Unsupervised Identification of Tissues in Zebrafish
Open Access
- Author:
- Shah, Harshil Jignesh
- Graduate Program:
- Electrical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 14, 2016
- Committee Members:
- Dr David Miller, Thesis Advisor/Co-Advisor
John F Doherty, Thesis Advisor/Co-Advisor - Keywords:
- Unsupervised
Image segmentation
Active contour
Level sets
Zebrafish - Abstract:
- This project is a part of the medical imaging research going on at Hershey Medical Center. Given an image of a zebrafish, the task is to classify the different tissues such as the notochord and spinal cord tissues of the zebrafish. Images of the zebrafish were obtained from Dr. Cheng at Hershey Medical Center. This is basically an image segmentation problem. We take an unsupervised approach to classify the tissues in the image. Active-contour models are used along with narrowband level sets to identify and segment out the notochord tissue from the image. The objective function of the contour is obtained from the active-contour models, which is minimized using level sets. Active-contour models also help in detecting the internal contours automatically. The aim is to experiment with different hyper parameters like the size of the mask, smoothening parameter and the number of iterations to achieve the best accuracy for tissue identification and segmentation. Accuracy of the proposed approach will be evaluated using True Positive Rate and the ratio of False Positive Count to the True Positive Count. This work will help speed up the ongoing genomics research at Hershey Medical Center.