Retinal Vessel Segmentation with Deep Learning Techniques

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Retinal Vessel Segmentation with Deep Learning Techniques"

By

Mr. Yishuo ZHANG


Abstract

The retinal fundus vasculature is the only vascular system that can be 
viewed in a non-invasive manner and morphology changes in micro-vascular 
provide pathological features and diagnostic information, which are 
commonly observed and studied in the ophthalmology examination. There is 
an urgent need for automatic retinal vessel segmentation methods from 
retinal images, nevertheless, the task is challenging. The main obstacles 
to vessel segmentation are poor image quality, tiny structures of vessels, 
and various abnormalities, which pose difficulties for detecting vessels. 
With the emergence of deep learning techniques, promising progress has 
been witnessed in the field of retinal vessel segmentation, although the 
precise segmentation for vessels especially micro-vascular remains 
unsolved. Concretely, Existing methods provide unsatisfactory performance 
on micro-vascular; CNNs based methods fail to maintain the graphical 
structure of vessels; deep learning based methods are highly dependent on 
perfect annotations which are expensive to acquire.

In this study, we further explore deep learning techniques for retinal 
vessel segmentation with a focus on several issues in the task. First, we 
propose a U-Net based method with deep supervision to preserve spatial 
information in deep layers. We specifically design a vessel-dependent loss 
to deal with different errors in cases of thick vessels and thin vessels. 
The proposed method provides better performance on vessel segmentation. 
Second, we design a novel method for exploring graphical structures of 
vessels by graph convolutional networks, which further improved the 
performance of vessel segmentation. The proposed method produces vessels 
with improved graphical structures. Third, we proposed a Transformer-U-Net 
hybrid model which improves segmentation results by a global receptive 
field. We also mitigate an imbalance issue between thin vessels and thick 
vessels by a skeleton-aware decoder, in which the model learns the 
skeleton of vessels as an auxiliary task. The proposed method has been 
evaluated in different imaging modalities and achieve promising 
performances. Fourth, we present a robust learning scheme under the 
condition of wrong annotations. A self-attention module with a label 
adaptation scheme increases the robustness toward label noise and enables 
learning with imperfect annotations. In conclusion, this study will not 
only provide cutting-edge techniques for retinal vessel segmentation but 
also offer insights into the segmentation of tiny tubular structures.


Date:			Monday, 16 May 2022

Time:			2:00pm - 4:00pm

Zoom Meeting: 
https://hkust.zoom.us/j/91671008002?pwd=LzZrRjRwUFJhaWpxLzFSZFpFNld5QT09

Chairperson:		Prof. Jianan QU (ECE)

Committee Members:	Prof. Albert CHUNG (Supervisor)
 			Prof. Chi Keung TANG (Supervisor)
 			Prof. Hao CHEN
 			Prof. Huamin QU
 			Prof. Tsz Wai WONG (CBE)
 			Prof. Pheng Ann HENG (CUHK)


**** ALL are Welcome ****