Near-duplicate contamination inflates brain MRI benchmark performance: a size-matched, multi-seed MedNeXt-B study.
- 1The First Clinical Medical College, The Affiliated People's Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian, China.
- 2Department of Interventional Radiology, The Affiliated People's Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian, China.
Abstract
Background
Pre-partitioned brain tumor MRI collections may place exact or near-duplicate slices on both sides of a benchmark split, but the magnitude and uncertainty of the resulting performance difference are rarely measured.Methods
A unified MedNeXt-B 2D model produced a four-class label and a tumor mask from a shared encoder. For each near-duplicate-screened fold, we constructed a replacement-based contaminated condition of identical size and tumor-class-by-plane composition; this condition was not the original unscreened training pool. Validation folds and the frozen 995-slice test set were identical. Constituent source-dataset matching was not possible because BRISC2025 provides no per-image source provenance. Both conditions were trained with seeds 42, 1,337, 2,024, 3,407, and 5,090. Primary uncertainty summaries used a 10,000-resample hierarchical bootstrap over seeds and inferred pseudo-patient clusters.Results
Across five seeds, screened accuracy was 0.9652 ± 0.0023 and mean class Dice was 0.7894 ± 0.0031 (mean ± sample SD). Accuracy was higher in the replacement-based contaminated condition by a mean difference of 0.0261 ± 0.0031 (range 0.0211-0.0281; hierarchical 95% CI, 0.0207-0.0318); corresponding mean differences were 0.0315 ± 0.0049 for mean class Dice (range, 0.0252-0.0375; hierarchical 95% CI, 0.0238-0.0394) and 0.0746 ± 0.0055 for meningioma recall (range, 0.0660-0.0792; hierarchical 95% CI, 0.0590-0.0911). The screening surrogate recovers only 5.8% of known same-patient slice pairs in an independent calibration dataset, so the screened condition may retain undetected related slices, and the intervals do not establish patient-level independence.Conclusion
Retaining detected test-linked near-duplicates was associated with higher slice-level benchmark performance in this replacement-matched exploratory analysis.
Related
- Comparative performance of nine large language models on professional Parkinson’s disease tasks: a multidimensional benchmark study
- Diagnostic and treatment recommendation performance of multimodal large language models in failed or painful total hip arthroplasty: an expert-rated benchmark study.
- Clinical Scores and Machine Learning at Matched Variable Domains in Acute Appendicitis: Performance of Recommended Rule-Out Thresholds in a CT-Verified Cohort
- Sport-related neurocognitive performance and repetitive head-impact exposure in elite Muay Thai athletes: a matched athletic-control study.
- AI-powered optical coherence tomography shows similar performance to intravascular ultrasound for plaque burden quantification: head-to-head benchmark study against histopathology of native coronary arteries.
BibTeX
@article{huang2026nearduplicate,
title = {{Near-duplicate contamination inflates brain MRI benchmark performance: a size-matched, multi-seed MedNeXt-B study.}},
author = {Yuhan Huang and Zhipeng Sun and Jinghui Chen and Tao Yang and Lianxin Xie and Xinchen Lin},
journal = {Frontiers in medicine},
year = {2026},
month = oct,
doi = {10.3389/fmed.2026.1955863},
url = {https://europepmc.org/article/MED/42923803},
note = {CC BY},
}
RIS
TY - JOUR
TI - Near-duplicate contamination inflates brain MRI benchmark performance: a size-matched, multi-seed MedNeXt-B study.
AU - Yuhan Huang
AU - Zhipeng Sun
AU - Jinghui Chen
AU - Tao Yang
AU - Lianxin Xie
AU - Xinchen Lin
PY - 2026
DA - 2026/10/08
JO - Frontiers in medicine
DO - 10.3389/fmed.2026.1955863
UR - https://europepmc.org/article/MED/42923803
AB - Background Pre-partitioned brain tumor MRI collections may place exact or near-duplicate slices on both sides of a benchmark split, but the magnitude and uncertainty of the resulting performance difference are rarely measured.Methods A unified MedNeXt-B 2D model produced a four-class label and a tumor mask from a shared encoder. For each near-duplicate-screened fold, we constructed a replacement-based contaminated condition of identical size and tumor-class-by-plane composition; this condition was not the original unscreened training pool. Validation folds and the frozen 995-slice test set were identical. Constituent source-dataset matching was not possible because BRISC2025 provides no per-image source provenance. Both conditions were trained with seeds 42, 1,337, 2,024, 3,407, and 5,090. Primary uncertainty summaries used a 10,000-resample hierarchical bootstrap over seeds and inferred pseudo-patient clusters.Results Across five seeds, screened accuracy was 0.9652 ± 0.0023 and mean class Dice was 0.7894 ± 0.0031 (mean ± sample SD). Accuracy was higher in the replacement-based contaminated condition by a mean difference of 0.0261 ± 0.0031 (range 0.0211-0.0281; hierarchical 95% CI, 0.0207-0.0318); corresponding mean differences were 0.0315 ± 0.0049 for mean class Dice (range, 0.0252-0.0375; hierarchical 95% CI, 0.0238-0.0394) and 0.0746 ± 0.0055 for meningioma recall (range, 0.0660-0.0792; hierarchical 95% CI, 0.0590-0.0911). The screening surrogate recovers only 5.8% of known same-patient slice pairs in an independent calibration dataset, so the screened condition may retain undetected related slices, and the intervals do not establish patient-level independence.Conclusion Retaining detected test-linked near-duplicates was associated with higher slice-level benchmark performance in this replacement-matched exploratory analysis.
SN - 2296-858X
ER -