Diabetic retinopathy screening using deep learning for multi-class imbalanced datasets.

Area of Science:

• Biomedical imaging
• Computer-aided diagnosis
• Machine learning

Background:

• Diabetic retinopathy (DR) screening relies on medical eye imagery for detecting vascular damage.
• Deep learning has advanced computer-aided diagnosis for DR but faces challenges like imbalanced datasets and inconsistent annotations.
• These challenges adversely impact deep learning model performance in DR detection.

Purpose of the Study:

• To conduct a comparative analysis of state-of-the-art methods for diabetic retinopathy detection.
• To address the impact of class imbalance in deep learning models for DR screening.
• To establish a baseline for future research on deep learning architectures for imbalanced datasets in this domain.

Main Methods:

• Comparative analysis of various state-of-the-art deep learning methods.
• Utilized three benchmark datasets for diabetic retinopathy: Kaggle DR detection, IDRiD, and DDR.
• Evaluated methods across classification, object detection, and segmentation tasks.

Main Results:

• Identified the impact of class imbalance on deep learning model performance in DR detection.
• Provided an extensive comparative analysis of different methods on benchmark datasets.
• Established a baseline for future research in DR screening using imbalanced data.

Conclusions:

• The comparative analysis serves as a foundation for future research in diabetic retinopathy detection.
• Highlights the need for robust deep learning approaches to handle imbalanced datasets in medical imaging.
• Aims to guide the selection of appropriate deep learning architectures for imbalanced DR datasets.