A dangerous eye condition known as diabetic retinopathy (DR) affects those who have diabetes and can cause blindness and visual loss. It is necessary to identify and treat DR in a timely manner to prevent vision loss. At present, fundus scans of the eye are used by qualified ophthalmologists to detect DR. The need for DR assessment, yet, outweighs the supply of qualified personnel according to the rapidly increasing percentage of people suffering diabetes, necessitating the creation of automatic DR detection devices. In this work, we offer a novel approach to DR detection that utilizes multitask learning. The suggested model is trained to execute several tasks concurrently, such as Gaussian blur, gray scaling, and DR classification, using a sizable dataset of fundus images. In comparison with conventional single-task models, the performance of the model improves and it is capable to make more accurate predictions by simultaneously optimizing all tasks. Our multitasking learning strategy beats state-of-the-art DR detection algorithms with regard to accuracy and F1 score, according to findings from experiments on a publically available DR dataset. Furthermore, the model’s Gaussian blur capabilities allow it to give illustrations for its predictions that can help physicians better grasp the foundation for diagnosing DR patients. To sum up, our suggested multitask learning method for DR identification shows encouraging outcomes and may prove to be a practical and successful DR tool for screening in a medical setting.