Oral Candidacy - Lin Yang
|Start:||4/26/2017 at 3:30PM|
|End:||4/26/2017 at 6:30PM|
|Location:||181 Fitzpatrick Hall|
Faculty and students are welcome to attend the presentation portion of the defense.
April 26, 2017 3:30 pm 181 Fitzpatrick
Adviser: Dr. Danny Chen
Dr. Sharon Hu Dr. Siyuan Zhang Dr. Walter Scheirer
New Approaches for Image Segmentation, Enhancement, and Analysis in 3D Brain Images
With the recent advances of optical tissue clearing technology, current imaging modalities are able to image whole mouse brain in 3D with single-cell resolution. This capability facilitates many exciting studies, such as researches on brain tumor metastasis and brain immune systems. In our preliminary research work, we have developed new methods to enhance, segment, and analyze 3D fluorescence microscopy images of mouse brain. By extracting quantitative information from such 3D images, many biological questions can be answered and many biological hypotheses can be generated and tested.
Due to the staining issue and the light-scattering issue, 3D images usually contain severe background noise. This background noise remains a significant obstacle for visualization and segmentation of these high-resolution 3D images. Thus, in the first step, We developed a new method that combines one-class learning and spatial filtering to remove background noise both accurately and fast.
To achieve quantitative analysis, we need to segment objects (e.g., cells, tumors, and blood vessels) in these images. Although deep learning based pixel-level segmentation has been well studied in computer vision area, 3D instance-level segmentation remains a challenge. Thus, in the second step, we developed a new method based on fully convolutional networks and k-terminal cut to achieve 3D instance-level cell segmentation.