Tumor Budding Detection in H&E-Stained Images Using Deep Semantic Learning

Citation

Banaeeyan, Rasoul and Ahmad Fauzi, Mohammad Faizal and Wei, Chen and Knight, Debbie and Hampel, Heather and Frankel, Wendy L. and Gurcan, Metin N. (2020) Tumor Budding Detection in H&E-Stained Images Using Deep Semantic Learning. In: IEEE Region 10 Conference (TENCON) 2020, 16-11-2020, Virtual, Osaka, Japan.

[img] Text
126.pdf - Published Version
Restricted to Repository staff only

Download (477kB)

Abstract

Tumor buddings (TB), a special formation of cancerous cells that bud from the tumor front, are fast becoming the key indicator in modern clinical applications where they play a significant role in prognostic and evaluation of colorectal cancers in histopathological images. Recently, computational methods have been rapidly evolving in the domain of digital pathology, yet the literature lacks computerized approaches to automate the localization and segmentation of TBs in hematoxylin and eosin (H&E)-stained images. This research addresses this very challenging task of tumor budding detection in H&E images by presenting different deep learning architectures designed for semantic segmentation. The proposed design for a new Convolutional Neural Network (CNN) incorporates convolution filters with different factors of dilations. Multiple experiments based on a newly constructed colorectal cancer histopathological image collection provided promising performances. The best average intersection over union (IOU) for TB of 0.11, IOU for non-TB of 0.86, mean IOU of 0.49 and weighted IOU of 0.83 were observed.

Item Type: Conference or Workshop Item (Paper)
Uncontrolled Keywords: Tumor Budding Detection, Deep Learning, Colorectral Cancer, Digital Pathology
Subjects: R Medicine > RC Internal medicine > RC0254 Neoplasms. Tumors. Oncology (including Cancer)
Divisions: Faculty of Engineering (FOE)
Depositing User: Ms Rosnani Abd Wahab
Date Deposited: 12 Oct 2021 05:23
Last Modified: 26 Oct 2021 04:03
URII: http://shdl.mmu.edu.my/id/eprint/8537

Downloads

Downloads per month over past year

View ItemEdit (login required)