An exploratory analysis of the incidence and morphological classification of artifacts in high-frequency pressure-flow and oximetry physiological signals in biomechanical neural injury

High-frequency physiological signals are crucial in guiding prognosis and diagnosis following neurological injury. Artifacts occurring during recording have long undermined the utility and accuracy of such signals in downstream clinical and research applications. Historically, artifacts have been retrospectively identified manually using a legacy data acquisition (DAQ) platform. In this study, we aimed to conduct an exploratory analysis of incidence and morphological classification of artifacts to better contextualize the scope of artifact incidence and compare legacy annotations to detailed annotations conducted using a custom-built platform. Manual annotation of artifacts in the Winnipeg Acute TBI dataset was conducted using a legacy annotation paradigm and a more detailed annotation method in intracranial pressure (ICP), arterial blood pressure (ABP), and bilateral regional oxygen saturation (rSO 2 ) signals collected from moderate-to-severe TBI patients. Incidence of general artifacts as well as defined sub-types were analyzed. A comparative analysis of the annotation between the two methods was also conducted. The median artifact incidence rate, as identified in the detailed annotations conducted using the custom-built platform across 154 traumatic brain injury (TBI) patients, was 4.23% (IQR: 1.85% – 11.16%) in ABP signals, 3.27% (IQR: 1.03% – 8.93%) in ICP signals for per patient median recording durations of 79.4 h (IQR: 37.6–132.9 h). Median artifact incidence was 4.25% (IQR: 0.60% – 11.43%) in viable left-sided rSO 2 signals ( n = 129), and 5.12% (IQR: 1.14% – 12.30%) in viable right-sided rSO 2 signals ( n = 126). Incidences of identified artifact sub-types were also reported. The comparison of annotation methods indicated that annotations conducted using the legacy paradigm produced comparatively lower annotation boundary precision leading to greater amounts of variability. This study represents a vital preliminary exploratory analysis of artifact incidence in high-frequency cerebral physiologic data. It highlights the ubiquity of artifacts in cerebral bio-signals, underscores the variation in artifact morphology, and emphasizes the need for reliable and more exhaustive annotation practices. Future work is required to bolster manual annotation accuracy and reliability, as well as to better define specific morphological artifact sub-types.