Abstract:
This study aims to explore the efficacy of convolutional neural networks (CNNs), a deep
learning methodology, in the early diagnosis of keratoconus, a severe corneal disorder with
potential to lead to blindness. Predominantly manifesting in the second decade of life,
keratoconus is a non-discriminatory disease affecting individuals across all sexes and races.
The focus of this research was to assess and compare the diagnostic capabilities of three distinct
deep CNN architectures: MobileNetV2, InceptionV3, and VGG16, in identifying keratoconusrelated pathologies. Through rigorous experimental analysis, it was observed that the
MobileNetV2 model demonstrated superior performance in detecting keratoconus, achieving
an impressive accuracy rate of 97.94%. This rate not only surpasses that of its D-CNN
counterparts but also underscores the model's stability, robustness, and potential for real-world
application in precise keratoconus identification. Further, the InceptionV3 D-CNN model
exhibited exceptional performance in terms of precision, recall rates, and F1 scores. These
metrics significantly endorse the model’s utility as an effective diagnostic tool, emphasizing
its ability to accurately identify keratoconus cases while minimizing the incidence of false
positives and negatives. The findings from this study highlight the practical applicability and
reliability of the MobileNetV2 model in the early and accurate detection of keratoconus,
showcasing its potential as a significant contribution to the field of ophthalmological
diagnostics.
