A Semi-automatic Indexing Pipeline for Medical Document Retrieval in Resource-constrained Settings

Abstract

Medical document indexing can benefit from both automation and human feedback. This research develops a semi-automatic indexing pipeline (SIP) for medical document retrieval in resource-constrained settings. The SIP includes an affordable and efficient automated process for preparing and indexing continuing medical education documents and a human feedback loop to validate recommended terms. It leverages pre-trained Named-entity Recognition models to identify appropriate terms from the MeSH vocabulary and higher-level subject terms from UMLS. The SIP achieved a precision of 59%, a recall of 64%, and an F1 score of 61% based on the expert evaluation of 124 distinct medical documents. The combination of automation with a human expert feedback loop demonstrates a model strategy for an affordable and practical approach to document indexing in resource-limited yet critical services. The SIP may be extended to other environments and information sources to improve the efficiency and accuracy of information retrieval.