Automated diagnosis system for age‐related macular degeneration using hybrid features set from fundus images. Issue 1 (26th June 2020)

• Record Type: Journal Article
• Authors: Khalid, Samina; Akram, Muhammad Usman; Shehryar, Tehmina; Ahmed, Waqas; Sadiq, Marium; Manzoor, Mahak; Nosheen, Nelam
• Is Part Of: International journal of imaging systems and technology. Volume 31:Issue 1(2021)
• Page Start: 236
• Page End: 252
• Publication Date: 2020-06-26
• Subjects: AMD -- AFIO -- drusen -- fundus images -- HOG -- textural features; Imaging systems -- Periodicals; Image processing -- Periodicals; 621.367
• DOI: 10.1002/ima.22456
• Languages: English
• ISSNs: 0899-9457

Abstract

Abstract: Impairment to macula can cause loss of central vision. There are various macular disorders that can affect macular region and if not treated at an early stage can cause irreversible central vision loss. Age‐related macular degeneration (AMD) disorder is one of the most threading macular disorder. Bright lesion, drusens presence in macular region is known as the hallmark of AMD disorder. This bright lesion differentiation from other bright lesion like exudates is important for accurate diagnosis of AMD. Focus of this article is automated diagnosis of affected macular region by applying a hybrid features set containing textural, color, and structural/shape features for more accurate detection of AMD at an early stage using fundus images. These features also help to distinguish drusens from exudates. The proposed algorithm at first stage, detect macular region from input fundus image and then perform features extraction based on textural pattern, edge, and structural properties of macular region to classify abnormal macula from normal macula. For classification, we have used support vector machine (SVM), K‐nearest neighbor and neural networks but SVM classifier achieves high accuracy. The proposed algorithm is tested on publicly available STARE and locally available AFIO datasets. Attained sensitivity, specificity, and accuracy of our proposed system are 97.5%, 95% and 95.45%, respectively, when applied on STARE dataset. When we have applied our proposed system on AFIO dataset, we have attained sensitivity, specificity, and accuracy of 93.3%, 92% and 92.34%, respectively.