Semantic-based approach for predicting venous thromboembolism using Kohonen self organized map neural network

摘要

This paper presents a novel semantic-based approach to analyze clinical documentation in order to predict the first occurrence of symptomless unprovoked venous thrombo-embolism (VTE) in patients. The goal of this work is the attempt to save lives in cases of symptomless unprovoked VTEs that could be fatal from first occurrence. Using the Unified Medical Language System (UMLS) and MetaMap API, our semantic approach extracts hidden factors from unstructured text that might be very critical in the diagnosis or identification of potential risks of having a VTE. We use Kohonen self-organizing map to predict a patient's potential to develop VTE based on the similarity between the identified hidden risk factors in the clinical notes section from their medical records and the clinical notes used in the learning phase. Our system achieved 80% accuracy with 50% precision in prediction.