Medical Prescription Recognition using Machine Learning

• Published in: 2021 IEEE 11th Annual Computing and Communication Workshop and Conference (CCWC) pp. 0973 - 0979
• Main Authors: Hassan, Esraa, Tarek, Habiba, Hazem, Mai, Bahnacy, Shaza, Shaheen, Lobna, Elashmwai, Walaa H.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.01.2021
• Subjects: Biomedical imaging; Classification; CNN; Conferences; Feature extraction; Handwriting recognition; Machine learning; Medical services; OCR; Optical distortion
• DOI: 10.1109/CCWC51732.2021.9376141

Admittedly, because of how busy doctors are nowadays, they tend to scribble unreadable prescribed medicines which leads to the problem of misinterpreting medicine names. Patients are sometimes curious to know information about their prescribed medicines before purchasing them. Recently, developers have been searching for a method to address this problem efficiently but, no technique leads to full recognition of medicine names due to the bad handwriting of doctors and its variety so that leads us to machine learning where the system will learn various types of handwritings for the same medicine to be able to recognize new handwritings. This paper proposed a system that presents a solution for both the pharmacist and the patient through a mobile application that recognizes handwritten medicine names and returns a readable digital text of the medicine and its dose. The System identifies the medicines' names and the doses for the collected data set with some, pre-processing techniques like image subtraction, noise reduction, and image resizing. After that, the pre-processed images will undergo some processing as it will be classified and feature extracted through Convolutional Neural Network and finally Optical Character Recognition technique applied on the medicines with low accuracy in the post-processing phase to identify their names by comparing the result with the dataset containing all the medicines. This will help in diminishing the instances of distortion of medication names assisting pharmacists in limiting their doubts. The proposed system tested on different real cases, and accuracy has reached 70% using (CNN) model.