| dc.description.abstract | The integration of digital health solutions into healthcare systems has become
increasingly vital, particularly in creating reminders for enhancing medication adherence
among patients. Optical Character Recognition (OCR) technology plays a crucial role in
automating the extraction of textual information from images, yet its application in
processing medical prescriptions remains underdeveloped. Current solutions often face
challenges related to the variability in printed prescription formats and the lack of patientoriented reminder creation tools, leading to inefficiencies in medication management. This
study addresses these limitations by developing a novel OCR system that leverages Google
Vision API and GPT-3.5-turbo to accurately recognize and extract prescription details
across diverse formats. Compared to other studies, this application excels in several areas:
Firstly, it can effectively detect both Vietnamese and English text, offering vital support
for Vietnamese users - a feature lacking in many existing applications. Secondly, the
unique combination of Google Vision and GPT-3.5 enhances the accuracy and versatility
of text recognition. The accuracy of the key-value pairs extraction model, when tested on
a dataset of 97 printed prescriptions, is 90.2%; on the other hand, when tested on 11
handwritten prescriptions, it is 25.7%. The average input time by using OCR (15.63s) is
substantially faster than manual input (241.5s). The evaluation conducted by 17 testing
users yielded an overall mean score of 3.29 out of 5.00 by using Mobile App Rating Scale
(MARS), with the app's functionality receiving the highest praise from the users. This
concludes that the scanning feature in Mediscan can effectively extract necessary
information from printed prescriptions to create a reminder, helping reduce the time of
manual inputting and human error. The contributions of this study are threefold: it
introduces an advanced OCR feature tailored for multilingual recognition, integrates this
feature into a user-friendly mobile application (Mediscan), and demonstrates superior
performance in terms of accuracy and processing time compared to existing solutions | en_US |
