Automated Detection of Keratoconus using EfficientNetB7 : A Deep Learning Approach

A progressive and degenerative eye condition called keratoconus is illustrated by the cornea’s thinning and bulging into a cone-like form. The consequence of this deformity is severe visual impairment, including distorted vision, increased sensitivity to light, and potential vision loss if left untreated. Early and precise detection of keratoconus is crucial for improved monitoring and therapeutic outcomes, as timely intervention can prevent disease progression and improve patient outcomes. In this work, we implemented a deep learning approach leveraging the use of EfficientNet-B7 architecture to detect and classify keratoconus, suspect, and normal corneas. EfficientNet-B7, known for its superior performance and scalability, was employed as the backbone of our convolutional neural network (CNN). The dataset used for this study comprised multiple samples, each consisting of seven distinct corneal maps, including topographic, tomographic, and pachymetric representations. These maps provided a holistic view of corneal structure and variations, allowing the model to depict the complex patterns linked to keratoconus. The proposed methodology has achieved a validation accuracy of 97%. This study demonstrates how deep learning can significantly advance eye care, offering precise and early detection of keratoconus by incorporating corneal maps that can improve the quality of life of individuals at risk for keratoconus. This advancement highlights the importance of integrating advanced AI models into clinical practice for the early diagnosis and management of ocular diseases.