This repository contains the implementation of convolutional neural network for optic disk and cup segmentation from given fundus images
Images were cropped to nearest square size and re-sized to a dimension of (512, 512). The different lighting conditions and intensity variations among images across various databases were circumvented by perform-ing normalization of the histogram using Contrast Limited Adaptive HistogramEqualization (CLAHE). 2 different images were generated by varying parameters such as clip value & window level while performing CLAHE. Along with CLAHE, spatial co-ordinates information were also provided to thenetwork. This additional information aided in learning relative features (i.e. disklocation with respect to fovea)
57 layered deep network was used for segmentation of optic disk and cup. Network architecture is illustrated in figure below...
Mask generation used for reducing false positives predicted by network...
Image on left shows raw data and image on left shows model predictions...
git clone https://github.com/koriavinash1/Optic-Disk-Cup-Segmentation.git
cd Optic-Disk-Cup-Segmentation
pip install -r requirements.txt
./src consists all source codes
./src/segmentation code for all segmentation work
./src/classification code for glaucoma screening : TODO
Tune parameters and run Main.py for executing task
If any comments or issues, pull requests/issues are Welcomed....
Thankyou
- Avinash Kori (avinashgkori@smail.iitm.ac.in)