Systems and methods for natural language processing-based classification of electronic medical records

Inventors

Verhoef, Kahla • Kwon, Michelle

Assignees

Egnite Inc

Abstract

Systems and methods of the present disclosure enable improved natural language processing of patient-related medical information for clinical decision support. To do so, a processor receives patient data including a written report, and accesses a dictionary of terminology associated with a disease. The terminology includes descriptors indicative of categories of the disease. The processor inputs the written report into a tokenization function to output tokens by parsing word patterns in the written report, and generating the tokens from the word patterns. The processor determines a presence in the written report of each descriptor based on the tokens and determines a category-specific score associated with each category based on the presence of the descriptors. The processor determines a category recommendation score indicative of a particular category based on the category-specific scores and generates a category recommendation representing the particular category based on the category recommendation score.

Core Innovation

The invention relates to an NLP-based clinical decision support method in which patient data comprising at least one written report associated with at least one patient is received via an electronic health record (EHR) system. The method accesses a dictionary of terminology associated with mitral regurgitation (MR), where the terminology includes a plurality of descriptors indicative of categories associated with MR, and the written report is parsed to produce a set of tokens from at least one word pattern.

The method inputs the tokens of the descriptors into a plurality of natural language processing pipelines, with each respective pipeline associated with at least one respective detected category and configured to separately and in parallel detect category-specific descriptors. The method determines category-specific scores using at least one category-specific rule parameter, and determines at least one recommended category score using at least one recommendation rule parameter based at least in part on the category-specific scores.

A category recommendation representing a particular category of MR is generated for the patient based at least in part on the recommended category score. The method causes a graphical user interface (GUI) associated with the EHR to display EHR data augmented with the category recommendation score to present to a user the category recommendation as clinical decision support for administration of transcatheter edge-to-edge repair (TEER), and TEER is administered to repair the mitral valve in response to the category recommendation for a particular category of MR.

Claims Coverage

The document explicitly provides one independent method claim, with dependent claims that refine processing of descriptors, define MR category structures and derived recommendations, and incorporate EHR display and TEER administration. Overall, the inventive features cover descriptor-token NLP parsing, parallel category-specific NLP pipelines, rule-parameter scoring and recommendation, EHR GUI augmentation, and TEER administration based on the recommendation.

The independent claim requires a method that converts descriptor terminology into tokens, uses multiple parallel category-specific NLP pipelines to detect MR category-specific descriptors, applies category-specific rule parameters and recommendation rule parameters to produce a recommended category score, generates a category recommendation, and then uses that recommendation score within an EHR GUI to provide clinical decision support for administering TEER to repair the mitral valve.

Stated Advantages

Documented Applications