An Improved Microaneurysm Detection Model Based on SwinIR and YOLOv8

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a diabetes complication characterized by microvascular changes.
• Microaneurysms (MAs) are early indicators of DR, necessitating precise detection in retinal images.
• Accurate MA detection is vital for timely DR diagnosis and management.

Purpose of the Study:

• To develop an automated model, MA-YOLO, for detecting microaneurysms (MAs) in fluorescein angiography (FFA) images.
• To enhance MA detection accuracy by integrating super-resolution and specialized detection layers.
• To validate the MA-YOLO model's performance against existing state-of-the-art methods.

Main Methods:

• Utilized SwinIR for super-resolution reconstruction of FFA images to enhance MA feature details.
• Integrated a novel MA detection layer into the YOLOv8 architecture to address small feature detection and information loss.
• Employed transfer learning and optimized the loss function with Wise-IoU to improve model generalization and robustness.

Main Results:

• The MA-YOLO model achieved superior performance in MA detection compared to SSD, RetinaNet, YOLOv5, YOLOX, and YOLOv7.
• Achieved optimal performance metrics: 88.23% recall, 97.98% precision, 92.85% F1 score, and 94.62% AP.
• Demonstrated effectiveness in handling geometric factors and sample imbalance through loss function optimization.

Conclusions:

• The proposed MA-YOLO model is highly effective for automated microaneurysm detection in FFA images.
• This AI-driven tool can assist ophthalmologists in diagnosing and monitoring the progression of diabetic retinopathy.
• The study highlights the potential of integrating advanced AI techniques for improved ophthalmic diagnostics.