PharmaChecked: A Low-Cost Visual Identification System for Counterfeit Pharmaceuticals

Overview

Counterfeit drugs have become increasingly prevalent globally, posing a serious threat to patient safety and healthcare systems, especially in lower-resourced areas. Counterfeit medications can cause significant harm ranging from over/under dosage to containing misleading ingredients or contamination. With the limited availability of laboratory drug verification methods that are often costly, there is a need for effective, affordable and scalable tools to assist in identifying counterfeit medications. This project aimed to address this need by developing an efficient and low-cost system to identify counterfeit drugs through noninvasive, visual data. This visual system used a public dataset of pre-labeled image samples of both authentic and counterfeit drugs, including differences in camera angle and lighting. Using a Convolutional Neural Network (CNN) architecture for binary classification, the system took image data and learned to discriminate based on features such as pill morphology and industry-standard packaging characteristics. Through extensive training, validation, and performance evaluation, the visual system was able to achieve 94% accuracy and a near-perfect ROC-AUC value of 0.9835, showing a strong ability to discriminate between samples. The system was able to accept images input by the user and output results quickly, making it highly scalable given large image datasets. Currently this system is best used as a preliminary screening or secondary confirmation tool to assist healthcare professionals and pharmacists in detecting counterfeits. With further development, it could be implemented as a standalone tool to enhance drug security and patient safety.