Microaneurysms segmentation with a U-Net based on recurrent residual convolutional neural network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Microaneurysms (MAs) are critical indicators for early diabetic retinopathy diagnosis.
• Manual annotation of MAs is time-consuming and requires expert knowledge.
• Automated MA segmentation is challenging due to low image contrast and small MA size.

Purpose of the Study:

• To develop an automated method for microaneurysm segmentation.
• To enhance the performance of deep learning models for medical image segmentation.

Main Methods:

• Proposed a novel deep learning architecture, deep recurrent U-Net (DRU-Net).
• Integrated deep residual learning and recurrent convolutional operations into the U-Net framework.
• Evaluated DRU-Net on the E-Ophtha and IDRiD public datasets.

Main Results:

• DRU-Net achieved superior performance compared to existing methods like U-Net, FCNN, and ResU-Net.
• Achieved 0.9999 accuracy and 0.9943 AUC on the E-Ophtha dataset.
• Obtained an AUC of 0.987 on the IDRiD dataset, representing the first reported MA segmentation results for this dataset.

Conclusions:

• DRU-Net demonstrates state-of-the-art performance in microaneurysm segmentation.
• The proposed architecture effectively accumulates features for improved MA detection.
• DRU-Net offers a promising solution for automated early detection of diabetic retinopathy.