henhao
Sign in
BioinformaticsbioRxiv
Preprint

Adversarial Deep Structural Networks for Mammographic Mass Segmentation

Wentao ZhuiD, Xiaohui S. XieiD

PDFDOI
Post to X
Threads on XPost to BlueskyThreads on Bluesky
Structured summary available · Jump to summary ↓

Abstract

Mass segmentation is an important task in mammogram analysis, providing effective morphological features and regions of interest (ROI) for mass detection and classification. Inspired by the success of using deep convolutional features for natural image analysis and conditional random fields (CRF) for structural learning, we propose an end-to-end network for mammographic mass segmentation. The network employs a fully convolutional network (FCN) to model potential function, followed by a CRF to perform structural learning. Because the mass distribution varies greatly with pixel position, the FCN is combined with position priori for the task. Due to the small size of mammogram datasets, we use adversarial training to control over-fitting. Four models with different convolutional kernels are further fused to improve the segmentation results. Experimental results on two public datasets, INbreast and DDSM-BCRP, show that our end-to-end network combined with adversarial training achieves the-state-of-the-art results.

02 /Specification & Classification

Click any tag to browse matching papers
Citations
68
OpenAlex Topics

Abstract summary

Verified abstract synthesis

Claude Opus 5.5 · xhigh effort · 2 Nov 2026 · from the abstract · CC BY-NC 4.0

An end-to-end fully convolutional network with a positional prior, conditional random field structural learning, and adversarial training segments mammographic masses on INbreast and DDSM-BCRP.

Mammographic mass segmentation extracts morphological features and regions of interest for mass detection and classification, but small dataset sizes and position-dependent mass distributions complicate model training. Combining a position-aware fully convolutional network, a conditional random field, and adversarial training addresses overfitting and structural consistency in an end-to-end architecture.

Key results

  1. ClaimedAn ensemble of four end-to-end fully convolutional networks with conditional random fields and adversarial training segments mammographic masses on two public benchmarks.

    Evaluated on the public INbreast and DDSM-BCRP datasets; specific segmentation metrics and numerical baseline comparisons are not reported in the abstract. · Abstract

Methods, data and reproducibility

An end-to-end architecture combines a fully convolutional network (FCN) incorporating a pixel-position prior to model potential functions with a conditional random field (CRF) for structural learning, trained with an adversarial loss to mitigate overfitting and fused across four models with distinct convolutional kernels.

  • INbreast public mammogram dataset
  • DDSM-BCRP public mammogram dataset
Methods & tools
fully convolutional network (FCN), conditional random field (CRF)
Reproducibility gaps
Abstract only; no code URL, dataset links, or quantitative evaluation metrics are provided in the abstract.
Limitations (2)
  • Mammogram datasets used for training have small sample sizes, which the authors address through adversarial training. · stated by the authors
  • The abstract reports no quantitative segmentation scores, baseline methods, or dataset sample sizes. · noted in this summary

Who should read:Read this if you develop deep learning models for medical image segmentation on small imaging cohorts.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISAbdelhafiz '20Zhu '16Hatamizadeh '19Banerjee '24Rakhlin '18Hatamizadeh '19
69% semantic matchBioinformatics
Dina Abdelhafiz, Jinbo Bi, Reda A.A. Ammar+2 more·BMC Bioinformatics·Dec 1, 2020· 67 citations