A Vision Language Correlation Framework for Screening Disabled Retina.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinopathy encompasses retinal disabilities leading to vision impairment or blindness.
• Optical coherence tomography (OCT) aids in early detection of retinal abnormalities.
• Existing autonomous screening systems often lack clinical manifestations, limiting their utility for physicians.

Purpose of the Study:

• To develop a novel framework for recognizing retinal disabilities by fusing retinal imagery with clinical prompts.
• To enhance autonomous retinal screening systems by incorporating clinical manifestations for better alignment with ophthalmologists' grading.

Main Methods:

• A novel framework utilizing vision-language correlation between retinal images and clinical prompts was developed.
• The framework was rigorously tested on six public datasets.
• Blind testing experiments were conducted with expert clinicians to evaluate clinical significance.

Main Results:

• The proposed framework outperformed state-of-the-art methods across multiple metrics on six public datasets.
• Blind testing demonstrated a statistically significant correlation coefficient of 0.9185 and 0.9529 with two expert clinicians.
• The system showed high accuracy in recognizing different types of retinal disabilities.

Conclusions:

• The novel framework effectively integrates retinal imagery and clinical information for accurate retinopathy screening.
• The system's performance in blind tests suggests its potential for real-world clinical deployment.
• This approach bridges the gap between automated analysis and clinical relevance in retinal disease diagnosis.