Characterization of intracranial pressure (ICP) signals

Abstract Intracranial pressure (ICP) monitoring is a mainstay of neurosurgical diagnostic and therapeutic procedures. With the development of telemetric monitoring devices in the last decade, ICP monitoring has become feasible in a broader clinical setting, with patients undergoing ICP monitoring with mobile equipment either in-hospital or in the home setting, where a larger variety of ICP waveforms exist. Currently, the identification of these waveforms, the so-called macro-patterns lasting seconds to minutes, is primarily based on visual inspection. This process is not only slow but also subject to investigator bias due to interpretation subjectivity and dependence on experience. The need for objective and more automated identification of these variations emerges as a potential tool for better understanding the physiological underpinnings of the patient’s clinical state.
This thesis, divided into three main objectives, presents a new methodology that serves as a foundation for future objective and reproducible macro-pattern identification in the ICP signal with the hope to better understand the morphological characteristics and distribution of these macro-patterns in the ICP signal and their clinical significance.
First, we establish the motivation for the need of algorithm development for the extraction of macro-patterns from the ICP signal as an insight into the brain function. The results show that the current terminology and descriptions of B-waves no longer adequately address the ICP waveforms found in the clinical practice today. Discrepancy also exists regarding their origin. Our results found that ICP B-waves, also observed in healthy subjects during sleep, are associated both with respiratory disturbances and vascular contribution of flow velocity in a limited frequency range.
Second, a new data quality pipeline is designed that integrates all data validation checks to ensure high data quality. This pipeline includes artefact removal based on empirical mode decomposition, which is able to handle the nonlinearity and nonstationarity properties of ICP signals. The method is applied to ICP signals before macro-pattern identification, to mitigate the possibility of artefacts masking the true signal.
Finally, a method based on k-Shape clustering is developed to identify the most encountered macro-patterns in ICP signals. We found a total of seven macropatterns—with varying occurrence and distribution—that describe our ICP signals. These results may be considered benchmarks for the discussed shape clustering method that will be used in our ongoing research. These building blocks together with additional retrospective data could allow the identification of more unencountered macro-patterns besides the seven
proposed in this dissertation.
In conclusion, this thesis proposes a new, objective, and more automated method to identify macro-patterns in ICP signals. It walks through all steps from initial ICP recording to the characterization of the ICP signal, including the validation of the data quality. Thanks to this method, disease entities are likely to be identifiable based on the internal distribution and weighting of specific ICP macropatterns. This information aims at optimizing both disease and treatment identification.