Published Paper Details:

Corneal disease, such as endothelial dysfunction and other dystrophies, are the leading causes of visual impairment and blindness worldwide. The mainstay of treatment for corneal disease is corneal transplantation. Unfortunately, the success of this approach is limited by restrictions in tissue availability and immune rejection as well as difficulties managing postoperative recovery. The advances in the diagnosis, treatment, and management of corneal disease have burgeoned due to the introduction of machine learning (ML) techniques. Although they have promising potential, existing ML models are encumbered by several challenges, including limited and imbalanced datasets, inconsistent data quality, lack of interpretability, and poor integration of multimodal data. These limit the accuracy and clinical utility of existing solutions. In this paper we review the current status of ML for corneal disease research, highlight challenges that are common in this space, and identify future directions that should be explored in order to address these challenges. Proposed approaches include data sharing and collaboration between research groups, improved data augmentation, advanced techniques such as transfer learning and few-shot learning, and interpretable models. Encouragement and consideration of such issues will motivate future ML studies to help establish a more reliable, generalizable, and clinically useful application of ML in corneal disease management.