Devasish Mahato, Disha Dudhal, Dhanashree Revagade and Yesoda Bhargava presented a research paper on "A Method to Detect Inconsistent Annotations in a Medical Document using UMLS" in 11th Forum for Information Retrieval Evaluation 2019 (FIRE'19) held at Indian Statistical Institute (ISI) Kolkata from December 12-15, 2019.
Abstract of the paper: Information retrieval from clinical documents relies heavily on annotated corpus. Any inconsistency in annotations, in form of heterogeneous annotations for similar concepts, could be detrimental to the quality of information retrieved. In extreme cases, this may lead to incorrect deductions about patient's medical history and result in erroneous decisions. In the present work, a complete end-to-end system that identifies inconsistencies in a clinically annotated document is presented and analysed. Unified Medical Language System (UMLS) is used to identify medical concepts in the clinical document. The output is presented in a clustered format wherein, each cluster identifies a unique medical concept and contains its semantic synonyms. For each semantic synonym, inconsistent annotations and the sentences in which they occur in the document are listed. The work could be useful for annotation experts who need automated tools to verify their work.
The full paper can be found on the link