Automatic microaneurysm detection in fundus image based on local cross-section transformation and multi-feature...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Retinal microaneurysms (MA) are early indicators of DR.
• Detecting small, low-contrast MAs in fundus images is challenging.

Purpose of the Study:

• To develop an automated method for detecting microaneurysms (MA) in retinal fundus images.
• To improve the accuracy and efficiency of early diabetic retinopathy (DR) detection.

Main Methods:

• A two-stage approach: MA candidate extraction and classification.
• Utilized local minimum region extraction and block filtering for candidate identification.
• Employed local cross-section transformation (LCT) to enhance MA features.
• Trained an under-sampling boosting-based classifier (RUSBoost) for final detection.

Main Results:

• The method was evaluated on three public datasets (e-ophtha-MA, DiaretDB1, ROC training set).
• Achieved high sensitivity at low false positive rates.
• Obtained FROC scores of 0.516, 0.402, and 0.293, comparable to state-of-the-art methods.

Conclusions:

• Local cross-section transformation effectively enhances MA discrimination.
• The proposed method demonstrates robust and consistent state-of-the-art performance.
• Automated MA detection aids in early DR diagnosis and vision preservation.