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

Fundus image disease diagnosis and quality assessment have emerged as essential tasks in medical image analysis. High-quality fundus images provide clear and well-defined pathological features, thereby enhancing diagnostic performance; similarly, diagnostic capability is regarded as one of the key criteria for assessing fundus image quality (FIQ). Motivated by this observation, we propose a dual-task collaborative optimization network (DTCONet) to explore the intrinsic interplay between quality assessment and disease diagnosis, enabling mutual promotion and performance enhancement. Specifically, DTCONet adopts a dual-branch feature extraction framework to strengthen the model's perception of fine structures and pathological characteristics. A dual-task module is then designed to process quality assessment and disease diagnosis in parallel by sharing fundus image representations. Furthermore, a collaborative optimization module is introduced to fully exploit the strong correlation between the two tasks. This study provides new insights into the relationship between FIQ assessment and disease diagnosis. Extensive quantitative and qualitative experiments on five widely used medical image datasets demonstrate the effectiveness and generalizability of the proposed DTCONet.