Multi-Feature Extraction with Ensemble Network for Tracing Chronic Retinal Disorders Muhammad Zubair Khan 1 , Yugyung Lee 2 , Arslan Munir 3 and Muazzam Ali Khan 4 Abstract—The retina manifests a vital role in tracing chronic retinal disorders. It is located near the optic nerve that transforms the captured light into neural signals. The most prominent chronic eye diseases exhibit themselves in the retina. The analysis of a retina for detecting disease symptoms is quite challenging. Most of the prior methods developed using shallow and deep learning algorithms primarily emphasized single feature extrac- tion for disease diagnosis. The underlying article has designed an ensemble network for extracting multiple retinal features using a single comprehensive platform. It includes a set of models that reflect feature-based needs to prevent intensity loss, micro- vessels overlap, and data redundancy. The proposed method has experimented with prominent benchmark datasets developed for vessels tree, optic disc/cup, and arteries/veins extraction. It is also compared with other methods and achieved promising results. Our platform is helpful for physicians to trace the variations in the retina of subjects facing chronic retinal disorders. Index Terms—deep learning, e-health, image analysis, retinal features, semantic segmentation. I. I NTRODUCTION The advancement of AI algorithms has inspired researchers and scientists to apply these techniques in medical diagnosis, monitoring, and treatment. A diverse range of methods are designed to target a set of medical challenges. The effort is conducted to develop minimalistic solutions with reduced computational complexities in medical image analysis. The core emphasis is to localize the anatomical structures and cat- egorize each pixel in a pre-defined set of classes, a critical step towards the image-guided diagnosis. The publicly available shared pool of resources, including datasets, algorithms, and source code, has motivated and appreciated the deep learning community to contribute in applied optics. It is found that these techniques have provided a solid foundation in tracing chronic ocular disorders such as diabetic retinopathy (DR), hypertensive retinopathy (HR) and glaucoma [1]–[7]. These disorders can affect both central and peripheral vision and are prominent in subjects with diabetes, hypertension, and intraocular pressure. The DR creates a blockage in vessels caused by an increased amount of blood glucose. It prevents the smooth supply of oxygen and essential nutrients from 1 Muhammad Zubair Khan, School of Computing and Engineering, Univer- sity of Missouri-Kansas City, Kansas City, USA (mkzb3@mail.umkc.edu) 2 Yugyung Lee, School of Computing and Engineering, University of Missouri-Kansas City, Kansas City, USA (leeyu@umkc.edu) 3 Arslan Munir, Department of Computer Science, Kansas State University, Manhattan, USA (amunir@ksu.edu) 4 Muazzam Ali Khan, Department of Computer Science, Quaid-i-Azam University, Islamabad, Pakistan (muazzam.khattak@qau.edu.pk) nourishing the human retina. In HR, chronic hypertension results in arterial and venous occlusion. The extended pressure in arteries and veins ruptures the vessels and causes blood and other fluids to enter the retina. The study conducted in [8] showed that the subjects with retinopathy are potentially at a high risk of vision loss, ischemic stroke, and heart failure. The intraocular pressure in glaucoma, developed by aqueous humor, results in optic nerve deterioration. It produces blind spots in the visual field and gradually progresses into perma- nent blindness. Detecting symptoms of these ocular disorders at an early stage through retinal features variation is vital to prevent vision loss. The initial step to analyze the root cause is to precisely segment the retinal features. Common symptoms reflected through retinal features include neovascularization, microaneurysms, arterial and venous occlusion, optic nerve damage, and change in the optic disc (OD) to optic cup (OC) ratio. The underlying article has proposed an ensemble network that can segment multiple features for detecting early symptoms of retinal diseases. The main contribution of this article includes the following: 1) A multi-feature extraction algorithm is proposed to trace the symptoms of chronic retinal disorders. 2) The region of interest and feature to extract influence the training mode and model selection. 3) The method produces a context-aware vessels tree, pre- vents artery/vein overlapping, and eliminates redundant pixels for effective optic disc/cup extraction. 4) The method is evaluated on benchmark datasets publicly available with different focal points and dimensions. II. RELATED WORK Deep semantic segmentation has a vital role in medical image analysis. It extracts the regions of interest required for medical diagnosis. A core concept is to effectively label every pixel to its corresponding class and elevate the system precision. Among several architectural solutions proposed by the AI community, the U-Net model [9] is the most prominent technique applied for medical image analysis. It is capable of operating with limited data samples. In [10], a method to analyze non-trivial pathologies with an effective response towards the central vessel reflex phenomenon is highlighted. A deformable U-Net architecture [11] is developed to capture the context knowledge by integrating low and high-end feature maps. The authors of [12] have offered a deeper architectural version to reduce the vanishing gradient problem with ade- quate feature representations. The method applied the residual and recurrent blocks along with the baseline model proposed