Paper Details
Abstract
Medication non-adherence is a significant public health challenge in Vietnam, exacerbated by complex prescription protocols, limited patient education, and inadequate digital health infrastructure. Manual transcription of printed prescriptions introduces errors and increases patient burden, especially in resource-limited clinics where staff are overworked and patients may face health literacy challenges. This work introduces an advanced Optical Character Recognition (OCR) and Information Extraction (IE) system to automatically convert printed Vietnamese prescriptions into structured digital formats, supporting integration with medication reminder systems and electronic health records. The system uses state-of-the-art OCR models, including VietOCR and PaddleOCR. By tailoring language models for Vietnamese script and accommodating diverse prescription layouts, the approach ensures robust text recognition even with low-quality or noisy scans. An innovative edge computing framework distributes processing tasks across NVIDIA Jetson Nano devices, reducing reliance on central servers, lowering latency, and improving performance in areas with limited internet connectivity, such as rural clinics. Experimental results demonstrate that splitting prescription images with intentional overlap for parallel edge processing reduces total processing time by approximately 30% while maintaining high recognition accuracy. This approach addresses the unique challenges of digitizing handwritten and printed Vietnamese prescriptions, enabling practical, cost-effective deployment in underfunded healthcare settings. The proposed system seeks to improve patient adherence, reduce medication errors, and strengthen healthcare integration, advancing scalable, affordable eHealth solutions tailored for developing countries.