Explainable Prediction of Medical Codes through Automated Knowledge Graph Curation Framework

Medical Coding (MC) converts medical diagnoses, procedures, equipment, and services into alphanumeric codes. Automated Medical Code prediction systems use Machine Learning and Deep Learning techniques. They could save insurance companies, Government agencies, and medical staff from the hassle of reading lengthy summaries to prepare insurance claims for reimbursement of expenses and fees. If not wholly correct, these machine-understandable codes can still help the medical coders to ease their task of predicting accurate medical codes. Despite being helpful, labor-saving, and efficient, these decision support systems fizzle out due to scarcity of domain knowledge. Knowledge Graphs containing a network of concepts, their meta-data, and a hierarchy of relationships in specific domains can help build the applications of Computer Assisted Coding (CAC) with much more accurate and precise results glued with Explainability. Unfortunately, domain-specific Knowledge Graphs (KG) creation is a grueling task, as it requires healthcare experts' intervention to annotate the medical concepts. Our proposed approach is to create a framework for the Automated Generation of Knowledge Graphs, which is yet a manual and time-consuming process, with the help of Natural Language Processing, Ontology-based information retrieval, and a semantic enrichment process. The created domain-specific Knowledge graph is then fused with a Deep Learning model. Predictions made by the Deep Learning model were compared before and after the consolidation of the domain-specific Knowledge Graph. We created a web-based application to save users from the complexity of the Deep learning model and knowledge graphs. Results proved the significance of the combination of Knowledge graphs and Artificial Intelligence. Produced results yield increased precision and accuracy with fewer false-positive results.