XAI Meets Ophthalmology: An Explainable Approach to Cataract Detection Using VGG-19 and Grad-CAM

Cataract is the most common cause of blindness in the world, and early detection is critical for successful treatment. Deep learning algorithms have recently demonstrated promising results in automated cataract detection from fundus images. Yet, the inability of these black box models to be interpreted raises concerns regarding their clinical application. In this paper, we present an explainable approach for detecting cataracts using the VGG-19 convolutional neural network (CNN) and the Gradient-weighted Class Activation Mapping (Grad-CAM) visualization technique. We trained and tested our model using 2,112 high-resolution fundus images from a publicly available dataset. According to our findings, the proposed method obtains a very high accuracy of 97%. Furthermore, the Grad-CAM visuals show the image regions that contribute to the model's decision, providing insights into the diagnosis process and increasing the model's trustworthiness. The findings of this study could assist in improving patient outcomes and lowering healthcare costs by allowing for the early detection and diagnosis of cataracts. Overall, the combination of VGG-19 with Grad-CAM offers a viable option in the medical domain for identifying cataracts and understanding CNN - based decisions.