COMPARATIVE EVALUATION AND MOBILE INTEGRATION OF VISION-ENABLED LLMS FOR HANDWRITING RECOGNITION IN MEDICAL PRESCRIPTIONS

Abstract

This study presents a comparative evaluation and mobile integration of modern vision-enabled Large Language Models (LLMs) for handwriting recognition in medical prescriptions. The research addresses the critical issue of illegible handwritten prescriptions in Indonesian healthcare by leveraging pretrained multimodal models that inherently combine visual and language processing capabilities. Unlike traditional approaches that require extensive manual preprocessing and feature extraction, the proposed system utilizes state-of-the-art LLMs to directly interpret handwritten text with minimal intervention. A series of experiments are conducted using a diverse dataset of handwritten prescription images, where various LLMs are benchmarked on evaluation metrics such as accuracy. The comparative analysis reveals significant variations in model performance, highlighting the robustness and efficiency of certain models in handling the complexities of diverse handwriting styles and localized medical terminology. Furthermore, the integration of the best-performing model into a mobile application demonstrates the practical viability of the approach, offering improved accessibility and streamlined workflow for both healthcare professionals and patients. These findings not only underscore the potential of vision-enabled LLMs in medical document processing but also pave the way for broader adoption in clinical settings.