MIRank-KNN: multiple-instance retrieval of clinically relevant diabetic retinopathy images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Diabetic retinopathy (DR) is a leading cause of blindness in adults.
• Early detection and treatment of DR are crucial for preventing vision loss.
• Computer-aided diagnosis offers potential improvements in DR screening and diagnosis.

Purpose of the Study:

• To develop an automated and unsupervised approach for retrieving clinically relevant fundus images for DR screening.
• To enhance the efficiency of DR screening and diagnosis through improved image retrieval.

Main Methods:

• A multiclass multiple-instance framework was proposed for image retrieval, considering localized DR lesions.
• A novel feature space was developed using modified color correlograms and steerable Gaussian filter responses.
• Fast radial symmetric transform points were utilized for feature selection.

Main Results:

• The proposed approach demonstrated superior performance compared to existing methods in experiments.
• Real DR images from five diverse datasets were used for validation.
• The method effectively retrieves clinically relevant images for DR diagnosis.

Conclusions:

• The developed automated image retrieval system shows significant promise for improving DR screening efficiency.
• The multiclass multiple-instance framework and novel feature space are effective for DR image analysis.
• This approach has the potential to aid clinicians in faster and more accurate DR diagnosis.