TOWARDS EFFICIENT MEDICAL IMAGE SEGMENTATION: OVERCOMING DATA AND ANNOTATION LIMITATIONS

PhD Thesis Proposal Defence


Title: "TOWARDS EFFICIENT MEDICAL IMAGE SEGMENTATION: OVERCOMING DATA AND 
ANNOTATION LIMITATIONS"

by

Mr. Yi LIN


Abstract:

The integration of computer-aided medical image analysis systems into clinical 
practice is pivotal for tasks such as disease diagnosis, treatment planning, 
and prognosis prediction. Deep learning-based methods have emerged as a 
promising approach for medical image analysis, showcasing significant 
potential. However, the performance of these deep learning models heavily 
relies on the quality and quantity of the available training data. In practical 
scenarios, obtaining high-quality medical images can be both expensive and 
time-consuming, while expert annotation of these images is labor-intensive and 
prone to errors. Therefore, it is imperative to develop effective and efficient 
methods for medical image analysis that can achieve satisfactory performance 
even with limited data and annotations, while also being computationally 
effective in their development and deployment. In this thesis, our focus 
centers around three crucial research topics within the realm of effective and 
efficient medical image analysis: data-efficient learning, label-efficient 
learning, and model-efficient learning. In data-efficient learning, we delve 
into techniques that increase the diversity of the dataset through data 
augmentation, and explore efficient strate-gies for optimal utilization of the 
limited available data during the training process. In label-efficient 
learning, we investigate methods to reduce the annotation cost by leveraging 
partial labels, and study ways to effectively incorporate these partial labels 
into the training process to maximize their utility. In model-efficient 
learning, we undertake the design and examination of lightweight and compact 
models specifically tailored for medical image analysis tasks. These models aim 
to strike a balance between computational efficiency and performance. To 
validate the effectiveness and efficiency of the proposed methods, we conduct 
extensive experiments across various medical image analysis tasks, including 
nuclei segmentation, organ segmentation, and dose prediction. Through our 
research, we aim to contribute to the development of effective and efficient 
medical image analysis methods, enabling accurate and practical solutions in 
clinical settings. By addressing the challenges of limited data and 
annotations, and optimizing the computational aspects of model development, we 
strive to make significant advancements in the field of medical image analysis.


Date:                   Wednesday, 14 August 2024

Time:                   10:00am - 12:00noon

Venue:                  Room 3494
                        Lifts 25/26

Committee Members:      Dr. Hao Chen (Supervisor)
                        Dr. Dan Xu (Chairperson)
                        Dr. Long Chen
                        Prof. Tim Cheng