Detection of Cataracts from Retinal Fundus Image Using Ensemble Machine Learning Approach

Cataract is a common eye condition that causes clouding of the eye's natural lens, resulting in blurry vision and it is very common in older ages people. However, it can also occur in younger aged people due to various genetical effects, diabetes, smoking, and prolonged exposure to sunlight. Cataracts can be treated through surgery to remove the clouded lens and replaced by an artificial one, which can significantly improve the vision. This problem can be reduced through early detection, regular eye examinations etc. In this research, a Contrast Limited Histogram Equalization (CLAHE) of retinal fundus images is applied for a better presentation of the above cataract effects. These improved images are fed to a machine learning-based model for effective detection. Finally, the ensembling of the proposed machine learning-based classifiers is performed for the accurate detection of cataracts. The proposed model was tested experimentally on a real dataset and achieves excellent performance with accuracy, precision, and recall scores of 99.67%, 100%, and 99.25%, respectively which outperform the baseline methods. The robustness of the proposed ensemble model is evaluated through 5-fold cross-validation and achieved an average accuracy of 99.6%. The implementation code can be accessed from the GitHub repository: https://github.com/nahid-tech/cataract-detection.