Digitization and Classification of Prescription and Medical Reports of Patients Using Deep Learning

Abstract

In this paper, digitization and classification of medical reports and prescriptions of patients have been
proposed using deep learning technique. Two main contributions have presented in two stages. In first
stage, object character recognition technique has implemented for digitization of prescriptions and medical
reports. In second stage, a novel architecture of Convolutional Neural network has been proposed. The
training dataset has synthesized by collecting prescriptions and medical reports on paper from different
patients. For finding the suitable solution, both single-channel and multi-channel Convolutional Neural
network model (CNN) has been implemented. Experimental results show that the multi-channel
Convolutional Neural Network model with Word2Vec embedding offers maximum accuracy of 93%. The
digitization of medical documents using the proposed deep learning technique is useful for alerting patients
about timing of medication and keeps report safe for future use. It is also useful in multispecialty hospitals
for categorization of patient and administration.