Extreme Learning Machine (ELM) Algorithm for the Detection of Diabetic Retinopathy

Blood vessel recognition is crucial for many medical image analysis applications, such as diagnosis and screening for diabetic retinopathy. However, manual de-tection can take a while. To address this, re-searchers have developed a method that uti-lizes an Extreme Learning Machine (ELM) in a supervised manner to segment blood vessels in the retina. The process involves several stages, including pre-processing images, unsupervised and supervised learning, and post-processing techniques. During the pre-processing stage, various techniques such as image cropping, color modification, enhancement of contrast, Gabor processing, and rectification using half-waves are used to pre-process the images. The pre-processed images are utilized to create a feature vector and used for Principal Component Analysis (PCA). Pixels are separated into non-vessel and vessel clusters using clustering based on K-means on the PCA result. The ensemble classification process is then applied to the non-vessel region using root-guided decision trees with bagging. Finally, the outcomes of the ensemble classification and clustering methods are combined to produce the final segmentation image. Post-processing techniques, such as morphological approaches, are used to improve the vessel-segmented output. The proposed method is verified using publicly available photos from the DRIVE database and achieves an accuracy of 95.36% with higher sensitivity, indicating that more vessel pixels are detected effectively.