Machine Learning Driven Hybrid Method for Retinal Vessel Segmentation

Abstract

The early diagnosis of ophthalmic disease like diabetic retinopathy, glaucoma and hypertension is of higher importance by segmentation of the retinal vessels. The conventional methods of segmentation are usually not very robust to noise and morphological variation of the vessels. This research paper suggests that a hybrid approach involving machine learning and a combination of deep convolutional neural networks (CNNs) and conventional image processing (e.g., morphological filtering, adaptive thresholding) can be used to enhance the accuracy and generalization. Experimental performance on publicly obtained datasets (e.g., DRIVE, STARE, CHASE_DB1) show better performance than the current ones, and they have high sensitivity, specificity and area under the ROC curve (AUC). The new provided hybrid approach is effective and computationally efficient in clinical approaches.