AI algorithm shows high accuracy for autonomous diagnosis of keratoconus

ByRobert Linnehan

Add topic to email alerts

Receive an email when new articles are posted on

Please provide your email address to receive an email when new articles are posted on .

Subscribe

Added to email alerts

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

Back to Healio

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

Back to Healio

An artificial intelligence algorithm designed for the autonomous diagnosis of keratoconus was able to correctly differentiate all patients with keratoconus from healthy subjects, according to a speaker at the virtual American Society of Cataract and Refractive Surgery meeting.

“The artificial intelligence algorithm is a novel technique with excellent accuracy, sensitivity and specificity in the diagnosis of keratoconus,” Eyüp Özcan, MD, FEBO, said.

The algorithm was able to diagnose 42 of 43 keratoconic eyes, achieving an accuracy rate of 98%, a sensitivity rate of 95% and a specificity rate of 94%. No study before this had used anterior segment OCT scans for the autonomous detection of keratoconus, he said.

Özcan and colleagues at Bascom Palmer Eye Institute built a deep learning algorithm from images captured from a Bioptigen high-definition OCT (Leica Microsystems). The images were labeled with keratoconus diagnosis and used to build a deep learning algorithm, he said.

“We used transfer learning using the pretrained network from images existing in ImageNet. We then implemented the algorithm into MATLAB. We split the training data set into three sets: 80%, 10% and 10% in training, validation and testing, respectively,” he said.

The Bascom Palmer AI 1.0 deep learning model compared 43 eyes of 23 subjects with 7,740 images vs. 28 eyes of 28 subjects with 5,040 images. Diagnosis results of the deep learning model were compared with the clinical diagnosis given by board-certified corneal specialists at Bascom Palmer Eye Institute, Özcan said.

The algorithm was able to correctly differentiate all patients with keratoconus from healthy subjects, he said. – by Robert Linnehan

Reference:

Özcan E. Accuracy of an artificial intelligence algorithm in the autonomous diagnosis of keratoconus. Presented at: American Society of Cataract and Refractive Surgery meeting; May 16-17, 2020 (virtual meeting).

Disclosure: Özcan reports no relevant financial disclosures.