Objective: The objective of this study is to propose a novel methodology for intracranial pressure (ICP) waveform subpeak identification by incorporating arterial blood pressure (ABP) and electrocardiogram (ECG) signals from patients who have undergone traumatic brain injury (TBI).
Methods: This approach consisted of 1) multimodal signal pre-processing and initial manual ICP waveform morphology labeling, 2) semi-supervised training of a support vector machine (SVM) ICP waveform morphological classifier, and 3) a dynamic time warping barycenter averaging (DBA) based ICP waveform template generation and derivative dynamic time warping (DDTW)-driven ICP waveform subpeak mapping from template to incoming processed waveforms.
Results: This proposed framework was evaluated on 30,000 ICP waveforms and resulted in an overall subpeak identification accuracy score of 98.2%.
Conclusion: The results showcased an improvement over existing methodologies and showed resilience to variations in ICP waveform morphologies from patient to patient due to the incorporation of a subject matter expert (SME) to accommodate new and unseen ICP morphologies.
Significance: The robustness of this comprehensive approach enabled the analysis of ICP morphological features over time to provide clinicians with crucial insights regarding the development of secondary pathologies in patients and facilitate monitoring their physiological state.