A Survey on Deep Domain Adaptation in Medical Image Segmentation

PhD Qualifying Examination


Title: "A Survey on Deep Domain Adaptation in Medical Image Segmentation"

by

Mr. Zengqiang YAN


Abstract:

Despite the advancement of deep learning in automatic medical image 
segmentation, performance would significantly degrade when applying the trained 
deep models to new data acquired from different machines or institutions than 
the training data. One straightforward way is to generate manual annotations 
for every new target domain and train new models, which is costly and 
time-consuming. The variations in multi-center data in medical image analysis 
have brought the necessity of domain adaptation.

The aim of this survey is to give an overview of deep domain adaptation with a 
specific task on medical image segmentation. After a general motivation, we 
first formulate the domain adaptation and the transfer learning problems. 
Second, we overview the current deep domain adaptation frameworks for medical 
image segmentation. More specifically, the current deep domain adaptation 
algorithms are classified into three categories namely supervised, 
semi-supervised and unsupervised methods, based on the availability of manual 
annotations of the target domain. Finally, we discuss several potential 
research directions in deep domain adaptation for medical image segmentation. 
Different from the current deep domain adaptation frameworks for medical image 
segmentation, we propose a weakly supervised domain adaptation framework by 
introducing additional constraints.

Keywords: Domain Adaptation, Adversarial Learning, Medical Image Segmentation


Date:			Monday, 10 September 2018

Time:                  	4:00pm - 6:00pm

Venue:                  Room 4475
                         Lifts 25/26

Committee Members:	Prof. Tim Cheng (Supervisor)
 			Prof. Albert Chung (Chairperson)
 			Dr. Pedro Sander
 			Prof. Weichuan Yu (ECE)


**** ALL are Welcome ****