, multi-reader target construction, and Dice aggregation alter reported CT lung nodule segmentation performance. Using the same 2D/2.5D setup, we compare patient-level and random sample-level splitting and evaluate frozen LIDC-IDRI models on MSD Task06 Lung. The leakage audit finds no train–test patient or image overlap for the patient-level protocol, whereas the sample-level protocol shares 568 of 579 test patients and 724 image identifiers with training. The sample-level split obtains higher internal LIDC-IDRI scores, but this apparent advantage reverses externally, where the patient-level model performs better. A controlled same-image experiment further shows that differences among union, majority-vote, and intersection targets shrink substantially when case selection is fixed. These results are empirical observations from the investigated LIDC-IDRI–MSD setting and emphasize the need to report split granularity, leakage, target construction, metric aggregation, and external-validation details explicitly.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_071.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=9Cag7FQxyc
BibTex
@InProceedings{NanMih_Data_MICCAISAT2026,
author = { Nan, Mihai AND Chiru, Cătălin-Mihail AND Mocanu, Irina AND Florea, Adina Magda},
title = { { Data leakage and external validation in lung CT segmentation } },
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}
}