Detection of regions of interest in retinal images using artificial neural networks and K-means clustering

The paper presents a new method for detecting and localizing regions of interest (ROI) in retinal images. In the learning phase, the focus is on an efficient feature selection based on K-means clustering of feature values and on artificial neural networks (ANNs) for image processing and textural features computation. Finally, a voting scheme identifies the regions of interest. The experiments were conducted to detect the optic disc, macula, exudates and hemorrhages. The results obtained on 150 test images show the efficiency of the proposed method in terms of accuracy, compared with other similar researches.