Past Events

PhD Defense

Instance Segmentation for Biological Image Analysis

 

Download as iCal file

Friday, March 12, 2021, 01:00pm - 03:00pm

 

Speaker: Jingru Yi

Location : Remote via Webex

Committee

Prof. Dimitris N. Metaxas (Advisor)

Prof. Konstantinos Michmizos

Prof. Sungjin Ahn

Prof. Huang, Sharon Xiaolei (Penn State)

Event Type: PhD Defense

Abstract: Image-based instance segmentation is a task that differentiates and classifies objects at the pixel level. This task is of great significance for many biological applications (e.g., the study of neural cell interactions), which require precise segmentation of biological objects. However, it is challenging to capture fine-grained object details in biological images and separate the attached or overlapping objects due to their weak boundaries. This thesis presents a series of instance segmentation methods for biological images that address these challenges. We focus on three representative biological subjects, including neural cells, plants, and cell nuclei, to illustrate the image-based challenges and our methods' effectiveness. In particular, neural cells feature irregular shapes with tiny and slender protrusions; plant leaves typically contain thin stalks and overlapping regions; cell nuclei are usually distributed as clusters and have various shapes. In the first chapter, we present an overview of existing instance segmentation methods and their applications in biological image analysis. In the following chapters, we introduce a series of instance segmentation methods that distinguish objects from a global view and classify object pixels within bounding boxes. We study two kinds of object detectors: anchor-based and keypoint-based detectors. We explore strategies to remove undesired neighboring objects within a region of interest (ROI). On this basis, we further enhance the segmentation quality around the uncertain areas (e.g., boundary) via an auxiliary point-wise feature refinement module. Through extensive experiments, we show the superiority of our methods by comparing them to state-of-the-art approaches.

 

Meeting link: https://rutgers.webex.com/rutgers/j.php?MTID=m0cb3fd10ea2f62afe2d3aafc8a7c100d

Meeting number: 120 120 1689
Password: 4paSnRChZ35

Join by video system
Dial This email address is being protected from spambots. You need JavaScript enabled to view it.
You can also dial 173.243.2.68 and enter your meeting number.

Join by phone
+1-650-429-3300 USA Toll
Access code: 120 120 1689