Disclosures: Lin reports consulting for Johnson & Johnson and speaking for Aerie Pharmaceuticals.
August 25, 2021
1 min read

Artificial intelligence may accurately identify glaucoma medication bottles

Disclosures: Lin reports consulting for Johnson & Johnson and speaking for Aerie Pharmaceuticals.
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

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

Artificial intelligence has demonstrated the ability to differentiate bottles of ophthalmic medications and could feasibly function on a smartphone platform, according to a study published in Ophthalmology Glaucoma.

Researchers pretrained seven convolutional neural network (CNN) models to classify five common ophthalmic medications using a set of 2,250 images captured by mobile phones. The two top-performing networks were included in separate iOS apps to test their ability to classify 1,500 images not included in the training set.

“We evaluated several known CNN algorithms to see which algorithm achieves the

best accuracy and computing efficiency to accomplish this task,” study co-author Ken Y. Lin, MD, PhD, told Healio/OSN. “Computing efficiency is important because our goal is to have a CNN that is both accurate and sufficiently lightweight to run on a smartphone device.”

The sets included bottles of brimonidine tartrate, dorzolamide-timolol, latanoprost, prednisolone acetate and moxifloxacin.

MobileNetV2 and ResNetV2 produced the most favorable results, with cross-validation accuracy of 0.974 and 0.961, respectively. MobileNetV2 had a significantly greater prediction accuracy than ResNetV2 once the networks were loaded into apps. MobileNetV2 had an image prediction accuracy of 0.86 compared with 0.68 for ResNetV2.

“There is an inherent irony in the design of most medication eye drop bottles. The writing is often too small, and the color coding on the bottle cap has been shown to be an unreliable feature to identify medications, especially for glaucoma patients who often have reduced contrast sensitivity and color vision. Yet, these patients depend critically on the correct use of these medications to maintain their vision,” Lin said.

The lack of a large, openly available data set was a limitation of the study, so the researchers had to create one. Other limitations include variations in different bottles of the same medication and the technological literacy and smartphone access needed by the target demographic to use such an app. Future directions include expanding the scope of research to include more medications.