Improved Diabetic Retinopathy Detection Using ERXG-PS Ensemble Algorithm and Modified Principal Component Analysis.

One of the diseases that affect the vision of human is diabetic retinopathy (DR). Numerous techniques were evolved for the identification of DR, but they are not easily accessible and unknown for many of the patients and this prevents them from getting help at the right time. If diabetes mellitus (DM) is left untreated for a long time, it results vision loss, thus the early identification of this disease is a necessary process. To identify and classify the DR from the datasets machine learning algorithms are commonly used. These algorithms do not contain proper preprocessing and feature extraction techniques, which leads to inaccurate detection and classification. Therefore, for the effective classification of DR, this paper proposes a novel Ensemble Random Extreme Gradient-based Puzzle Search (ERXG-PS) algorithm. Initially, data preprocessing is done in this paper using techniques like normalization, noise elimination, and grayscale conversion. Then feature extraction is done using modified principal component analysis (MPCA). The extracted images are fed into the classification model for the classification of DR. Finally, the extracted features are provided for classification, where the classification phase uses a novel ERXG-PS algorithm to classify DR and healthy images from diabetic retinopathy dataset, Diabetic Retinopathy MessidorEye_Pac_Pre-processed dataset, and IDRiD datasets. Then, the classified images are post-processed for the segmentation of the optic disc, blood vessels, and exudates of retinal images. The proposed technique attains an accuracy of 93.67% in the experiments conducted and it proves to be fit for practical implementation in the medical field. [ABSTRACT FROM AUTHOR]