Artificial intelligence-based grading of human cataracts: A feasibility study

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Human lens opacification, or cataracts, is a leading cause of visual impairment.
• Accurate grading of cataract severity is crucial for timely intervention.
• Current grading methods can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate an AI algorithm for automatic classification of human lens opacification (cataracts).
• To assess the feasibility of using anterior segment optical coherence tomography (OCT) for automated cataract grading.
• To evaluate the performance of a deep learning model in classifying nuclear opalescence (NO) and posterior subcapsular cataract (PSC).

Main Methods:

• A dataset of 1,802 unprocessed OCT images from 901 eyes was utilized.
• Cataract severity was graded using the Lens Opacities Classification System (LOCS) III.
• A convolutional neural network was trained and validated for automated classification of NO and PSC.

Main Results:

• The AI achieved 86.4% accuracy for posterior subcapsular cataract (PSC) classification, with a true positive rate of 81.8% and a true negative rate of 90.9%.
• Nuclear opalescence (NO) classification accuracy was 60.3%, with most misclassifications within one severity category.
• The algorithm demonstrated potential for automated grading of specific cataract types.

Conclusions:

• Deep learning models can effectively classify certain types of cataracts using OCT imaging.
• Automated cataract grading holds potential to improve diagnostic efficiency and consistency.
• Further research is warranted to refine algorithms and expand their application to other cataract types.