1. Feature subset selection through nature inspired computing for efficient glaucoma classification from fundus images.
- Author
-
Singh, Law Kumar, Khanna, Munish, and Singh, Rekha
- Subjects
ARTIFICIAL intelligence ,OPTIMIZATION algorithms ,COMPUTER-aided diagnosis ,FEATURE selection ,MACHINE learning ,TONOMETERS - Abstract
The selection of the most efficient features for glaucoma identification is the subject of our investigation because this disease is rapidly increasing worldwide. This disease causes lifelong blindness due to damage to the eye's optical nerve. Ophthalmologists have traditionally used tonometry, pachymetry, and other methods to measure intraocular pressure in order to diagnose patients. Yet each of these judgments takes time, requires high professional experience, and can be open to human error (inter-observer variability). Therefore, scholars are currently engaged in the domain of medical imaging, specifically focusing on the analysis of retinal images for the purpose of predicting glaucoma. This research also has the same objective and aims to address the aforementioned challenges. This empirical study proposes an artificial intelligence-based computer-assisted diagnosis (CAD) system which is built to overcome these difficulties by providing the best features for machine learning techniques for categorizing subject retinal pictures as "healthy" or "sick". This study presents a new set of reduced hybrid features that were selected from an initial set of 36 features extracted from fundus images of benchmark datasets that belonged to different classes to categorize patient fundus images into two categories: "healthy" or "infected." The nature inspired computing-based Emperor Penguin Optimization (EPO) algorithm and the Bacterial Foraging Optimization (BFO) algorithm are utilized to implement feature selection (FS) process. Additionally, a novel hybrid algorithm combining these two techniques is also proposed. Seven machine learning (ML) classifiers are engaged to compute eight statistically based performance metrics along with execution time computation, and a comparison of those metrics is also provided in a detailed fashion. The recommended method exhibits a fortunate performance with the highest specificity of 0.9940, sensitivity of 0.9347, and maximum accuracy of 96.55%. Expert medical practitioners who are overworked may receive assistance from the proposed system in making the optimal decisions to preserve human vision. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF