The computer-aided analysis is poised to play an increasingly prominent role in medicine and healthcare. Benefiting from the increasing computing power, various machine learning frameworks have been developed in the biomedical field, bringing significant improvements in real-world clinical applications. However, for many diseases, the development of these life-supporting algorithms is still in its infancy. To bridge this gap:
This thesis is dedicated to the development of efficient algorithms for better image intervention and addressing data quality challenges in machine learning algorithms to provide direct guidance for real-world clinical applications.
With the above goals, three topics are explored in depth. First, we develop a novel tissue analysis framework for cardiac substrate identification and tissue heterogeneity assessment. In particular, we creatively used model uncertainty to measure tissue structure information, offering a means of extracting the tissue heterogeneity information in a non-invasive way for real-time imaging and processing.
The tissue analysis framework in the first aim is based on the fully supervised technique, which relies heavily on the availability of large-scale datasets with accurate annotations. Such high-quality datasets are extremely time-consuming to acquire, especially for biomedical segmentation tasks. To lessen the need for the labeling process, we further develop three weakly supervised learning frameworks to address data and labeling challenges caused by limited data resources.
Finally, we develop an in-vivo tissue analysis framework on cardiac datasets, aiming to provide real-time guidance for clinical ablation procedures. Our models could contribute to the improvement of ablation treatment by identifying the ablation targets and avoiding critical structures within the hearts.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/wzcd-fx21 |
| Date | January 2023 |
| Creators | Huang, Ziyi |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.002 seconds