Analysis of intracranial pressure recordings: Comparison of PCA and signal averaging based filtering methods and signal period estimation

• Main Authors: Calisto, A, Galeano, M, Bramanti, A, Angileri, F, Campobello, G, Serrano, S, Azzerboni, B
• Format: Conference Proceeding
• Language: English
• Subjects: Eigenvalues and eigenfunctions; Filtering algorithms; Finite impulse response filter; Iterative closest point algorithm; Noise; Principal component analysis
• ISSN: 1094-687X; 1558-4615
• DOI: 10.1109/IEMBS.2010.5627420

Intracranial pressure monitoring is a common used approach for neuro-intensive care in cases of brain damages and injuries or to investigate chronic pathologies. Several types of noises and artifacts normally contaminate ICP recordings. They can be sorted in 2 classes, i.e. high-frequency noises (due to measurement and amplifier devices or electricity supply presence) and low-frequency noises (due to unwanted patient's movement, speeches, coughing during the recording and quantization noise). Thus, deep investigations on ICP components aimed to extract features from ICP signal, require a denoised signal. For this reason the authors have addressed a study upon the most common filtering techniques. On each ICP recording we have performed 4 configurations of filters, which involve the use of a FIR filter together with Signal Averaging filters or PCA based filters. Next step is period estimation for absolute minima detection. The results obtained by the algorithm for automatic ICP marking are compared to those ones obtained from manual marking (peaks are manually identified and annotated by a brain surgeon). The procedure is repeated varying the filters sliding window size to minimize the mean square error. The results show how the configurations FIR filter + Signal averaging provides smaller mean squared error (MSE=118.84[sample 2 ]) than the others 3 configurations FIR filter + PCA filter based (MSE=135.29-147.15[sample 2 ]).