Automatic Detection of Microaneurysms in Fundus Images Using an Ensemble-Based Segmentation Method
Vidas Raudonis,
Arturas Kairys,
Rasa Verkauskiene,
Jelizaveta Sokolovska,
Goran Petrovski,
Vilma Jurate Balciuniene,
Vallo Volke
Affiliations
Vidas Raudonis
Automation Department, Kaunas University of Technology, 51368 Kaunas, Lithuania
Arturas Kairys
Automation Department, Kaunas University of Technology, 51368 Kaunas, Lithuania
Rasa Verkauskiene
Institute of Endocrinology, Lithuanian University of Health Sciences, 50140 Kaunas, Lithuania
Jelizaveta Sokolovska
Faculty of Medicine, University of Latvia, 1004 Riga, Latvia
Goran Petrovski
Center of Eye Research and Innovative Diagnostics, Department of Ophthalmology, Oslo University Hospital and Institute of Clinical Medicine, Faculty of Medicine, University of Oslo, 0372 Oslo, Norway
Vilma Jurate Balciuniene
Lithuanian University of Health Sciences, 44307 Kaunas, Lithuania
Vallo Volke
Faculty of Medicine, Tartu University, 50411 Tartu, Estonia
In this study, a novel method for automatic microaneurysm detection in color fundus images is presented. The proposed method is based on three main steps: (1) image breakdown to smaller image patches, (2) inference to segmentation models, and (3) reconstruction of the predicted segmentation map from output patches. The proposed segmentation method is based on an ensemble of three individual deep networks, such as U-Net, ResNet34-UNet and UNet++. The performance evaluation is based on the calculation of the Dice score and IoU values. The ensemble-based model achieved higher Dice score (0.95) and IoU (0.91) values compared to other network architectures. The proposed ensemble-based model demonstrates the high practical application potential for detection of early-stage diabetic retinopathy in color fundus images.
