Ensemble based adaptive over-sampling method for imbalanced data learning in computer aided detection of...

Area of Science:

• Ophthalmology
• Computer Science
• Machine Learning

Background:

• Diabetic retinopathy (DR) is a leading cause of blindness, necessitating early detection.
• Automated screening systems for DR can aid ophthalmologists and prevent vision loss.
• Microaneurysms (MAs) are early indicators of DR, but their detection is challenged by high false positive rates due to class imbalance.

Purpose of the Study:

• To develop an effective automated system for microaneurysm detection in diabetic retinopathy.
• To address the class imbalance problem inherent in identifying microaneurysms.
• To enhance the performance of extreme learning machines (ELM) for DR screening using novel ensemble methods.

Main Methods:

• Proposed an ensemble-based adaptive over-sampling algorithm (ASOBoost) to mitigate class imbalance.
• Utilized ensemble frameworks including Boosting, Bagging, and Random Subspace for microaneurysm detection.
• Combined adaptive over-sampling with ensemble techniques to reduce induction biases and improve classification.

Main Results:

• The ASOBoost method demonstrated superior performance compared to existing class imbalance learning methods.
• Achieved higher Area Under the ROC Curve (AUC) and G-mean values, indicating improved detection accuracy.
• Effectively reduced false positives in microaneurysm identification, crucial for reliable DR screening.

Conclusions:

• Ensemble-based adaptive over-sampling is a promising approach for addressing class imbalance in DR detection.
• The proposed ASOBoost method enhances the generalization and classification performance of ELMs for automated DR screening.
• This work contributes to more accurate and efficient automated systems for early diabetic retinopathy detection and management.