A Controllable Semi-automated Breast Cytology Images Segmentation

Conference

Haddag A., Bénamrane N.
University of Sciences and Technology of Oran - Mohamed Boudiaf, 2023 Dec

Cite

APA Click to copy
A., H., & N., B. (2023). A Controllable Semi-automated Breast Cytology Images Segmentation. University of Sciences and Technology of Oran - Mohamed Boudiaf.

Chicago/Turabian Click to copy
A., Haddag, and Bénamrane N. “A Controllable Semi-Automated Breast Cytology Images Segmentation.” University of Sciences and Technology of Oran - Mohamed Boudiaf, 2023.

MLA Click to copy
A., Haddag, and Bénamrane N. A Controllable Semi-Automated Breast Cytology Images Segmentation. University of Sciences and Technology of Oran - Mohamed Boudiaf, 2023.

BibTeX Click to copy

@conference{haddag2023a,
  title = {A Controllable Semi-automated Breast Cytology Images Segmentation},
  year = {2023},
  month = dec,
  institution = {University of Sciences and Technology of Oran - Mohamed Boudiaf},
  author = {A., Haddag and N., Bénamrane},
  howpublished = {Preprint},
  month_numeric = {12}
}

ℹ️

This paper is still in preprint phase. I can't share any further information due to the respect of the confidentiality imposed by the laboratory and other authors.

Abstract

Breast cancer remains a prevalent global health concern predominantly affecting women. Its detection often relies on different medical modalities such as cytology, which analyzes cellular morphology at a microscopic level. However, segmentation of cells within cytology slides is hindered by challenges such as low contrast, fuzzy borders, and overlapping cells. Current trends in research emphasize the development of automated end-to-end solutions based on deep learning methods, yet these approaches lack explainability by design. In this paper, we propose a novel approach that revisits classic segmentation techniques - like color spaces separation, watershed, regions selection, etc. that offer physicians and cytopathology experts an open-box tool for cell segmentation. Our method provides a controllable semi- automated process, ensuring control in diagnosis, unlike prevailing black-box deep learning models.