Citation

Bannah, Hasanul (2026) Breast cancer nuclei segmentation and classification in histopathology images using deep learning. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

The expression of progesterone receptors (PR) is an important biomarker in breast cancer that affects decisions about hormonal therapy. The Allred scoring method is used in modern clinical practice to measure progesterone receptors using immunohistochemistry (IHC). Pathologists manually analyze both the quantity and intensity of stained nuclei. This method requires substantial effort and may vary among observers or even within the same observer, especially in marginal circumstances. This thesis presents an automated framework for PR-IHC analysis, comprising three fundamental stages: nuclei segmentation, nuclei classification, and scoring. A deep learning model based on Cellpose, leveraging its spatial vector flow representation, was used to achieve accurate nuclear segmentation in intricate PR-IHC tissue contexts characterized by overlapping structures, indistinct borders, and diverse shapes. A color deconvolution approach was applied to isolate the DAB (3,3′-diaminobenzidine) chromogen, enabling accurate classification of nuclear staining intensity into four categories: Strong, Moderate, Weak, and Negative. The Allred score was then computed by combining the proportion of positively stained nuclei with their corresponding intensity levels. We used 250 high-resolution PR-IHC images that experts manually marked up to make sure the system was correct. The results of the experiment show that the segmentation worked well, with a Precision of 0.8882, a Recall of 0.8215, an F1 score of 0.8535, and an IoU of 0.7445. A Macro-F1 score of 0.9487 shows that the classification accuracy across the four staining categories is quite high. The automated Allred scores showed more than 90% agreement with pathologist evaluations at the slide level, ensuring that treatment recommendations were consistent. This research introduces a completely automated, nuclei-resolved Allred scoring system for PR-IHC whole-slide images, combining segmentation, classification, and scoring into a unified workflow. The proposed approach reduces observer variability, improves reproducibility, and offers a simple, clinically useful alternative for digital pathology workflows. This study shows that artificial intelligence can be a useful and reliable tool for predicting and planning treatment for breast cancer.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: Q325.73 .H37 2026
Uncontrolled Keywords:	Deep learning (Machine learning)
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	22 May 2026 04:50
Last Modified:	22 May 2026 04:50
URII:	http://shdl.mmu.edu.my/id/eprint/15899

Downloads

Edit (login required)