DR. PRASANNA VENKATESH RAMESH

Dr. SATHYAN PARTHASARATHY, DR. SHRUTHY VAISHALI RAMESH, Dr. R. RAMESH

Abstract

Purpose: A novel convolutional neural network (CNN) approach in detecting glaucomatous damage was employed, with confocal fundus images, to overcome blackbox dilemma in artificial intelligence (AI). In addition to identification of glaucoma, this tool intended to identify signs ranging from trivial splinter hemorrhages to catastrophic glaucomatous optic atrophy, due to customised human annotations. Methods: 1400 high resolution fundus images were utilised; 1120 images (80%) for training and 280 images (20%) for testing. 25 signs pertaining to glaucoma were annotated. You Only Look Once 5 algorithm was used for detection. The testing images were split into three groups: 90, 100 and 90 for three runs performed, once every 15 days. Results: Tests showed consistent increments from 94.5% to 98.5% accuracy in predicting diagnosis and intricate signs. Conclusion: With constant training via feedback mechanism, there was an upsurge in prediction accuracy, which helped overcome blackbox dilemma.

Full Text