Identification of suitable fundus images using automated quality assessment methods

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal image analysis is vital for diagnosing eye diseases.
• Automated systems require high-quality input images for reliable diagnosis.
• Current methods for retinal image quality assessment (IQA) need improvement.

Purpose of the Study:

• To develop and evaluate a novel approach for assessing retinal image quality.
• To classify retinal images into 'good,' 'bad,' and 'outlier' categories.
• To identify medically suitable images for automated retinal diagnosis.

Main Methods:

• A three-class grading system (good, bad, outlier) was implemented.
• A dataset of 216 retinal images (Diabetic Retinopathy Image Database) was created.
• The approach was validated using public datasets (DRIvE, Standard Diabetic Retinopathy Database Calibration level 1).

Main Results:

• The highest F1 scores achieved were 99.60% (good), 96.50% (bad), and 85.00% (outlier).
• The suitable image detection accuracy reached 98.08%.
• The proposed method demonstrated high performance in classifying retinal image quality.

Conclusions:

• The developed retinal image quality assessment approach is effective for identifying suitable images.
• This method can significantly enhance the performance of automated retinal analysis systems.
• The approach is robust and can be integrated into existing diagnostic platforms.