Exudation-of-Eye

Using UNet for detection of Exudation. Medical Imaging

Prerequisites

• Python
• Keras (with tensorflow-gpu preffered)
• Opencv,Numpy,Pandas

Test and Train dataset

Fundus images of Retina. Train - 191 Test - 25 Preprocessing includes finding contours/radius of the eyeballs and normalizing the dataset to that radius.
Augmentation on the dataset includes rotation, gamma, colour shift and contrast.

CNN Model and Performance

Used UNet for detecting exudations in fundus images. Performance of dice coefficient close to 0.45-0.6 respectively for training and testing dataset and the accuracy on the test data is close to 94% on the testing data[binary classification accuracy of existence of exudation in the retina images].