Efficient multi-kernel multi-instance learning using weakly supervised and imbalanced data for diabetic retinopathy...

Area of Science:

• Ophthalmology
• Computer Science
• Machine Learning

Background:

• Diabetic retinopathy (DR) is a major cause of vision loss, necessitating effective screening and diagnosis.
• Traditional DR screening faces challenges with accurate lesion detection (micro-aneurysms, hemorrhages) and false positive reduction.
• Weakly supervised learning is crucial for DR diagnosis due to limitations in labeled data.

Purpose of the Study:

• To address the challenges in supervised learning for diabetic retinopathy diagnosis.
• To develop a weakly supervised multi-class DR grading method using a multi-instance learning framework.
• To improve the accuracy and efficiency of DR screening through advanced computational methods.

Main Methods:

• Formulated weakly supervised multi-class DR grading as a multi-class multi-instance problem.
• Proposed a multi-kernel multi-instance learning method utilizing graph kernels.
• Implemented bi-level re-sampling (instance-level under-sampling, bag-level over-sampling) to enhance model performance.

Main Results:

• Achieved an overall classification accuracy of 0.916 on the Messidor dataset.
• Obtained an Area Under the Curve (AUC) of 0.957, indicating high diagnostic performance.
• Demonstrated superior performance compared to baseline and state-of-the-art methods.

Conclusions:

• The proposed multi-kernel multi-instance learning framework effectively handles imbalanced and weakly supervised data for DR grading.
• The bi-level re-sampling strategy significantly improves diagnostic performance in DR detection.
• This method offers a promising advancement for automated diabetic retinopathy screening and diagnosis.