Natural Language Processing for Diagnosis and Risk Assessment of Cardiovascular Disease

Abstract

Even with better tools for diagnosis and prevention, heart disease continues to be a major killer. Diagnosis and prevention of heart disease require detection and awareness of risk factors. Disease progression modeling and clinical decisionmaking can both benefit from the automatic identification of these risk markers from clinical records. Many studies have tried, without success, to uncover the characteristics that put people at risk for cardiovascular disease. Many hybrid systems have been suggested that attempt to integrate knowledgedriven and data-driven approaches, however due to the usage of Machine Learning, rules, and dictionaries, methods, they frequently require substantial human input. Natural language processing and deep learning methods can be used to extract useful information from clinical tales. By using state-of-the-art methods involving stacked word embeddings to the 2014 i2b2 challenge, this article hopes to contribute to the state of the art in this area. The suggested model makes use of a stacking method that combines different embeddings (CHARACTERBERT Embedding) to better understand the data. Our model shows substantial improvement after using this strategy to the i2b2 heart disease risk factors challenge dataset. An F1 score of 92.65% is an excellent result. When compared to competing models and systems created for the 2014 i2b2 challenge, our proposed model fares exceptionally well.

Tasks

Reproductions