Retinal image quality assessment based on image clarity and content

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Automated retinal screening systems require high-quality images to prevent misdiagnosis.
• Poor image quality is a significant challenge in automated retinal image analysis.
• Existing methods may lack efficiency or comprehensive quality assessment.

Purpose of the Study:

• To introduce a novel no-reference transform-based algorithm for retinal image quality assessment (RIQA).
• To address multiple quality issues including sharpness, illumination, homogeneity, field definition, and content.
• To develop a computationally inexpensive yet effective RIQA method for automated screening.

Main Methods:

• Utilized wavelet-based features for evaluating image sharpness and illumination.
• Designed a retinal saturation channel combined with wavelet features for homogeneity assessment.
• Employed color information to differentiate retinal from non-retinal images and a classifier for overall quality evaluation.

Main Results:

• Individual feature sets achieved an area under the receiver operating characteristic curve (AUROC) above 0.99 on diverse datasets.
• The collective feature-based classifier demonstrated superior performance compared to existing literature methods.
• The algorithm effectively and comprehensively addressed various quality issues.

Conclusions:

• The proposed transform-based RIQA algorithm is highly accurate and efficient.
• It is suitable for integration into automated screening systems to improve diagnostic reliability.
• This method offers a robust solution for ensuring the quality of retinal images in clinical practice.