Segmentation of Blood Vessels, Optic Disc Localization, Detection of Exudates and Diabetic Retinopathy Diagnosis from Digital Fundus Images
Soham Basu, Sayantan Mukherjee, Ankit Bhattacharya, Anindya Sen
){kind=icon}
){kind=icon}
Abstract: Diabetic Retinopathy (DR) is a complication of long-standing, unchecked diabetes and one of the leading causes of blindness in the world. This paper focuses on improved and robust methods to extract some of the features of DR, viz. Blood Vessels and Exudates. Blood vessels are segmented using multiple morphological and thresholding operations. For the segmentation of exudates, k-means clustering and contour detection on the original images are used. Extensive noise reduction is performed to remove false positives from the vessel segmentation algorithm's results. The localization of Optic Disc using k-means clustering and template matching is also performed. Lastly, this paper presents a Deep Convolutional Neural Network (DCNN) model with 14 Convolutional Layers and 2 Fully Connected Layers, for the automatic, binary diagnosis of DR. The vessel segmentation, optic disc localization and DCNN achieve accuracies of 95.93%, 98.77% and 75.73% respectively.
 (a) |
 (b) |
 (c) |
 (d) |
 (e) |
 (f) |
 (g) |
Blood vessel segmentation algorithm
 (a) |
 (b) |
 (c) |
 (d) |
 (e) |
 (f) |
 (g) |
(a) Original image, (b) Grayscale of (a), (c) Result of k-means clustering, (d) Generated Template, (e) Template Matching result (using NCCOEFF; notice the OD region has highest similarity), (f) Marking OD and its center, (g) Masking OD region.
Optic Disc Localization algorithm
 (a) |
 (b) |
 (c) |
 (d) |
 (e) |
(a) Original Image, (b) K-means clustering result, (c) Extracting the exudates from (b) and thresholding, (d) Logical OR of (c) and the images containing the smallest exudates, (e) Final segmentation result after OD masking.
Exudates Detection algorithm
Proposed Neural Network architecture with the corresponding kernel/Filter sizes (k), number of feature maps (n) and strides (s) specified for each convolutional layer.
The DRIVE dataset was used for the evaluation of the proposed method.
The segmentation result for the best case (left) with ground truth (right)
The proposed method had an accuracy of 98.77% on the IDRiD Segmentation Dataset.
Optic Disc masked image (left) with the original image (right)
We used the IDRiD Segmentation Dataset for testing our proposed exudates detection algorithm.
 (a) |
 (b) |
(a), (b) Result of proposed exudates detection method (left) with original image (right).
The IDRiD Disease Grading dataset was used to train and evaluate the proposed network. It had an accuracy of 75.73% on a test set comprising 25% of the entire dataset.
DCNN training results with Training Accuracy, Test Accuracy and Training Loss
Basu, S., Mukherjee, S., Bhattacharya, A., Sen, A. (2021). Segmentation of Blood Vessels, Optic Disc Localization, Detection of Exudates, and Diabetic Retinopathy Diagnosis from Digital Fundus Images. In: Pan, I., Mukherjee, A., Piuri, V. (eds) Proceedings of Research and Applications in Artificial Intelligence. Advances in Intelligent Systems and Computing, vol 1355. Springer, Singapore. https://doi.org/10.1007/978-981-16-1543-6_16
For any queries, please contact: soham.basu07@gmail.com