[Development and application of a fundus image quality assessment system based on computer vision technology].

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Diabetic retinopathy screening relies on high-quality fundus images.
• Manual quality assessment of fundus images is time-consuming and subjective.
• Automated systems are needed to improve efficiency and objectivity in image quality control.

Purpose of the Study:

• To develop and validate a computer vision-based system for automated fundus image quality assessment.
• To compare the system's accuracy against manual evaluations by professional readers.
• To assess the system's objectivity and efficiency in classifying image quality.

Main Methods:

• A four-module system was developed: preprocessing, quality evaluation, content detection, and result output.
• The system automatically identifies key features like the optic disc and macula.
• A dataset of 2,397 fundus images from type 2 diabetes patients was used for validation, comparing system results with 12 professional readers.

Main Results:

• The system achieved an overall consistency rate of 97.54% compared to manual assessments.
• High agreement was observed for qualified (96.86%) and unqualified (99.82%) images.
• The system rapidly evaluated images (under 1 second per image) and identified common quality issues like blurriness, poor illumination, and lack of key structures.

Conclusions:

• The developed fundus image quality assessment system demonstrates high accuracy and objectivity.
• It provides results consistent with expert ophthalmologists' judgments.
• The system offers a reliable and efficient tool for quality control in fundus image analysis, particularly for diabetic retinopathy screening.