Title :
Predicting the occurrence of acute hypotensive episodes: The PhysioNet Challenge
Author :
Chiarugi, F. ; Karatzanis, I. ; Sakkalis, V. ; Tsamardinos, I. ; Dermitzaki, Th ; Foukarakis, M. ; Vrouchos, G.
Author_Institution :
Inst. of Comput. Sci., Found. for Res. & Technol. - Hellas, Heraklion, Greece
Abstract :
The PhysioNet Challenge 2009 addresses the prediction of acute hypotensive episodes (AHEs), which are serious clinical events since they could result in multiple organ failure and eventually in death. This objective is pursued with two different events: (a) event 1: the separation of records with critical AHE (subgroup H1) in the forecast window (FW), the one-hour period immediately following a specified time T0, from records from patients with no documented AHEs at any time during their hospital stay (subgroup C1) and (b) event 2: the separation of records with an AHE in the FW (group H) from records without any AHE in the FW (group C). Both events have been approached, using a subset of information common to the whole dataset, through the extraction of significant features from the last hours before T0 of the ABP and HR time series, linearly interpolated in the empty intervals and processed with a median filter for suppressing most artifacts. Decision tree classifiers based on these features have been designed for event 1 and 2, having better performances than classifiers based on support vector machine. The H1/C1 classifier (event 1) correctly classified all cases of the learning set (15 H1, 15 C1) producing also a perfect score on test set A. The H/C classifier (event 2) correctly classified 91.67% of the cases in the training set (30 H, 30 C) and obtained a score of 75% on test set B.
Keywords :
biological organs; blood pressure measurement; cardiology; decision trees; electrocardiography; feature extraction; medical signal processing; signal classification; support vector machines; ECG; PhysioNet Challenge 2009; acute hypotensive episodes; decision tree classifiers; feature extraction; forecast window; linear interpolation; median filter; multiple organ failure; support vector machine; time 1 h; training set; Classification tree analysis; Data mining; Decision trees; Feature extraction; Hospitals; Information filtering; Information filters; Support vector machine classification; Support vector machines; Testing;
Conference_Titel :
Computers in Cardiology, 2009
Conference_Location :
Park City, UT
Print_ISBN :
978-1-4244-7281-9
Electronic_ISBN :
0276-6547