Deep Learning-Based Recognition of Pharmaceutical Pills using Convolutional Neural Networks

Abstract

Drug identification technology is a critical step that pharmacies take to ensure they are dispensing the correct drug. Photos of pills can be used to determine what medication it is, and which manufacturer produced it. Elderly patients are more likely to lose or misplace their medicines, especially if they are using a pill organizer. With the proper technology, patients will be able to determine which drugs are which. Convolutional Neural Networks identified medications from images with over 90% accuracy. This algorithm allows for better verification processes in the pharmacy and provides a tool for patients and global supply chain professionals with practical decision-support. Existing safeguards such as barcoding and electronic prescribing reduce, but do not eliminate, the risk of administering the wrong medication, particularly in environments complicated by polypharmacy and globalized supply chains. The need for reliable, scalable verification at the pill level has therefore become a pressing concern in pharmacovigilance.