**Deep Neural Network for Retinal Disease**
Mitha Ann Philip, Paul Rad
*Open Cloud Institute, University of Texas at San Antonio, San Antonio, Texas, USA*
{jrb468, Paul.Rad}@utsa.edu
**Project Definition:** The retinal blood vessels in our eye are good references for evaluating / monitoring human health. For correct and early diagnose of the disease, accurate segmentation of the retinal vessel is required to facilitate detection of various pathological modifications. Retinal imaging requires a robust technique that can accurately segment down the entire retinal blood vessel information even in varied imaging conditions such as lower contrast, non-uniform illumination and images with significant changes on the retinal image.
A tensorflow implementation of deep learning neural network based vessel segmentation method could be a robust technique towards identifying the retinal blood vessels from such retinal images efficiently. In this automatic retinal vessel extraction technique, the first approach could be an image enhancement step that could prepare the input for feature extraction / detection purpose. This can then be followed by directing the input to a convolutional neural network to extract the retinal vessel information and use it for disease classification.
**Outcome:** Segmented Retinal fundus image and disease classification for diabetic retinopathy.
**Dataset:** The image data can be found in the following links:
• DRIVE Database: http://www.isi.uu.nl/Research/Databases/DRIVE/.
• STARE Databse: http://cecas.clemson.edu/~ahoover/stare/
• CHASE Database: https://blogs.kingston.ac.uk/retinal/chasedb1/
</span>
</div>
[1]: Automated characterization of blood vessels as arteries and veins in retinal imagesv