Medicine Recommendation System For Diabetes Using Prior Medical Knowledge
Mulubrhan Ayalew Wedagu, Dehua Chen, Muhammad Ather Iqbal Hussain, Tsegay Gebremeskel, Mayugi Tanguy Orlando, Arslan Manzoor · 2020
Medicine recommendation system could assist doctors in making an accurate diagnosis and predict medicine. According to CDC reports, more than 250,000 people die by medication error; for that matter, many deep learning models were previously proposed to solve the high order correlations of diabetic medicine for diabetic patients. However, they are failed to address the benefit of prior medical knowledge from experienced doctors. This study proposes a recommendation method called (Diabetes Medicine Recommendation System) DIMERS model, which combines a prior medical knowledge of doctors with bidirectional Long Short-Term Memory (BiLSTM). In this study, DIMERS, first, preprocesses general tests of diabetic patient test results and medicines with a complex data achievement strategy. After that, we use a weighted block with prior medical knowledge to enhance the learning and explainable abilities of deep neural networks (DNN). Our study's overall performance was excellent. We train and validate using the real world Ruijin hospital and drug bank datasets that contain 25,280 patients within three months gaps and 63 types of diabetes medicines.