More about HKUST
Data-efficient Deep Learning for Precise Medical Image Analysis
Speaker: Dr. Xi Wang The Chinese University of Hong Kong Title: "Data-efficient Deep Learning for Precise Medical Image Analysis" Date: Monday; 23 Dec 2024 Time: 3:30pm - 4:30pm Join Zoom Meeting https://hkust.zoom.us/j/93624265291?pwd=K21sNWNFenFYZjlTWDRPMTRwMTY5dz09 Meeting ID: 936 2426 5291 Passcode: 456088 Abstract: Automated medical image analysis holds great significance in clinical practice, as it can considerably improve diagnostic efficiency and accuracy. However, annotating medical images is prohibitively expensive, as it is often highly demanding, time-consuming, and labor-intensive--particularly for large-sized or high-dimensional medical images. As a result, the annotations in collected datasets are often imperfect, meaning that images are either coarsely labeled or partially labeled. In this talk, I will share our work on weakly and semi-supervised deep learning for the analysis of large-sized or high-dimensional medical images. First, I will discuss several weakly-supervised deep learning methods that use coarse labels for whole-slide image and optical coherence tomography image analysis. Next, I will focus on semi-supervised deep learning techniques that leverage large amounts of unlabeled data to enhance fully supervised models, with applications in disease classification and organ segmentation. The class imbalance issue is also addressed. All of the aforementioned approaches have been extensively evaluated on both in-house and public datasets, achieving state-of-the-art performance. Finally, I will conclude by outlining future work, which aims to extend these studies into areas such as multi-modal learning, longitudinal learning, and foundation model analysis. ***************** Biograhpy: Dr. Xi Wang received her Ph.D. degree in Computer Science and Engineering (CSE) from The Chinese University of Hong Kong (CUHK) in 2020. After that, she worked as a post-doctoral fellow at Stanford University and CUHK. Her research interests are at the intersection of computer science, medical science, and physics, with special emphasis on deep learning, medical image analysis, and complex systems, strongly driven by the goal of improving human health. Over the past decade, She has been dedicated to developing novel deep learning approaches to streamline the workflow of clinical diagnosis, treatment planning, and prognosis for patients with chronic diseases, cancer, infectious diseases, and neurological diseases with high accuracy and efficiency at the lowest cost. She is currently working on longitudinal learning based on time series data for disease progression forecasting and computational methodology related to single-cell and spatial genomics.