No-reference quality index for color retinal images.

Area of Science:

• Ophthalmology
• Medical Imaging
• Image Processing

Background:

• Retinal image quality assessment (RIQA) is crucial for accurate medical diagnosis by ophthalmologists and automated systems.
• Measure-based RIQA offers advantages over classification-based methods, providing quantitative confidence scores and guiding image enhancement.
• Existing methods may not adequately address variations in illumination and image quality.

Purpose of the Study:

• To introduce a no-reference numeric quality index for retinal image sharpness.
• To enhance the index with a homogeneity parameter for unevenly illuminated regions.
• To validate the index's performance against subjective scores and other quality measures.

Main Methods:

• A no-reference retinal image sharpness index was computed using wavelet decomposition.
• A homogeneity parameter, derived from the retinal image saturation channel, was incorporated to handle illumination variations.
• The proposed index was tested on two datasets with varying resolutions and quality grades.

Main Results:

• The developed quality index demonstrated a strong, statistically significant correlation (Spearman's > 0.8, p < 0.001) with subjective human quality scores.
• Multiclass classification using the index achieved high F-measures (0.84 and 0.95) on high and low-resolution datasets, respectively.
• The proposed index outperformed existing retinal image quality measures in both performance and speed.

Conclusions:

• The novel numeric quality index provides a reliable and efficient method for assessing retinal image sharpness.
• The index is effective across different image resolutions and quality levels.
• This tool can aid in diagnosis confidence, enhancement evaluation, and comparison of image processing techniques.