Fast and Accurate Ophthalmic Medication Bottle Identification Using Deep Learning on a Smartphone Device

Area of Science:

• Ophthalmology
• Computer Science
• Artificial Intelligence

Background:

• Glaucoma medication adherence is crucial for preventing vision loss.
• Accurate identification of ophthalmic medications is essential for patient safety.
• Deep learning offers potential for automated medication identification.

Purpose of the Study:

• To assess the accuracy and efficacy of deep learning models, specifically convolutional neural networks (CNNs), for identifying glaucoma medication bottles.
• To develop an algorithm for predicting ophthalmic medication bottles using a mobile image-based dataset.

Main Methods:

• Seven CNN models were pretrained and retrained on 2250 mobile images of 5 ophthalmic medications.
• Model performance was evaluated using k-fold cross-validation (k=10).
• Top 2 models were embedded into iOS apps and tested on 1500 additional images.

Main Results:

• MobileNet V2 achieved the highest k-fold cross-validation accuracy (0.974) and shortest processing time (3.45 sec/image).
• In app-based evaluation, MobileNet V2 showed significantly higher prediction accuracy (0.86) than ResNet V2 (0.68).
• Processing times for app-embedded models were comparable (0.32 sec/image for MobileNet V2, 0.31 sec/image for ResNet V2).

Conclusions:

• Retrained MobileNet V2 accurately identifies ophthalmic medication bottles and functions within a smartphone environment.
• This study serves as a proof-of-concept for a CNN-based smartphone application.
• The application aims to empower patients by reducing medication errors.