Citation

Bannah, Hasanul and Ahmad Fauzi, Mohammad Faizal and Mansor, Sarina and Nabi, Md Serajun and Ur Rehman, Zaka and Looi, Lai Meng (2025) Breast cancer nuclei segmentation in ER-IHC images using deep Learning. In: 2024 Multimedia University Engineering Conference (MECON), 23-25 July 2024, Cyberjaya, Malaysia.

Text IEEE Xplore Full-Text PDF_32.pdf - Published Version Restricted to Repository staff only Download (2MB)

Abstract

Breast cancer is the leading cause of cancer deaths among women, accounted for more than 2.3 million new cases and 666,103 deaths in 2022. [1] Early detection is critical in improving survival, but traditional manual biopsy review is time consuming and expensive, especially in the early stages. But digital pathology plays a crucial role in automatically detecting breast cancer. In this work, deep learning-based techniques for automatically detecting breast cancer nuclei are thoroughly examined. Treatment can be guided by the expression of the estrogen receptor (ER), which is the most potent prognostic indicator in the diagnosis and prognosis of breast cancer. 220 Regions of Interest (ROIs) from 44 Whole Slide Images (WSIs) are included in our collection; each ROI has pixel-level ground truth annotations. We use a hybrid technique that takes advantage of the complementary strengths of two models: a customized U-Net and Cellpose. The custom U-Net is utilized to accurately segment irregularly shaped overlapping nuclei, and Cellpose has wide use across cellular morphologies. Using the two models within an ensemble model improves segmentation performance and generalizability over a wide range of tissue architectures. Quantitative analysis confirms the effectiveness of our methodology, achieved through a mean F1 score of 0.7468 and a mean intersection over union (IoU) of 0.6018 for the test set. The most compelling aspect was that the highest-scoring image (Test Image 42) achieved an F1 score of 0.8761 and IoU of 0.7795, demonstrating the model’s potential to achieve high-precision segmentation. These results confirm the ability of hybrid deep learning approaches to revolutionize AI-driven digital pathology for breast cancer diagnosis and treatment planning.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Deep Learning, breast cancer,
Subjects:	R Medicine > RC Internal medicine > RC0254 Neoplasms. Tumors. Oncology (including Cancer)
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	17 Mar 2026 07:12
Last Modified:	17 Mar 2026 07:53
URII:	http://shdl.mmu.edu.my/id/eprint/15522

Downloads

Edit (login required)