Email: zilinwan @ umich . edu
I am a Master's student in CSE at the University of Michigan, Ann Arbor. I am broadly interested in computer vision and deep learning, particularly training models with limited/no supervision. In the meantime, I am excited to find out how state-of-the-art algorithms can help doctors gain insights into fatal diseases.
I am currently working with Prof. Stella Yu on unsupervised object discovery. I am also a research assistant at C. Galban Lab, Michigan Medicine, where I work on pulmonary 3D CT scan segmentation using deep learning. Before that I was a summer intern at Genentech, working on OCT-A image segmentation.
I previously graduated Summa Cum Laude from the Ohio State University majoring in CSE.
- EECS598 - Action and Perception: Graduate Student Instructor, University of Michigan, Winter 2023.
- SI670 - Applied Machine Learning: Instructional Aide, University of Michigan, Fall 2021.
- CSE3521/5521 - Introduction to Artificial Intelligence: Grader, the Ohio State University, Spring 2020.
Pulmonary Artery-Vein Segmentation in 3D Computed Tomography Images
We propose iSparseUnet, a 3D Unet-like semantic segmentation model that is highly efficient when the positive class masks are sparse. We benchmark the model on A/V segmentation, which demonstrated that our model only need 0.7 seconds and 16 GB of GPU memory to segment a CT scan of ~34 million voxels on a NVIDIA A40 while achieving SOTA performance.
Machine Learning Based FAZ Segmentation in OCTA images
We propose a multi-task learning algorithm for foveal avascular zone(FAZ) segmentation that learns to segment FAZs, FAZ boundaries, and vessels at the same time. We demonstrate that the model could pick up meaningful domain knowledge and achieve near perfect segmentation performance.