Relaxation and Restriction for Medical Image Segmentation with Convolutional Neural Networks

PhD Thesis Proposal Defence


Title: "Relaxation and Restriction for Medical Image Segmentation with 
Convolutional Neural Networks"

by

Miss Pei WANG


Abstract:

Deep learning has achieved great success in different tasks for natural 
image data in recent years. However, the segmentation of medical image 
data appears to be challenging because of the limited manually labeled 
data from the medical experts, the pathological changes and the 
morphological variation of the target objects, and the random noise 
associate with the imaging systems. Therefore, the effectiveness of 
convolutional neural networks (CNN) cannot be fully achieved. In this 
thesis, we propose different approaches and techniques in two major 
categories, which are relaxation and restriction in CNN models to promote 
the performance in medical image segmentation.

For multi-modality and multi-class brain tumor segmentation on magnetic 
resonance images, the challenges lie in the severe data imbalance among 
different tumor sub-regions, and the great variety in terms of tumor 
location, size, shape, and appearance. Therefore, we relax the boundary 
constraints for the tumor sub-regions to better recognize the overall 
structure of the tumor. A novel loss function is proposed to enforce more 
attention on the harder classes automatically during training, and a 
symmetrical attention module is presented to restrict the possible tumor 
location. The experimental results on publicly available datasets from 
real patents validate the effectiveness of these proposed approaches.

Colon gland instance segmentation of histological images is a crucial step 
for colorectal cancer diagnosis, but accurate segmentation of extremely 
deformed glands in highly malignant cases or some rare benign cases 
remains to be challenging. Therefore, we relax the input domain to 
incorporate the predicted clinical text information for high-level feature 
guidance of the gland morphology with different histologic grades. Besides 
the initial segmentation, it offers histologic grade diagnosis and 
enhanced segmentation for full-scale assistance. In the other approach, 
the gland segmentation is conducted under the restriction of hierarchical 
semantic feature matching from histological pairs in an attentive process, 
where both spatial details and morphological appearances can be well 
preserved and balanced, especially for the glands with severe deformation. 
A loss function is also introduced to enforce simultaneous satisfaction of 
semantic correspondence and gland instance segmentation on the 
pixel-level. The models successfully boost the segmentation performances 
on greatly mutated or deformed cases, and outperform the state-of-the-art 
approaches on public datasets from real patients.


Date:			Thursday, 6 August 2020

Time:                  	2:30pm - 4:30pm

Zoom Meeting:		https://hkust.zoom.us/j/4089239113

Committee Members:	Prof. Albert Chung (Supervisor)
  			Prof. Chiew-Lan Tai (Chairperson)
 			Prof. Pedro Sander
 			Dr. Sai-Kit Yeung


**** ALL are Welcome ****