The Library
Topology and attention in computational pathology
Tools
Qaiser, Talha (2019) Topology and attention in computational pathology. PhD thesis, University of Warwick.
|
PDF
WRAP_Theses_Qaiser_2019.pdf - Submitted Version - Requires a PDF viewer. Download (76Mb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b3489183~S15
Abstract
Histopathology serves as the gold standard in the process of cancer diagnosis and unravelling the disease heterogeneity. In routine practice, a trained histopathologist performs visual examination of tissue glass slides under the microscope. The objective of the visual examination is to observe the morphological appearance of tissue sections, analyse the density of tumour rich areas, spatial arrangement, and architecture of diferent types of cells. However, careful visual examination of tissue slides is a demanding task especially when workloads are high, and the subjective nature of the histological grading inevitably leads to inter- and even intra-observer variability. Attaining high accuracy and objective quantification of tissue specimens in cancer diagnosis are some of the ongoing challenges in modern histopathology. With the recent advent of digital pathology, tissue glass slides can now be scanned with digital slides scanners to produce whole slide images (WSIs). A WSI contains a high-resolution pixel representation of tissue slide, stored in a pyramidal structure and typically containing 1010 pixels. Automated algorithms are generally based on the concepts of digital image analysis which can analyse WSIs to improve the precision and reproducibility in cancer diagnostics. The reliability of the results of an algorithm can be objectively measured and improved against an objective standard.
In this thesis, we focus on developing automated methods for quantitative assessment of histology WSIs with the aim of improving the precision and reproducibility of cancer diagnosis. More specifically, the designed automated computational pathology algorithms are based on deep learning models in conjunction with algebraic topology and visual attention mechanisms. To the best of our knowledge, the applicability of attention and topology based methods have not been explored in the domain of computational pathology. In this regard, we propose an algorithm for computing persistent homology profiles (topological features) and propose two variants for effective and reliable tumour segmentation of colorectal cancer WSIs. We show that incorporation of deep features along with topological features improves the overall performance for tumour segmentation.
We then present the first-ever systematic study (contest) for scoring the human epidermal growth factor receptor 2 (HER2) biomarker on breast cancer histology WSIs. Further, we devise a reinforcement learning based attention mechanism for HER2 scoring that sequentially identifies and analyses the diagnostically relevant regions within a given image, mimicking the histopathologist who would not usually analyse every part of the slide at the highest magnification. We demonstrate the proposed model outperforms other methods participated in our systematic study, most of them were using state-of-the-art deep convolutional networks. Finally, we propose a multi-task learning framework for simultaneous cell detection and classifi- cation, which we named as Hydra-Net. We then compute an image based biomarker which we refer as digital proximity signature (DPS), to predict overall survival in diffuse large B-cell lymphoma (DLBCL) patients. Our results suggest that patients with high collagen-tumour proximity are likely to experience better overall survival.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software R Medicine > RB Pathology R Medicine > RC Internal medicine > RC0254 Neoplasms. Tumors. Oncology (including Cancer) |
||||
Library of Congress Subject Headings (LCSH): | Histology, Pathological, Cancer -- Histopathology, Diagnostic imaging -- Digital techniques, Topology | ||||
Official Date: | April 2019 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | Department of Computer Science | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Supervisor(s)/Advisor: | Rajpoot, Nasir M. (Nasir Mahmood) | ||||
Format of File: | |||||
Extent: | xvii, 141 leaves : illustrations, charts | ||||
Language: | eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |
Downloads
Downloads per month over past year