Posters

Automatic classification of corneal imagining using deep learning

Poster Details

First Author: I.Lavy ISRAEL

Co Author(s):    I. Magal   T. Levi Weinberg                 

Abstract Details

Purpose:

To evaluate the accuracy of corneal imaging diagnosis using Deep Learning

Setting:

A retrospective review of corneal imaging at Hadassah Medical Centre in 2019

Methods:

A method based on Deep learning tools was developed to automatically detect six conditions in 1021 corneal slit lamp images. The objectives of our model were to classify corneal images as: 1. Clear cornea. 2. Clear corneal graft 3. Corneal ulcer 4. Corneal Dendrite 5. Corneal dystrophy 6. Pterygium Main outcome was measured by area under the ROC curve.

Results:

Deep learning of the arithmetical mean output data of these six classification showed: 1. accuracy of 0.881 and top accuracy of 0.968 in detecting clear cornea. 2. accuracy of 0.875 and top K accuracy of 0.962 in detecting Clear corneal graft. 3. accuracy of 0.875 and top K accuracy of 0.962 in detecting corneal ulcer. 4. accuracy of 0.881 and top K accuracy of 0.962 in detecting corneal dendrite. 5. accuracy of 0.888 and top K accuracy of 0.975 in detecting corneal dystrophy. 6. accuracy of 0.875 and top K accuracy of 0.975 in detecting pterygium.

Conclusions:

Deep learning tools effectively discriminates different corneal images, and furthermore classifies the disease. It is suggested that this will enable early detection and timely intervention of corneal diseases, alleviating the strain on limited clinical resources and improving diagnosis quality

Financial Disclosure:

is employed by a for-profit company with an interest in the subject of the presentation