Dual-task collaborative optimization for fundus image disease diagnosis and quality assessment.

Area of Science:

• Medical image analysis
• Ophthalmology
• Computer vision

Background:

• Fundus image quality is crucial for accurate disease diagnosis.
• Diagnostic capability is a key metric for assessing fundus image quality (FIQ).
• Existing methods often address these tasks independently.

Purpose of the Study:

• To propose a dual-task collaborative optimization network (DTCONet) that explores the interplay between FIQ and disease diagnosis.
• To enhance the performance of both tasks through mutual promotion.
• To provide new insights into the relationship between FIQ assessment and disease diagnosis.

Main Methods:

• Developed a dual-branch feature extraction framework to capture fine structures and pathological characteristics.
• Designed a dual-task module for parallel processing of quality assessment and disease diagnosis using shared representations.
• Introduced a collaborative optimization module to exploit the correlation between the two tasks.

Main Results:

• The DTCONet demonstrated effectiveness in enhancing both fundus image quality assessment and disease diagnosis.
• Experiments on five datasets confirmed the model's performance and generalizability.
• The collaborative optimization approach showed significant improvements over independent task models.

Conclusions:

• The proposed DTCONet effectively integrates quality assessment and disease diagnosis for improved medical image analysis.
• The study highlights the synergistic relationship between FIQ and diagnostic performance.
• DTCONet offers a promising approach for advancing automated fundus image analysis in clinical settings.