[home] [Personal Program] [Help]

---

Session: Poster session I
Session starts: Thursday 24 January, 15:00

---

Guy Bogaarts ()
E. Gommer ()
D. Hilkman ()
V. van Kranen ()
W. Mess ()
J. Reulen ()

Abstract:
On Neonatal Intensive Care Units (NICU) many vital parameters are recorded but monitoring of brain function by Electroencephalography (EEG) is rare, mainly because signal interpretation requires expert visual inspection. In 1-6 % of newborns on the NICU (sub clinical) seizures occur and even more frequent in prematures and low-birth weight children [1]. Failure of detection and subsequent lack of treatment can result in brain damage. Automatic EEG analysis could enhance the application of NICU brain monitoring. Recently, a new seizure detection method was introduced using Support Vector Machines (SVM) [2]. Major aim of our project is to further optimize classification accuracy by introducing two pre-processing procedures. First, a Kalman filter was used to filter feature time series in order to reduce short false detections. Second, baseline feature correction was used to reduce inter patient differences. Data from 31 newborns aged between 0 and 6 months consisted of 46 single channel routine EEG recordings (average duration of 26 minutes) in which convulsions were annotated by visual inspection. A total of 122 features for neonatal seizure detection [3-4] are computed for 10s EEG epochs. Each epoch is represented by a feature vector that can either be used to classify the epoch or for training. Feature correction is performed using either all non-seizure data for training or a 3 minute baseline for testing. Kalman filtering is performed on test data only. This results in 4 train-test combinations: 1) no preprocessing, 2) baseline correction, 3) Kalman filtering, 4) baseline correction & Kalman filtering. By applying a threshold (T), classifier produced seizure probabilities can be transformed to binary decisions. Results show that baseline correction yields a significant increase in sensitivity from 47% to 59% (p<0.05). Sensitivity is further increased to 71% when Kalman filtering is used (p<0.05 ). Both pre-processing procedures together improved classification performance from 65% to 74% (p<0.05) correctly classified epochs. Why for some individual measurements performance decreases will be subject to future research.