Analysis of intracranial pressure pulse waveform in studies on cerebrospinal compliance: a narrative review

The earliest applications of pulse pressure analysis in compliance assessment were proposed in the late 1970s. Szewczykowski et al (1977) investigated the changes in pulse amplitude of ICP (AmpICP) associated with alterations in mean ICP to estimate cerebrospinal elastance. Based on the simplifying assumption that the cerebral fraction of cardiac stroke volume is constant in each heartbeat, and therefore AmpICP differs from elastance only by the constant factor 1/δ V (where δV denotes the intracranial volume increment in each cardiac cycle), the authors proposed to analyze elastance by calculating the slope of the amplitude–pressure (AMP–P) characteristic (figure 3(a)). The study showed that the AMP–P plot consists of two regions, representing the baseline state of nearly constant elastance and then a gradual increase in elastance, respectively. It has since been established that the AMP–P plot may contain one more region, matching the right-side deflection of the P–V curve at very high ICP, where AmpICP decreases again following the breakpoint associated with the collapse of cerebral blood vessels (Czosnyka et al 1996).

Zoom In Zoom Out Reset image size

Around the same time a study by Avezaat et al (1979) compared the changes in AmpICP and VPR at different mean ICP levels in dogs. The authors observed that AmpICP increases linearly with mean ICP up to the level of 60 mmHg, which led to the conclusion that the cerebral fraction of stroke volume can be assumed to be constant over a wide range of mean ICP and the AMP–P slope is an adequate measure of cerebrospinal elastance. Diverging trends in AmpICP and VPR over that threshold were attributed to a failure of cerebral autoregulation at highly elevated mean ICP. However, the authors noted that in order to use the AMP–P approach in clinical practice as an absolute measure of the volume–pressure relationship, the change in cerebral blood volume should be estimated for each patient at the beginning of the ICP recording. Moreover, the same group highlighted in a later study (Avezaat and Van Eijndhoven 1986) that the variability of pulsatile cerebral blood volume in humans is in fact quite high, meaning that the AMP–P slope cannot be taken as an absolute index of elastance without measuring the magnitude of change in cerebral blood volume in each cardiac cycle, a variable that is difficult to estimate and mostly ignored in studies on AmpICP.

Still, to date AmpICP is arguably the most frequently studied feature of the ICP pulse waveform (Wagshul et al 2011). Some authors propose to estimate AmpICP directly in the time domain, as the peak-to-nadir value of the pulsatile ICP signal (figure 4(a)) (Morgalla et al 1999, Eide 2006), while others employ spectral analysis to calculate AmpICP as the amplitude of the fundamental component of the signal's Fourier spectrum (figure 4(b)) (Czosnyka et al 1988). The two approaches have been said to produce strongly correlated results (Pearson correlation coefficient of 0.97) (Czosnyka et al 2007) and neither has been conclusively proven to offer more clinical benefit (Wagshul et al 2011), although some studies suggested that the frequency domain approach underestimates AmpICP when there is heart rate variability or waveform distortion (Holm and Eide 2008). It should also be noted that while in recent literature spectral AmpICP is almost exclusively calculated on 10 s fragments of the signal, the precise settings for the Fourier transform algorithm (e.g. window type and length) are not reported and there is no universal recommendation available.

Zoom In Zoom Out Reset image size

Over the last few decades changes in AmpICP have been investigated both in chronic conditions (Eide and Brean 2006a, Czosnyka et al 2009, Eide 2016) and acute brain insult (Czosnyka et al 1996, Eide and Sorteberg 2006, Eide et al 2011, Holm and Eide 2008, Hall and O'Kane 2016). In TBI, studies indicated that increased AmpICP is associated with severity of injury (Howells et al 2012) and worse outcome (Holm and Eide 2008). However, the validity of using elevated AmpICP as an indicator of reduced compliance in the clinical setting has been called into question. A study by Wilkinson et al (1979) noted that while increased AmpICP generally correlated positively with increases in mean ICP and intracranial volume, the large variability of responses rendered the parameter of limited clinical utility. It has also been shown in animals that the change in AmpICP during hypercapnia is related to cerebral arterial vasodilation rather than the steepening of the P–V curve (Portnoy and Chopp 1981), and there is a significant correlation between the slope of AMP–P characteristic and mean arterial pressure (Kaczmarska et al 2021), which highlights the importance of considering vascular factors in interpreting changes in AmpICP.

A different approach, proposed by Czosnyka et al (1988), is based on combining the information contained in both the P–V and AMP–P characteristics (figure 3(b)). The RAP index (whose name was derived from the common symbol for the correlation coefficient, R, and the words 'amplitude' and 'pressure') is defined as the moving Pearson correlation coefficient between mean ICP and AmpICP (usually expressed as 10 s averages) calculated over longer periods, usually 5 min. Positive values of RAP close to 0 suggest good compensatory reserve as the changes in AmpICP are not driven by oscillations in mean ICP (i.e. there is little to no correlation). Values close to 1 represent decreased compensatory reserve, with changes in mean ICP producing corresponding changes in AmpICP as expected in the steep part of the P–V curve. Finally, negative values are associated with disturbed cerebrovascular reactivity following the upper breakpoint of the P–V curve. While not a direct estimate of cerebrospinal compliance, RAP index serves as an indicator of the patient's 'working point' on the P–V curve and can be calculated continuously, therefore allowing for identification of patients with exhausted capacity for volume compensation.

Since its introduction in the 1980s, RAP index has been studied mostly in dysfunctions of CSF circulation (Kim et al 2009a, Weerakkody et al 2011), but also in relation to brain injury (Czosnyka et al 1994, Balestreri et al 2004, Timofeev et al 2008, Zeiler et al 2018b). The superiority of RAP index over other measures, primarily AmpICP, in clinical assessment of brain injured patients has been disputed due to its susceptibility to baseline effect errors, i.e. shifts in mean ICP that are either spontaneous or related to sensor location (Eide et al 2014, Hall and O'Kane 2016). It was noted that while AmpICP was nearly identical in two measurement sites, mean ICP changed substantially, producing differences in the RAP index large enough that they may alter patient management (Eide et al 2014). Another study demonstrated worse performance of the RAP index in terms of differentiating patients before and after surgery or thiopental treatment when compared to AmpICP (Howells et al 2012). However, RAP-weighted mean ICP (defined as ICP * (1 – RAP)) and sometimes termed 'true ICP') has been shown to outperform standard mean ICP in prediction of outcome after TBI (Czosnyka et al 2005, Calviello et al 2018, Zeiler et al 2019). Moreover, a similar approach was continued in defining the RAC index, a correlation-based measure calculated from AmpICP and cerebral perfusion pressure which theoretically includes information on both cerebrospinal compensatory reserve and cerebrovascular pressure reactivity (Zeiler et al 2018a), and the respiratory amplitude quotient (RAQ), calculated as the ratio of the amplitude of the respiratory wave to the amplitude of the respiration-induced variation in AmpICP that—unlike RAP—selectively reflects effects caused by respiration rather than all changes in mean ICP (Spiegelberg et al 2020).

In addition to providing an alternative method of calculating AmpICP (figures 4(a)–(b)), spectral analysis based on the Fourier transform was used to obtain a more detailed description of the changes the ICP pulse waveform undergoes at different mean ICP levels. Chopp and Portnoy (1980) proposed a systems analysis approach to characterize the ICP pulse waveform in relation to the ABP pulse, with the latter treated as input and the former as output of the cerebrospinal system. The authors reported enhanced transfer from ABP to ICP at increased mean ICP levels, suggesting that cerebral vasodilation which accompanies ICP elevation diminishes the ability of the arterioles to attenuate the arterial pulse. Takizawa et al (1987) investigated the ICP signal's Fourier spectrum at higher mean ICP and showed that the higher harmonics of ICP are reduced compared to the fundamental wave, leading to increased similarity between the ICP waveform and the simple sine wave. This finding was in line with the rounding of the ICP pulse observed in the time domain. Subsequent animal studies led to characterization of the intracranial system in basal conditions as a 'notch filter' attenuating the cardiac frequency more than the higher frequency components (Wagshul et al 2009) to protect cerebral vasculature from the highly pulsatile arterial input (Kim et al 2012). Reduction in cerebrospinal compliance has been shown to remove the 'notch filter' characteristics (Zou et al 2008). In TBI patients, Piper et al (1990) identified different types of transfer function spectra between ICP and ABP. Most importantly, they obtained contrasting results in patients with relatively low ICP, those with relatively low ICP but likely reduced vascular tone, and those with elevated ICP associated with vascular mechanisms. Generally, however, methods based on the transfer function from ABP to ICP (or the 'notch filter' properties of the system), although scientifically interesting, did not find wider application in clinical practice. The reason for this may lie in the fact that the relationship between ABP and ICP in the higher frequency range containing heart rate and its harmonics is by nature highly nonlinear, making both analysis and interpretation of the results very challenging.

A different approach was explored in research on the high frequency centroid (HFC), defined as the frequency-presented center of mass of the ICP power spectrum in the arbitrary 4–15 Hz range (Robertson et al 1989) (figure 4(c)). Bray et al (1986) examined the changes in HFC in relation to PVI and showed that HFC around 7Hz is considered normal whereas an increase up to 9 Hz corresponds to exhausted compensatory reserve. Later studies using HFC as a tool for continuous monitoring in TBI patients demonstrated that increased HFC correlates with higher mortality and occurrence of IH (Robertson et al 1989), and opposite direction of changes in HFC can differentiate transient from refractory hypertension episodes (Contant et al 1995). A different spectral centroid, called the higher harmonics centroid (HHC), was proposed to obtain a measure theoretically less dependent on heart rate than HFC (Berdyga et al 1993) (figure 4(d)). In contrast to HFC, which is calculated in the same frequency range regardless of the fundamental cardiac frequency, HHC attempts to take into account variations in heart rate by defining the analysis range in harmonic numbers (i.e. multiples of the fundamental frequency) rather than Hz. An increase in HHC can therefore be interpreted as an increase in the distortion of the waveform irrespective of heart rate. HHC calculated from the 2nd through 10th harmonic was shown to be positively correlated with mean ICP in TBI patients and to decrease significantly during ICP plateau waves (Zakrzewska et al 2021). Recently, assessment of both centroid metrics in the same database of TBI patients demonstrated that while elevated HFC is associated with mortality and poor outcome, HHC may potentially serve as an early warning sign of IH due to the occurrence of a breakpoint in the relationship between HHC and mean ICP at a level of approximately 25 mmHg (Uryga et al 2022).

However, it should be noted that the use of the Fourier transform requires that the signal in question satisfies the stationarity condition, meaning roughly that its spectral content remains unchanged over time. During continuous ICP monitoring, with frequent alterations in heart rate and rapid changes in the patient's condition, especially during the first days following acute brain insult, this requirement may not always be fulfilled, potentially making this approach imprecise for ICP analysis. Furthermore, spectral indices cannot be translated directly to a description of the waveform shape in the time domain, which presents a challenge for the clinician interpreting the results.

In the 1980s, Foltz and Aine (1981) proposed to analyze the upslope of the ICP pulse waveform defined as the pulse amplitude divided by latency of the initial ascending part of the pulse (i.e. from onset to the maximum). The authors noted that the difference in venous outflow between the inspiration (I) and expiration (E) phase of the respiratory cycle allows for the estimation of intracranial compliance related to venous volume. The I/E ratio was shown to differentiate hydrocephalic from nonhydrocephalic patients as well as specific types of hydrocephalus (Foltz et al 1990). In TBI patients, a positive correlation between the ICP slope and mean ICP was observed (Morgalla et al 1999, Westhout et al 2008). The slope (defined here as only the initial uptick of the waveform) was also demonstrated to decrease following decompressive surgery/evacuation and thiopental treatment (Howells et al 2012). However, despite its encouraging simplicity, this approach has not met wider application beyond the few cited studies.

As described earlier, under normal conditions the ICP pulse exhibits three distinct local maxima, called peaks P1, P2, and P3, whereas at elevated ICP the waveform becomes progressively more 'rounded' or 'monotonic', with the peaks gradually disappearing (Germon 1988). In 1983, Cardoso et al (1983) investigated the changes in the configuration of the characteristic peaks during alterations in mean ICP caused by different maneuvers. The authors noted that in addition to the previously known influence on AmpICP, hyperventilation-induced ICP reduction has a profound effect on the ICP pulse contour in the form of decreasing prominence of peak P2. It was suggested that since hyperventilation influences cerebral bulk volume (and with it, cerebrospinal compliance) through vasoconstriction of arterioles, the relative height of the first two characteristic peaks may provide an indirect measure of compliance (figure 5). Based on those observations, a study in normal pressure hydrocephalus (NPH) patients undergoing constant rate infusion studies compared changes in the P1/P2 ratio to the 'gold standard' method using a mathematical model of CSF dynamics (Kazimierska et al 2021a). It was demonstrated that all three approaches differentiate 'high' and 'low' compliance conditions associated with the baseline (low mean ICP) and plateau (high mean ICP) phase of the infusion test, respectively. Furthermore, resulting time trends of compliance estimators for individual patients were strongly correlated with each other, confirming the possibility of monitoring cerebrospinal volume compensation through analysis of the shape of the ICP pulse waveform.

Zoom In Zoom Out Reset image size

Meanwhile, the inverse P2/P1 ratio was proposed as a potential tool for prediction of impending ICP elevation. Studies in TBI patients demonstrated that the P2/P1 ratio is higher in patients who exhibit IH compared to those who do not (Fan et al 2008) and P2 elevation may be predictive of increased frequency of IH episodes (Mitchell et al 1997). Yet when used with a prediction threshold of 0.8, this approach showed low specificity, leading to the conclusion that elevated P2/P1 ratio alone is not a reliable indicator of increases in mean ICP as it was also present in the comparison data (Fan et al 2008). It should be noted, however, that the proposed threshold appears to have been chosen arbitrarily and possibly inappropriately. A P2/P1 ratio of 0.8 (equivalent to P1/P2 of 1.2) corresponds to a waveform with dominating peak P1 and slightly lower P2, which in turn corresponds to normal ICP; P2 can only be clearly said to dominate when the P2/P1 ratio exceeds 1. Accordingly, in the more recent studies on the P1/P2 ratio in NPH (Kazimierska et al 2021a, Ziółkowski et al 2021) almost all patients exhibited P1/P2 ratio < 1.2 (i.e P2/P1 > 0.8) at baseline and a reduction down to approx. 0.5 in the plateau phase. On the one hand, NPH and TBI are different clinical entities and patients in both groups may exhibit different pattern of changes in cerebrospinal compliance due to different underlying pathologies. On the other hand, one could argue that the apparent failure of the peak ratio method as a tool for IH prediction could be rectified by a more in-depth investigation of what constitutes an appropriate prediction threshold.

Nevertheless, while it has been confirmed that the ICP pulse waveform contains information about cerebrospinal compliance encoded in the height ratio of local maxima, accurate peak detection is a complex task due to the variety of pulse shapes observed in the clinical setting both in different patients and in the same patient over time (including also distorted waveforms that should be considered as artefacts but are not easily distinguished from valid pulses). A number of studies aimed to refine morphological analysis techniques in order to replace visual assessment of the waveform with reliable, repetitive automated tools. Hu et al (Hu et al 2009, but also Scalzo et al (2009, 2010), Asgari et al (2009)) developed an algorithm called Morphological Clustering and Analysis of Continuous Intracranial Pressure (MOCAIP) capable of assessing different pulse shape metrics, including peak height, latency, and curvature. Proposed as a generalized method of ICP pulse waveform analysis, MOCAIP-derived metrics have been used in a variety of scenarios, including prediction of ICP elevation (Hamilton et al 2009, Hu et al 2010) and identification of artefactual pulses (Megjhani et al 2019). In a small group of patients with slit ventricle syndrome, MOCAIP-based P2/P1 ratio was shown to correlate with enlargement of the lateral ventricles (Hu et al  2008). Various other attempts to solve the task of peak designation have also been reported (Calisto et al 2013, Elixmann et al 2012, Scalzo et al 2012, Lee et al 2016). Still, analysis of ICP peak configuration is yet to transition beyond the realm of research into the NCCU. The ICP signal is intricate and highly variable, and the computational algorithms proposed to date demonstrate varying levels of accuracy. Moreover, their technical complexity leads to limited understanding and acceptance in the medical community.

In order to mitigate the drawbacks of the peak detection method, a different approach based on machine learning was presented where peak identification is substituted with easy to interpret visual criteria for overall pulse waveform morphology. The concept has been presented in 1993 by Li et al (1993) who described 11 distinct types of ICP pulses in various pathological states and noted that changing configuration of characteristic peaks is associated with the progressive increase in mean ICP. In 2016, Nucci et al (2016) revived this concept by isolating four classes of ICP pulse shapes in NPH patients and proposing an artificial neural network capable of classifying the waveforms. The categories ranged from normal waveforms with three peaks arranged in a descending saw-tooth pattern to fully rounded pathological waveforms with only one defined maximum. Using the proposed classification method, the authors compared the dominant ICP pulse waveform class with other parameters describing CSF circulation during constant rate infusion of fluid. The results showed that pathologically altered pulse waveform morphology at baseline is indicative of disturbances in intracranial elastance revealed by the infusion test.

Recently, this method was adapted for application in continuous monitoring in brain injured patients using deep neural networks to classify ICP pulse shapes (Mataczyński et al 2022a). In this approach, the full long-term (i.e. spanning several days) high-resolution ICP recording is presented to the algorithm which first identifies individual pulse waveforms and then assigns each of them to one of the four classes proposed by Nucci et al (2016) or marks the pulse as an artifact and excludes it from further analyses (figure 6). The incorporation of pulse detection and artifact removal steps allows for the raw ICP recording to be processed in near real-time without manual inspection, potentially enabling continuous assessment of cerebrospinal compensatory reserve that could complement ICP measurement. Results of initial studies showed that TBI patients with unfavorable outcome present pathologically altered pulse shapes more often than those with good outcome (Kazimierska et al 2021b, Mataczyński et al 2022a) and taking into account the combination of increased mean ICP and pathologically altered ICP pulse shape enables more accurate mortality prediction in TBI patients than just elevated mean ICP (Mataczyński et al 2022b).

Zoom In Zoom Out Reset image size

However, straightforward use of the classification results at single pulse level requires simultaneous tracking of the occurrence of each of the four valid waveform types. Therefore, it is not easy to interpret by clinicians and leads to a loss of information on pulse shape variability over time. In order to present a simple, continuous index of compliance, a summary measure called pulse shape index (PSI) was proposed. PSI is calculated as the weighted average of class numbers and the percentage of pulses assigned to given class in a moving 5 min window. It is presented on a linear scale from 1 (only waveforms of class 1, indicating normal compliance) to 4 (only pathologically altered waveforms of class 4 associated with reduced compliance), and the removal of artifacts prior to analysis potentially improves the reliability of the calculations. Studies in a large, multi-center cohort of patients from the CENTER-TBI project showed that PSI is significantly elevated in patients who died after TBI even at relatively low mean ICP levels (below 15 mmHg) (Uryga et al 2022) while in patients who survived increases in PSI are only observed as mean ICP rises, in accordance with the P–V curve (unpublished data). Furthermore, elevated PSI was observed in patients with midline shift >5 mm and mass lesions >25 cm³ visible in computed tomography scans, indicating that this index reflects the volume imbalance in the intracranial space.

As described in the systems analysis approach (Chopp and Portnoy 1980, Takizawa et al 1987), pulsatile changes in ICP are believed to originate in arterial pulsations transmitted throughout cerebral vasculature. Due to the constraints of the skull boundary enclosing the intracranial space, direct measurement of ABP at the level of the brain is not routinely performed in humans for practical and ethical reasons, and ABP monitored peripherally differs significantly from the pressure registered centrally (O'Rourke 2009). However, transcranial Doppler (TCD) ultrasonography allows for noninvasive measurement of cerebral blood flow velocity (CBFV) in large cerebral arteries, primarily the middle cerebral artery, i.e. relatively close to the site of the ICP sensor, and therefore allows for the ICP signal to be investigated in relation to a different pulsatile component of the system.

A study by Kim et al (2015) in NPH patients analyzed spectral phase shift between CBFV and ICP recorded during the infusion test. The authors noted that at baseline, phase shift between the pulses was negative, with ICP trailing behind CBFV. Phase shift decreased further as mean ICP increased, indicating an increase in the time delay between the signals which may be attributed to progressive rounding of the ICP waveform. In the time domain, the relationship between the pulses was investigated using the ratio of pulse slopes (RPS), defined as the ratio of the inclinations from pulse onset to pulse maximum of CBFV and ICP (Ziółkowski et al 2021). It was shown that RPS decreases with increasing mean ICP, which corresponds to relatively small changes in the position of the systolic peak of CBFV but a substantial right-shift in the position of the maximum of the ICP pulse. Furthermore, both spectral phase shift and RPS at baseline exhibited significant negative correlation with elasticity revealed through analysis of the entire infusion test recording (Kim et al 2015, Ziółkowski et al 2021). This approach could therefore constitute a screening tool for reduced compliance that eliminates the need for volume manipulation. However, although TCD measurements are relatively widespread in the clinical setting, also in the NCCU, they require a trained operator and are difficult to perform over long periods, which limits the applicability of these indices in monitoring of acute brain injury patients in whom continuous compliance assessment would be of most benefit.

The CBFV signal has also been used to study relative changes in cerebral compartmental compliances: compliance of the cerebrospinal space (C_i ) and compliance of the cerebral arterial bed (C_a ) (Kim et al 2009b, 2012). C_i is calculated as the ratio of pulse amplitudes of changes in cerebral arterial blood bolume (CaBV) to ICP and C_a as the ratio of pulse amplitudes of CaBV to ABP, where CaBV is estimated from CBFV with a model of cerebral circulation (Avezaat and van Eijndhoven 1984, Kim et al 2009b). Due to the unknown cross-sectional area of the artery insonated to obtain CBFV, neither parameter can be calibrated to enable comparison between patients, and the observations are limited to relative changes over time. Nevertheless, it has been shown that this method accurately identifies the trend of decreasing C_i and increasing C_a during ICP plateau waves in brain injury patients (Kim et al 2009b, 2012) and decrease in both C_i and C_a during refractory IH (Kim et al 2012). Furthermore, changes in C_i during infusion test in NPH patients were strongly correlated with the 'gold standard' compliance estimated using the Marmarou model of CSF circulation (Kazimierska et al 2021a).

In the NCCU, invasive measurement of ICP using ventricular catheters or intraparenchymal microsensors remains the current standard (Evensen and Eide 2020). Nevertheless, the requirement for surgical intervention to place the sensor intracranially and the risk of complications such as infection or hemorrhage limit the use of ICP measurements to severe conditions. Over the years there have been numerous attempts to introduce noninvasive ICP (nICP) estimates to clinical practice and therefore make this modality applicable to a wider population of patients. Some of those methods, such as magnetic resonance or computed tomography imaging and measurements of optic nerve sheath diameter, have the inherent disadvantage of only being suited to one-off assessment, making them a tool for triage rather than continuous monitoring. Various pulsatile signals that can be recorded continuously and noninvasively have been considered as potential sources of nICP: ABP, TCD-based CBFV recordings, diffusion correlation spectroscopy (DCS) measures of cerebral blood flow, near-infrared spectroscopy measures of hemoglobin concentration and oxygentation, and tympanic membrane displacement. TCD-derived pulsatility index was among the early measures correlated with ICP (Bellner et al 2004), but later studies did not confirm its utility (Behrens et al 2010, Figaji et al 2009). More recently, DCS has been proposed as a valid alternative to TCD in estimation of critical closing pressure, a parameter directly related to ICP (Farzam et al 2017, Wu et al 2021). Detailed description of all of those approaches is beyond the scope of this review as most of them produce only steady-state values rather than reflect the dynamic cardiac-induced changes; we refer the reader to existing reviews on the subject of ICP measurement (e.g. Robba et al 2016, Evensen and Eide 2020) and will focus only on the methods that allow for continuous estimation of nICP pulse waveform.

Among the modeling approaches, there have been attempts to estimate nICP from a combination of pulse ABP and TCD characteristics with a 'black box' model (i.e. a model using a set of mathematical transfer functions rather than a description of underlying physiology) (Schmidt et al 2000), a physiological model of cerebrovascular dynamics (Kashif et al 2012), and a simplified model using only estimated central aortic pressure waveform as input (Evensen et al 2018a). However, while the first two studies (as well as a different investigation comparing the two approaches (Cardim et al 2016)) asserted that the model generates full nICP waveforms, only beat-averaged results were presented, leaving the question of the accuracy of nICP pulse-derived indices unanswered. In the model using central aortic pressure without additional TCD-based metrics both the nICP waveform and derived parameters were shown to be too inaccurate to aid patient management (Evensen et al 2018a). Studies on nICP estimation from a model of ultrasound pulse wave propagation through the human head (Petkus et al 2002, Ragauskas et al 2003) showed good agreement between invasive and noninvasive traces of ICP waveforms, but did not attempt to derive quantitative metrics.

Using a different approach that estimates pulse nICP from the measurement of tympanic membrane (TM) pressure, based on the assumption that ICP is transmitted to the membrane through the inner ear via infrasonic waves, Lang et al (2003) observed that noninvasive HFC is equivalent to that obtained by invasive intraparenchymal monitoring. However, a later study by Evensen et al (2018b) reported that the cochlear aqueduct appears to act as a low-pass filter, resulting in substantial differences in time domain metrics. Similarly, Dhar et al (2021) showed a substantial discrepancy between the TM pulsation and expected ICP pulse shape in illustrative examples.

Of the most recent studies, two approaches deserve particular attention. Monitoring of nICP via a mechanical sensor placed in contact with cranial skin that detects beat-to-beat micrometric deformations of the skull has been shown to produce pulse morphology metrics (specifically, peak height ratio and peak latency) in good agreement with the 'gold standard' ventricular sensors (Brasil et al 2021b, de Moraes et al 2022, Brasil et al 2023). This novel approach was studied in various pathological conditions; in acute brain injury, the authors demonstrated that noninvasively obtained P2/P1 ratio over 1.2 is predictive of IH in patients with preserved skull integrity (Brasil et al 2021b). Furthermore, they showed that patients after decompressive craniectomy present significantly higher nICP P2/P1 ratio (i.e. more impaired cerebrospinal compensatory reserve) than those with intact skull (Brasil et al 2021a), suggesting that their method is capable of detecting disturbed intracranial volume–pressure relationships. On the other hand, a different group reported that nICP features obtained with this technique do not reliably predict IH in TBI patients (Ballestero et al 2023), pointing to the need for more detailed investigations.

Concurrently, Spiegelberg et al (2023) developed a device for noninvasive capacitive measurement of the head's dielectric properties from electrodes placed on the scalp. A modeling study confirmed that changes in the dielectric properties of the head correlate to the cardiac and respiratory cycles associated with changes in the volumes of cerebral blood and CSF (Karimi et al 2023). The signal provided by the device, denoted W, was shown through experimental hyperventilation studies to be, at least partially, of intracranial origin (Spiegelberg et al 2023) and the amplitude of its cardiac-related oscillations was demonstrated to change during head-up and head-down tilt testing in a way that reflects the alterations in blood and CSF volume induced by changes in body position (Boraschi et al 2023). The authors concluded that W may therefore contain meaningful information on craniospinal compliance; however, they have not yet reported comparative analyses between invasive and noninvasive methods of compliance assessment. Due to the limited signal-to-noise ratio and the undetermined influence of extracranial factors they have also not attempted to describe the waveform shape quantitatively (e.g. through the P1/P2 ratio).

Nevertheless, the techniques that enable noninvasive ICP pulse waveform monitoring are a relatively recent development and larger prospective studies proving their reliability are likely still forthcoming.