Local Structure Awareness-Based Retinal Microaneurysm Detection with Multi-Feature Combination.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Diabetic retinopathy (DR) is a complication of diabetes, with retinal microaneurysms (MA) as its initial clinical sign.
• Early detection of MA is crucial for timely diagnosis and treatment of DR.
• Existing algorithms often overlook the importance of local structural features in MA detection.

Purpose of the Study:

• To propose an efficient local structure awareness-based method for retinal MA detection.
• To introduce a novel local structure feature, the ring gradient descriptor (RGD), for improved MA identification.
• To enhance MA detection performance through a multi-feature combination approach.

Main Methods:

• Development of the ring gradient descriptor (RGD) to capture structural differences between MA and surrounding retinal tissue.
• Integration of RGD with salience and texture features.
• Utilization of a Gradient Boosting Decision Tree (GBDT) classifier for candidate MA classification.

Main Results:

• The proposed LSAMFC algorithm demonstrated improved performance in detecting retinal microaneurysms.
• Combining RGD with traditional features significantly enhanced the model's accuracy.
• The area under the receiver operating characteristic curve (AUC) increased from 0.9615 to 0.9751 on the e-ophtha MA dataset and from 0.9066 to 0.9409 on the ROC dataset.

Conclusions:

• The novel RGD feature effectively describes local structural information for MA detection.
• The multi-feature combination approach, incorporating RGD, offers superior performance compared to traditional methods.
• This technique holds promise for assisting clinicians in the early diagnosis and management of diabetic retinopathy.