Neovascularization Detection and Localization in Fundus Images Using Deep Learning

Abstract

Proliferative Diabetic Retinopathy (PDR) is a severe retinal disease that threatens diabetic patients. It is characterized by neovascularization in the retina and the optic disk. PDR clinical features contain highly intense retinal neovascularization and fibrous spreads, leading to visual distortion if not controlled. Different image processing techniques have been proposed to detect and diagnose neovascularization from fundus images. Recently, deep learning methods are getting popular in neovascularization detection due to artificial intelligence advancement in biomedical image processing. This paper presents a semantic segmentation convolutional neural network architecture for neovascularization detection. First, image pre-processing steps were applied to enhance the fundus images. Then, the images were divided into small patches, forming a training set, a validation set, and a testing set. A semantic segmentation convolutional neural network was designed and trained to detect the neovascularization regions on the images. Finally, the network was tested using the testing set for performance evaluation. The proposed model is entirely automated in detecting and localizing neovascularization lesions, which is not possible with previously published methods. Evaluation results showed that the model could achieve accuracy, sensitivity, specificity, precision, Jaccard similarity, and Dice similarity of 0.9948, 0.8772, 0.9976, 0.8696, 0.7643, and 0.8466, respectively. We demonstrated that this model could outperform other convolutional neural network models in neovascularization detection.

An Automatic Detection and Classification System of Five Stages for Hypertensive Retinopathy Using Semantic and Instance Segmentation in DenseNet Architecture

“…In addition, we have performed state-of-the-art comparisons with four recent HR systems. DL-based HR models are utilized, such as DenseHyper [ 31 ], Arsalan-HR [ 38 ], Kriplani-AVR [ 39 ] and Tag-Semantic [ 40 ] because of the ease of their implementation. Mostly, they were developed to detect lesions from retinograph images for diabetic retinopathy (DR) or HR stage classification.…”

“…In comparison to other state-of-the-art approaches for HR recognition that use deep learning (DL) models, few research efforts use DL approaches for classifying HR from fundus videos. These four DL-based HR models are utilized: DenseHyper [ 31 ], Arsalan-HR [ 38 ], Kriplani-AVR [ 39 ] and Tag-Semantic [ 40 ]. This is because of their ease of implementation.…”

“…Recently, the authors in Tag-Semantic [ 40 ] have developed a deep model based on pretrained CNN with an interpretable Tag-Semantic approach to recognize the stage of diabetic retinopathy (DR) instead of hypertensive retinopathy (HR). However, this Tag-Semantic system can be used to detect the two stages of HR in case of the detection of the HR-related lesions.…”

“…In that study, the authors aggregated the residual neural network with CNN and achieved 94% detection accuracy. More recently, in [ 40 ], the authors developed a Tang-Semantic system based on semantic segmentation by a CNN architecture for the detection and localization of diabetic retinopathy (DR)-related lesions. The Tang-Semantic approach is useful to detect DR, but it cannot be applied to detect HR due to the AVR ratio.…”

Dorsal Hand Vein Image Enhancement Using Fusion of CLAHE and Fuzzy Adaptive Gamma

“…Sensitivity (Se), accuracy (Acc), and dice coefficient (Dice) were handpicked as the key performance matrices. The three performance matrices were calculated as follows [ 36 ]: …”