Hidden Markov models for analyzing medical time series in order to detect nosocomial pneumonia |
Pneumonia - as an inflammatory illness of the lung - is a dangerous and often fatal disease. A special subclass, the ventilator associated pneumonia (VAP), is affecting up to one fifth of the patients at Intensive Care Units (ICU). Based on a two years dataset, collected at a large ICU, we investigate a new method for time series processing in order to develop an early warning system for developing pneumonia. The system focuses on the preonset phase of the disease to extrapolate the future’s course. We utilized the functionality of Hidden Markov Modelsand the stacking paradigm to categorize and forecast given time series of a patient. Finally we demonstrate the benefits of our approach with a set of real patient data.
Marek Opuszko, Johannes Ruhland, Franziska Oroszi, Michael Hartmann, Martin Specht, Friedrich Schiller |