Optimal Feature Selection and Extraction for Eye Disease Diagnosis

Optimal Feature Selection and Extraction for Eye Disease Diagnosis

Alli P. (Velammal College of Engineering and Technology, India) and S. K. Somasundaram (PSNA College of Engineering and Technology, India)
Copyright: © 2019 |Pages: 13
DOI: 10.4018/978-1-5225-5876-7.ch005
OnDemand PDF Download:
No Current Special Offers


Ophthalmologists utilize retinal fundus images of humans for the detection, diagnosis, and prediction of many eye diseases. Automatic scrutiny of fundus images are foremost apprehension for ophthalmologists and investigators. The manual recognition of blood vessels is most deceptive because the blood vessels in a fundus image are multifaceted and with low contrast. Unearthing of blood vessels proffers information on pathological transformation and can smooth the progress of rating diseases severity or mechanically diagnosing the diseases. The manual recognition method turns out to be annoying. Consequently, the automatic recognition of blood vessels is also more significant. For extracting the vessel in fundus images unswerving and habitual methods are obligatory. The proposed methodology is designed to effectively diagnose the eye disease by performing feature extraction succeeded by feature selection and to improve the performance factors such as feature extraction ratio, feature selection time, sensitivity, and specificity when compared to the state-of-art methods.
Chapter Preview

All eye blood vessels congregate at the Optic Disc (OD) whose detection also plays an imperative role in diagnosing the eye diseases. Existing template-based methodology (Aquino et al., 2010) uses Circular Hough Transform and edge detection techniques to segment the OD. An Optic Disc containing sub-image is wheedled out and its surrounding region are chosen but it does not automatically localize the selected features. An optimally comfortable morphological operator is used for exudates detection on diabetic retinopathy patients and low-contrast images (Sopharak et al., 2008). These mechanically detected exudates are validated by associating with expert ophthalmologists hand-drawn ground-truths but do not insert more detailed features to the system.

Exudates based diabetic macular edema detection in Fundus images (Giancardo et al., 2012) does not employ machine learning methods to categorize candidate exudates into true positives and false positives. Macular edema detection require extensive ground truth segmentations by experts which is problematic to attain and also fails to create a competitive diabetic retinopathy selection system to transparently diagnose the disease state. Combining algorithms (Qureshi et al., 2012) for automatic detection of optic disc and macula in Fundus images is used to discover the macula center. Combining algorithm arise issue in improving the sensitivity of tasks, such as the detection of the optic disc and blood vessels, and also fails to localize faint and small exudates.

Complete Chapter List

Search this Book: