Abstract
High-quality mammography screening depends on correct positioning,
yet existing AI systems often rely on black-box classifiers with limited cross-device generalizability. We propose a measurement-based, explainable assessment grounded in the Posterior Nipple Line (PNL) criterion, casting positioning as YOLO keypoint regression rather than classification. We evaluate a complete 2 × 3 factorial of two backbones (YOLOv8-L, YOLOv26-L) and three preprocessing strategies—no-op, global histogram equalization, and non-local means (NLM) denoising with wavelet edge boost and Contrast Limited Adaptive Histogram Equalization (CLAHE)—on VinDr-Mammo (n = 1,463 paired studies) and the external Chinese Mammography Database (CMMD, n = 95). Inclusion requires radiologist-labeled adequate MLO positioning, whereas the evaluation target is PNL consistency from expert keypoints (ΔD ≤ 10 mm), giving both consistent and inconsistent pairs (internal test: 117/29). The recommended NLM+Wav+CLAHE / YOLOv26-L configuration attains the lowest internal PNL error (2.52 mm), Bland–Altman 95% limits of agreement within ±10mm ([−6.60, +7.45] mm), and the strongest external agreement (unbuffered rule: accuracy 0.862, sensitivity 0.877, specificity 0.828), outperforming our reimplemented previously published classifier. In our runs advanced preprocessing helped YOLOv26-L internally but not YOLOv8-L, and the ordering reversed externally.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/Deep_Brea3th_009.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=dYORbOQz7i
BibTex
@InProceedings{ErkEne_Landmark_MICCAISAT2026,
author = { Erkan, Enes Can AND Denizoğlu, Nurper AND Seker, Mustafa Ege AND Meltem, Emine AND Aktufan Cerekci, Esma AND Şanal, Yaren Nas AND Alis, Deniz AND Aribal, Erkin AND Oksuz, Ilkay},
title = { { Landmark Detection for Automated Mammography Positioning Quality Assessment via Posterior Nipple Line Consistency } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}