Tasks related to the diagnosis of eye diseases and the assessment of image quality in fundus images have become crucial tools for assisting in medical diagnoses. High-quality fundus images can provide clear and definite pathological features to enhance disease diagnosis. Similarly, the diagnosability of images serves as a crucial criterion for assessing the quality of fundus images. Building on this characteristic, this paper proposes a dual-task collaborative optimization network (DTCONet). It leverages disease diagnosis indicators to enhance quality assessment and improves diagnostic accuracy through optimization, achieving mutual enhancement. First, we designed a dual-task module to process disease diagnosis and quality assessment tasks in parallel by sharing fundus image features. Second, we introduced a dual-task collaborative optimization module to delve into the strong correlation between these tasks and iteratively optimize results through loop learning. This research offers fresh insights into the interplay between disease diagnosis and quality assessment in fundus images. In addition, considering that fundus images have a large amount of detailed information, this paper uses the CA attention to improve the model’s ability to perceive small structures and pathological features in the image. Extensive subjective and objective experiments on three widely used medical image datasets demonstrate the effectiveness and generalization of our method.