Machine learning for intensive care units

Abstract

Intensive Care Units (ICUs) generate extensive time series data, particularly patient vitals, holding immense potential for meaningful insights crucial for timely interventions and enhanced patient outcomes. Conventional analytical approaches fall short in unraveling the complexity and structure of this data, often missing vital patterns for effective prognostication. Leveraging the prowess of deep neural network-based language models, this project harnesses a BERT-based transformer [2] for Masked Language Modeling (MLM) [1] and Next Sentence Prediction (NSP) tasks, specifically targeting shock prediction. The embeddings derived from this transformative approach open avenues for further exploration and advancements in this critical domain.