OCULAR DISEASE CLASSIFICATION BY USING GLOBAL-LOCAL MULTI-LABEL CLASSIFICATION NETWORK
Keywords:
Classification, Fundus images, Vision Transformers, Deep learning, Transfer learningAbstract
The timely detection of ocular fundus disease has emerged as a crucial concern in recent decades due to the lack of early-stage symptoms and the potential for severe effects such as blindness. In this context, it is imperative for physicians to accurately diagnose ocular fundus disease during its first phases, a task that is both labor-intensive and demanding. Various machine learning algorithms are employed to provide assistance to doctors through the implementation of helpful actions. Therefore, this study presents an innovative methodology for the classification of fundus pictures, encompassing eight distinct groups that include various pathologies as well as normal cases. The proposed model, referred to as the Global-Local Multi-Label Classification Network, incorporates both global and local perspectives in the learning process. This is achieved by leveraging a vision transformer for global understanding and employing convolutional layers for local analysis. The performance of this model is assessed using the ODIR-5K dataset, and the findings indicate that it outperforms existing state-of-the-art approaches across several measures.
