Coarse-to-fine classification for diabetic retinopathy grading using convolutional neural network

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a major complication of diabetes, often leading to blindness.
• Current DR grading methods are time-consuming and carry risks.
• Accurate DR grading is crucial for effective treatment.

Purpose of the Study:

• To develop an automated, accurate, and efficient tool for DR severity classification.
• To propose a hierarchically Coarse-to-fine network (CF-DRNet) for five-class DR grading.

Main Methods:

• A novel Coarse-to-fine network (CF-DRNet) using convolutional neural networks (CNNs).
• The Coarse Network classifies No DR vs. DR using an attention gate module.
• The Fine Network classifies four DR severity stages (mild, moderate, severe NPDR, PDR).

Main Results:

• CF-DRNet achieved high performance in classifying five stages of DR severity.
• The model demonstrated superior results compared to state-of-the-art methods on public datasets (IDRiD, Kaggle).
• Attention gate module effectively highlighted DR lesions and reduced background noise.

Conclusions:

• The proposed CF-DRNet offers an efficient and reliable automated solution for DR grading.
• This AI tool can significantly aid in clinical diagnosis and management of diabetic retinopathy.
• The hierarchical approach effectively captures DR grading complexity.