Real-Time CNN-Based Generic Medicine Name Classification for Prescription Recognition

Abstract

 
Illegible handwriting in prescriptions remains a prevalent issue in fast-paced healthcare environments, contributing to misinterpretation and medication errors. Recognizing the importance of prescription readability, this study developed and deployed a real-time web application that classifies 20 generic medicine names using a Convolutional Neural Networks (CNN) model, making the system more accessible for real-world use. A dataset of 2,100 images was collected from public and private hospitals in Quezon City, and expanded to 6,720 images using data augmentation techniques such as brightening, blurring, and noise reduction. Some medicines, such as Chlorpromazine and Hydroxyzine, showed slightly lower performance, suggesting the need for more diverse data. The results demonstrate the model’s reliability and potential for integration into hospital systems or pharmacy management software, offering a practical solution to reduce errors in medication dispensing. Future work could involve expanding the dataset and integrating the model with OCR or electronic health record (EHR) systems to support broader handwriting variations and real-time clinical workflows.