Vitals embedding for intensive care units

Abstract

Intensive Care Units (ICUs) collect vast amounts of data, including patient vitals, often recorded as time series data. This data can be analysed to extract meaningful data, aiding in timely intervention and improving patient outcomes. The complexity and quantity of this data also make it impossible to analyze manually. Traditional statistical methods are not adequate enough to extract important information and recognize patterns. Deep neural network-based language models however, excel at recognizing patterns. This project leverages this to train a BERT-based [1] transformer, procure the embeddings from this transformer and perform Shock Prediction task.