Abstract

There is significant interest in applying machine learning (ML) to data routinely collected in Intensive Care Units (ICUs) for clinical research aimed at improving healthcare. This data is crucial for predicting adverse events, ensuring timely treatment, and maintaining stable conditions for patient recovery. The broad availability of routinely collected data suggests that ML tools designed for this purpose could benefit a wide audience.

However, employing ML in this context presents challenges due to data quality issues and the need for model explainability. In this talk, we will present a novel ML approach that predicts medical events through learned latent representations of multivariate physiological time series using Long Short-Term Memory networks (LSTM) and a measurement of similarity in the patients' events. This method introduces a new way to predict events, particularly effective when high-resolution features are unavailable.

Comparisons with the state of the art demonstrate that this approach is more resilient to class imbalance compared to conventional methods. Our technique improves the analysis of physiological time series data routinely collected from ICU patients, contributing to more precise and reliable ML applications in healthcare.

Biography

Hollan Haule is a PhD student at The University of Edinburgh co-supervised by Dr Javier Escudero Rodriguez, Dr Chen Qin (now at Imperial) and Dr Milly Lo (NHS Lothian and Usher Institute). He received a BSc in Computer Engineering and Information Technology from University of Dar es Salaam, Tanzania and an MSc degree in Artificial Intelligence from University of Edinburgh, UK.