Detection and Removal of Non-Medical Reports from User Provided Paper-Based Documents for Generating an Efficient EHR System

The global healthcare landscape is rapidly transitioning to Electronic Health Records (EHR) systems. In this digital evolution, individuals store their historical medical records in both analog and digital formats. However, as EHR technology gains traction, users often inadvertently upload unrelated documents alongside their medical records. This unintended inclusion of irrelevant data introduces noise into the EHR databases, leading to decreased accuracy when employing machine learning models. While extensive research focuses on medical data extraction and analysis, the critical task of effectively filtering non-medical documents remains relatively under-explored. This study presents a simple machine learning and deep learning approach to discriminate legitimate medical documents from user-provided uploads. Using machine learning algorithms such as Logistic Regression, Support Vector Machines (SVM), Gradient Boosting, Extreme Gradient Boosting (XGBoost), Bidirectional Encoder Representations from Transformers (BERT), and Clinical BERT, our study identifies BERT as the most effective tools for this task achieving an accuracy of 98%.