Explainable Artificial Intelligence with Deep Learning Framework for Glaucoma Assessment on Fundus Images

The diagnosis of glaucoma is generally carried out by scanning the eye through special sensors such as the fundus scanner. This is followed by an expert analysis resulting in an outcome stating whether the eyes are affected or not. With the discoveries of state-of-the-art algorithms, now a days the assessment of such fundus images are done with improved accuracy. In this context, a framework has been proposed in this paper that involves two phases to not only assess the state of glaucoma but also it reveals the insight into the affected tiny patches of affected nerve tissues. For this, firstly, state-of-the-art deep learning concept is being used to make an efficient assessment of the glaucoma condition. Secondly, the concept of explainable artificial intelligence (XAI) is utilized to trace insight into the actually affected tiny region in the fundus image where the damage is realized. Suitable experimentation on a sample set of 1200 fundus images is performed. The overall rate of accuracy so obtained (using k-fold cross validation) is 93.25%. The performance of the proposed framework also reveals satisfactory output in terms of demonstrating the actual region of damage.