Comparison of Morphological and Wavelet Based Methods in Intracranial Pressure Signal Analysis

Analysis of Intracranial Pressure signal (ICP) to predict the abrupt increases is extremely important, because this sudden elevation can be life threatening and sign of secondary brain injury. Segments with three minute duration were then constructed from the ICP signal: One from the onset of amplitude elevation as part#0, with sets 5, 15, 25 minutes prior to the elevation as respectively, part#5, part#15 and part#25. Using Fourier transform, wavelet transform and dual tree complex wavelet transform, several features were extracted. Next, the most informative features are selected with regards to three different criteria: T-Test, Statistical Correlation, and Genetic Algorithm with the cost function of SVM classification error. Then these most informative features are classified using SVM. The accuracy rate of 73.38% was achieved. In the second method, using morphological analysis, ICP signal was segmented into several sub signals including two maximum points and two minimum points. Morphological features and Power spectrum sum of each sub signal were computed and classified using SVM. The accuracy rate achieved was 87.12%. The results prove that morphological analysis of ICP signal is more efficient.