Deep Learning-Based Metastatic Detection and Localization for Breast Cancer Diagnosis in Whole Slide Histopathology Images

Khaliliboroujeni, Sepideh

Deep Learning-Based Metastatic Detection and Localization for Breast Cancer Diagnosis in Whole Slide Histopathology Images

Khaliliboroujeni, Sepideh

Permalink

Publication Type:: Thesis
Issue Date:: 2026

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (15.47 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Khaliliboroujeni, Sepideh
dc.date.accessioned	2026-11-04T02:12:40Z
dc.date.available	2026-11-04T02:12:40Z
dc.date.issued	2026
dc.identifier.uri	http://hdl.handle.net/10453/197297
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Breast cancer is one of the most prevalent and lethal diseases among women worldwide, and early and accurate detection, particularly of metastasis, is crucial to improving patient survival rates. Deep learning techniques have revolutionized image analysis, yet their application to histopathology Whole Slide Images (WSIs) remains a challenge due to the massive size, complex tissue structures, and limited annotations of WSIs. Although advances such as patch-based classification, multi-scale modeling, and weakly supervised learning have improved accuracy and reduced annotation needs, challenges remain. Metastasis detection and localization approaches still face high computational demands, reliance on extensive annotations, and difficulty capturing global context and fine details. Patch-based methods are widely used as they enable manageable processing of large WSIs by dividing them into smaller regions. However, they often struggle with boundary precision and detecting small metastases, while reliance on high-resolution WSIs increases costs and limits scalability in clinical practice. This thesis addresses the core research question of how deep learning models can be optimized for metastasis detection and localization in large-scale histopathology images while ensuring efficiency, reliability, and clinical deployability in weakly supervised settings. To address this question, this thesis proposes two novel, efficient, weakly supervised deep-learning approaches for the detection and localization of metastases in WSIs. The first approach, detailed in Chapter 3, focuses on maximal computational efficiency, achieving efficient inference for large-scale WSI analysis using low-resolution inputs and minimal annotated data, making it suitable for rapid clinical triage. The second, enhanced framework (Chapter 4) prioritizes spatial accuracy and improved data efficiency by addressing the limitations of coarse patch-based analysis. It improves contextual modeling and boundary-level precision through advanced architectural components, enabling more refined localization of metastatic regions under weak supervision. Both newly proposed methods reduce computational and annotation costs by requiring only a fraction of labeled data. Evaluated on the CAMELYON16 dataset, they outperform state-of-the-art methods and expert pathologists in metastasis detection and localization, demonstrating robustness in large-scale, noisy, and limited-data environments.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/197297/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2026 Sepideh Khaliliboroujeni
dc.rights	au.edu.uts.lib/cph
dc.title	Deep Learning-Based Metastatic Detection and Localization for Breast Cancer Diagnosis in Whole Slide Histopathology Images	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Breast cancer is one of the most prevalent and lethal diseases among women worldwide, and early and accurate detection, particularly of metastasis, is crucial to improving patient survival rates. Deep learning techniques have revolutionized image analysis, yet their application to histopathology Whole Slide Images (WSIs) remains a challenge due to the massive size, complex tissue structures, and limited annotations of WSIs. Although advances such as patch-based classification, multi-scale modeling, and weakly supervised learning have improved accuracy and reduced annotation needs, challenges remain. Metastasis detection and localization approaches still face high computational demands, reliance on extensive annotations, and difficulty capturing global context and fine details. Patch-based methods are widely used as they enable manageable processing of large WSIs by dividing them into smaller regions. However, they often struggle with boundary precision and detecting small metastases, while reliance on high-resolution WSIs increases costs and limits scalability in clinical practice. This thesis addresses the core research question of how deep learning models can be optimized for metastasis detection and localization in large-scale histopathology images while ensuring efficiency, reliability, and clinical deployability in weakly supervised settings. To address this question, this thesis proposes two novel, efficient, weakly supervised deep-learning approaches for the detection and localization of metastases in WSIs. The first approach, detailed in Chapter 3, focuses on maximal computational efficiency, achieving efficient inference for large-scale WSI analysis using low-resolution inputs and minimal annotated data, making it suitable for rapid clinical triage. The second, enhanced framework (Chapter 4) prioritizes spatial accuracy and improved data efficiency by addressing the limitations of coarse patch-based analysis. It improves contextual modeling and boundary-level precision through advanced architectural components, enabling more refined localization of metastatic regions under weak supervision. Both newly proposed methods reduce computational and annotation costs by requiring only a fraction of labeled data. Evaluated on the CAMELYON16 dataset, they outperform state-of-the-art methods and expert pathologists in metastasis detection and localization, demonstrating robustness in large-scale, noisy, and limited-data environments.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/197297