Detection and location of microaneurysms in fundus images based on improved YOLOv4 with IFCM

Abstract Diabetic retinopathy (DR) is one of the leading causes blindness for people suffering from diabetes. Microaneurysm (MA) initial symptom DR. MA a lesion based disease which starts as small red spots on retina and increases in size DR progresses finally leads to blindness. So eliminating can effectively prevent at an early stage. However, due complex retinal structure, different brightness contrast fundus images with effects factors such patients, environment changes, difference acquisition equipment, it difficult existing detection algorithms achieve accurate results location. Therefore, algorithm improved YOLOv4 (YOLOv4‐Pro) was proposed. First, Fuzzy C‐Means (IFCM) clustering proposed optimize anchor parameters target samples improve matching between anchors feature graphs. In order control noise efficiency, median filtering method employed update criterion function original FCM algorithm, K‐means initialize clustering. Second, SENet attention module added backbone enhance key information suppress background, improving confidence effectively. Finally, spatial pyramid pooling (SPP) neck acceptance domain output characteristics network, profits separating important context information. The IFCM verified Kaggle dataset compared other methods. Experimental show that optimizing prior frame make suitable dataset, improves accuracy network by nearly 5%, provides nice performance location images. This would help ophthalmologists finding exact retina, thereby simplifying process any manual intervention.