XWH - 07 - 2 - 0118 TITLE : Early Support of Intracranial Perfusion

s Accepted or Presented since last Annual Report Prediction of MortalitySlaughter G, Kurtz Z, desJardins M, Hu PF, Mackenzie C, Stansbury L, Stein DM.2012 IEEE Biomedical Circuits & Systems conference, Nov 28-30, Taiwan Abstract— Real-time patient monitoring data collected over the course of trauma care are often large inquantities and require systematic representation that can combined temporal reasoning and validatedalgorithms to support clinical decision making. Available continuous vital signs, in many cases result infew state changes in the temporal range of interest; this work applies validated systematic algorithms to asmall set of vital signs data to identify patients at risk for mortality. Vital signs signals are used to train J48,naïve bayes, decision stump and SMO models. The InfoGain features selection algorithm was used toextract the best features using full run time-series data and feature generation permitted the features to betrained/tested on sensor data of any size, which dramatically improved the prediction classification of theJ48 algorithm. The evaluation of the models were done using leave-one-out cross validation. The quality ofthe classification was determined by the accuracy, precision and recall. Results show that the J48 algorithmcoupled with feature selection is a simple method for the identification of patients at increased risk formortality in trauma care. Real-time patient monitoring data collected over the course of trauma care are often large inquantities and require systematic representation that can combined temporal reasoning and validatedalgorithms to support clinical decision making. Available continuous vital signs, in many cases result infew state changes in the temporal range of interest; this work applies validated systematic algorithms to asmall set of vital signs data to identify patients at risk for mortality. Vital signs signals are used to train J48,naïve bayes, decision stump and SMO models. The InfoGain features selection algorithm was used toextract the best features using full run time-series data and feature generation permitted the features to betrained/tested on sensor data of any size, which dramatically improved the prediction classification of theJ48 algorithm. The evaluation of the models were done using leave-one-out cross validation. The quality ofthe classification was determined by the accuracy, precision and recall. Results show that the J48 algorithmcoupled with feature selection is a simple method for the identification of patients at increased risk formortality in trauma care. Exploiting Representational Diversity for Time Series ClassificationOates T, Mackenzie CF, Stein DM, Stansbury LG, DuBose J, Aarabi B, Hu P.ICMLA(International Conference on Machine Learning and Applications) 2012: Machine LearningEnsemble Methods and Applications track, December, Boca Raton, FL Abstract—More than a decade of research has produced numerous representations and similarity measuresto support time series classification and clustering. Yet most of the work in the field is so focused on therepresentation or similarity measure that it ignores the possibility of improving performance usingensembles of representations or classifiers. This paper explores ways of exploiting representationaldiversity for time series classification via ensembles of representations. We focuson the Symbolic Aggregate approXimation (SAX) discretization method coupled with the bag-of-patterns(BoP) representation because of their state-of-the-art performance in the single representation/classifier case. Experiments with a number of standard benchmark time series datasets and a new dataset ofvital signs collected from patients suffering from traumatic brain injury demonstrate the power of theensemble approaches, producing a single method that is often significantly better than vanilla SAX/BoPand compares favorably on a per dataset basis with the best methods reported in the literature for eachdataset.More than a decade of research has produced numerous representations and similarity measuresto support time series classification and clustering. Yet most of the work in the field is so focused on therepresentation or similarity measure that it ignores the possibility of improving performance usingensembles of representations or classifiers. This paper explores ways of exploiting representationaldiversity for time series classification via ensembles of representations. We focuson the Symbolic Aggregate approXimation (SAX) discretization method coupled with the bag-of-patterns(BoP) representation because of their state-of-the-art performance in the single representation/classifier case. Experiments with a number of standard benchmark time series datasets and a new dataset ofvital signs collected from patients suffering from traumatic brain injury demonstrate the power of theensemble approaches, producing a single method that is often significantly better than vanilla SAX/BoPand compares favorably on a per dataset basis with the best methods reported in the literature for eachdataset.