Future of neurocritical care: Integrating neurophysics, multimodal monitoring, and machine learning

Abstract

Multimodal monitoring (MMM) in the intensive care unit (ICU) has become increasingly sophisticated with the integration of neurophysical principles. However, the challenge remains to select and interpret the most appropriate combination of neuromonitoring modalities to optimize patient outcomes. This manuscript reviewed current neuromonitoring tools, focusing on intracranial pressure, cerebral electrical activity, metabolism, and invasive and noninvasive autoregulation monitoring. In addition, the integration of advanced machine learning and data science tools within the ICU were discussed. Invasive monitoring includes analysis of intracranial pressure waveforms, jugular venous oximetry, monitoring of brain tissue oxygenation, thermal diffusion flowmetry, electrocorticography, depth electroencephalography, and cerebral microdialysis. Noninvasive measures include transcranial Doppler, tympanic membrane displacement, near-infrared spectroscopy, optic nerve sheath diameter, positron emission tomography, and systemic hemodynamic monitoring including heart rate variability analysis. The neurophysical basis and clinical relevance of each method within the ICU setting were examined. Machine learning algorithms have shown promise by helping to analyze and interpret data in real time from continuous MMM tools, helping clinicians make more accurate and timely decisions. These algorithms can integrate diverse data streams to generate predictive models for patient outcomes and optimize treatment strategies. MMM, grounded in neurophysics, offers a more nuanced understanding of cerebral physiology and disease in the ICU. Although each modality has its strengths and limitations, its integrated use, especially in combination with machine learning algorithms, can offer invaluable information for individualized patient care.

Key Words: Neurocritical care, Critical care, Multimodal monitoring, Machine learning, Neurophysics, Cerebral hemodynamics, Cerebral energetics, Transcranial Doppler, Cerebral microdialysis, Near-infrared spectroscopy

Core Tip: This manuscript provided a comprehensive review of multimodal monitoring (MMM) in the intensive care unit, emphasizing the integration of neurophysics to optimize patient outcomes. It covered invasive and noninvasive neuromonitoring tools and highlighted the role of machine learning in real-time data analysis and interpretation from MMM tools, aiding in precise clinical decision-making. By integrating diverse data streams through MMM, machine learning algorithms enhance the understanding of cerebral physiology and disease, offering invaluable insights for personalized patient care in the intensive care unit. This integration aids the neurointensivist in more accurate neuroprognostication and in future avenues for targeted therapeutic interventions.

INTRODUCTION

The practice of neurocritical care in the intensive care unit (ICU) has seen remarkable advances in recent years, particularly in the field of multimodal monitoring (MMM). Based on the principles of neurophysics, these advances have expanded our ability to assess brain function and metabolic status in real time[1]. Despite the increasing complexity and sophistication of available technologies, clinicians still face considerable challenges in selecting and interpreting the most appropriate combinations of neuromonitoring modalities to improve patient outcomes[2].

Traditionally, neuromonitoring in the ICU has largely been dependent on invasive techniques, such as intracranial pressure (ICP) monitoring, jugular venous oximetry (SjvO₂), oxygen monitoring of brain tissue, thermal diffusion flowmetry (TDF), cortical and depth electroencephalography (dEEG), as well as cerebral microdialysis (CMD)[3]. While these methods promise precise measurements, they come with associated risks, including infection and hemorrhage[4]. Recently, however, a shift has been made toward the incorporation of noninvasive modalities into the neuromonitoring landscape. These include transcranial Doppler (TCD), tympanic membrane displacement (TMD), near infrared spectroscopy (NIRS), optic nerve sheath diameter (ONSD), positron emission tomography (PET), and autonomic assessment by analysis of heart rate variability (HRV)[5-7].

Even as the number of available neuromonitoring tools increases, so does the complexity of interpreting the generated data. In this context, machine learning and other data science tools have emerged as promising adjuncts to traditional methods[8]. These algorithms can filter the high-dimensional data generated by MMM to provide clinicians with predictive models for patient outcomes, thus facilitating individualized treatment strategies[9].

The objective of this review was to provide a critical overview of the current state of MMM technologies focused on ICP, cerebral electrical activity, metabolism/energetics, and autoregulation. Their neurophysical basis and clinical relevance in an ICU setting were examined. Additionally, the growing use of machine learning will be discussed by exploring how it can be integrated into neuromonitoring practices to assist clinicians in making timely and accurate decisions for individualized patient care.

CEREBROSPINAL FLUID HYDRODYNAMICS

The labyrinthine architecture of the human brain is governed not only by cellular interactions but also by physical principles that dictate fluid dynamics, pressure gradients, and metabolic exchanges. Among these, the physics of cerebrospinal fluid (CSF) plays a central role in maintaining intracranial homeostasis and is deeply interconnected with the parameters that we aim to monitor and modulate in the neuro ICU[10]. Understanding the physics of CSF has direct implications for interpreting invasive and noninvasive neuromonitoring data, helping to optimize patient outcomes[11].

CSF serves as a buoyant cushion for the brain, facilitates waste elimination, and acts as a conduit for the distribution of neuroactive substances. However, changes in CSF dynamics can have devastating consequences, manifesting as elevated ICP, impaired cerebral autoregulation, and downstream neurological deterioration[12]. Therefore, understanding the physics of CSF can provide a more comprehensive view of cerebral pathophysiology, especially when integrated with other neuromonitoring tools. In this section, the physical principles that govern the flow and pressure of the CSF including, but not limited to, hydrostatics, hydrodynamics, and coupling between the CSF and cerebral blood flow (CBF) were discussed.

The choroid plexus is a specialized network of blood vessels and epithelial cells located within the ventricular system of the brain. Its primary function is the production of CSF. These vascular tufts are enveloped by a layer of epithelial cells that act as a selectively permeable barrier between the blood and the CSF. Through a combination of active secretion, filtration, and reabsorption processes, the choroid plexus maintains the composition and volume of the CSF, which is crucial for cushioning the brain, providing nutrients, and removing waste products. The choroid plexus is found in all four ventricles of the brain: the two lateral ventricles, the third ventricle, and the fourth ventricle[13]. The total volume of the brain is 1400 mL, and the CSF comprises approximately 150 mL. The CSF production rate is approximately 0.35 mL/min (500 mL/d)[12].

The Monro-Kellie doctrine is a fundamental concept in neurocritical care that describes the relationship between ICP, brain tissue, blood, and CSF. According to this principle, the cranium is a rigid, nonexpandable structure that contains these three components. Any increase in one component must be compensated for by a decrease in one or both remaining components to maintain a constant ICP. Failure to achieve this balance leads to elevated ICP, which can result in neurological deterioration[14]. Lumbar cisterns act as important buffer zones within this framework. Located in the subarachnoid space of the lumbar region, these cisterns are CSF reservoirs that can accommodate excess fluid. When there is an increase in ICP due to cerebral edema, hematoma, or other pathology, the lumbar cisterns can expand to receive additional CSF, thus acting as a compensatory mechanism to help stabilize ICP. Their role as a ‘safety valve’ highlights the complexity and adaptability of the intracranial compartment in the regulation of ICP and further underlines the clinical importance of understanding CSF dynamics in the treatment of neurointensive care patients[15].

The rate of CSF formation is defined by the following equation: CSFprod = infusion (C_i − C_o/C_o). In this equation, infusion is the rate at which a substance (often a tracer) is infused into the system, C_i is the concentration of that substance in the inflow (usually arterial blood), and C_o is the concentration in the outflow (often venous blood or CSF). The equation suggests that the rate of CSF production is influenced by the rate of infusion and the gradient between the inflow and outflow concentrations, normalized to the outflow concentration[16]. This mathematical representation provides a way to quantify the dynamics of the CSF and can be particularly useful in experimental settings or clinical evaluations where precise measurements are needed. Highlighting the importance of the interaction between substance concentrations and infusion rates in determining the rate of CSF formation, providing a quantitative framework to understand this crucial physiological process[17].

Ekstedt et al[18] in Sweden were the first to use a constant pressure infusion methodology to better examine the hydrodynamics of the CSF. A rectilinear CSF pressure-flow relationship was observed. Therefore, when the ICP is greater than the sagittal sinus pressure, the flow through the sagittal sinus increases[19,20]. Furthermore, the rate of CSF production does not remain constant with increasing age. In patients with hydrocephalus, increasing age is inversely proportional to CSF production[21]. Arterial pulsations play a crucial role in the dynamics of CSF flow within the brain. With each cardiac cycle, arterial pulsations generate pressure waves that propagate through the parenchyma and influence the movement of the CSF. These pulsatile forces act as a driving mechanism that helps circulate CSF through the ventricular system and into the subarachnoid spaces surrounding the brain and spinal cord. The rhythmic expansion and contraction of the arteries contribute to the to-and-fro motion of the CSF, facilitating its distribution and aiding in the overall homeostasis of the intracranial environment. Disruptions in arterial pulsatility, as seen in conditions such as hypertension or arterial stiffness, can potentially affect the dynamics of the CSF and may be involved in various neurological conditions. Arterial hypotension decreases the resistance of CSF outflow by approximately 17% per 50 mmHg. Furthermore, hypercapnia increases resistance of CSF outflow (approximately 18%; 27-48 mmHg)[22].

The glymphatic system is a recently discovered pathway for waste clearance in the brain, which also involves a novel understanding of CSF absorption. According to glymphatic theory, CSF flows through the perivascular spaces surrounding the arteries, enters the brain parenchyma, and mixes with interstitial fluid[23]. This mixing facilitates the removal of soluble proteins and metabolites, including harmful substances like amyloid-β. From there, the fluid is directed to perivenous spaces and ultimately drains into the lymphatic system. The glymphatic process is believed to be more active during sleep and is crucial to maintaining brain health[24]. This theory provides an alternative or complementary explanation to traditional models, which have emphasized the drainage of the CSF primarily through the villi and arachnoid granulations. The role of the glymphatic system in the absorption of CSF within the parenchyma offers new avenues to understand and treat neurological diseases related to waste accumulation and fluid balance. Despite increasing interest in recent years given the discovery of meningeal lymphatic vessels, further mechanistic studies are needed to validate the hydrodynamics of the CSF within the glymphatic model[25].

CBF AND PERFUSION

The cerebrovascular tree, with its intricate and multifaceted architecture, is structured to efficiently supply blood to the various regions of the brain. The circle or polygon of Willis, a crucial arterial polygon at the base of the brain, acts as the primary hub for cerebral blood distribution, connecting the anterior and posterior blood supplies and providing collateral circulation, which is essential in cases of arterial occlusion. Branching out from the main arteries are smaller arteries and arterioles that penetrate the brain parenchyma, forming extensive capillary networks, where nutrient and gas exchange occurs. Subsequently, these capillaries converge into venules and veins, culminating in larger venous structures that facilitate the drainage of deoxygenated blood away from the brain[26].

Angelo Mosso (1846-1910), a distinguished Italian physiologist of the late 19th century, contributed significantly to our understanding of CBF and metabolism. Mosso was one of the pioneers who explored the dynamics of blood circulation in the brain, laying the foundational knowledge for modern neurophysiological research[27]. He ingeniously designed experiments in neurosurgical patients with skull defects and developed early instruments to investigate variations in cerebral blood volume and pressure in response to different physiological and psychological stimuli. Mosso invented the ‘human circulation balance,’ a noninvasive measure of CBF that demonstrated that emotional stressors and intellectual activity were associated with increased arterial pulsations of the brain. Mosso’s meticulous research demonstrated the relationship between cerebral activity and changes in blood flow, suggesting the concept of neurovascular coupling[28]. Blood flow within cerebral vessels is governed by the principles of fluid dynamics. According to Poiseuille’s law, the flow (Q) through a cylindrical tube is proportional to the fourth power of the radius (r) and the pressure gradient (ΔP) and is inversely proportional to the length (L) and viscosity (η) of the fluid[29]. Mathematically, this is represented as: Q = (πr⁴ ΔP)/8ηL

This equation illustrates how small changes in the radius of the vessel can have profound effects on blood flow, emphasizing the importance of vascular tone regulation in maintaining adequate cerebral perfusion. Furthermore, the Reynolds number, which quantifies the likelihood of laminar vs turbulent flow within a vessel, is pivotal in understanding the flow characteristics in cerebral vessels. Cerebral autoregulatory mechanisms work diligently to ensure constant blood flow despite fluctuations in systemic blood pressure, by adjusting vascular resistance through vasodilation and vasoconstriction of the arterioles. Any disruption in these mechanisms can lead to pathological conditions, such as hyperemia or ischemia, that can compromise the integrity of brain tissue and its function. In a resting awake state, the human brain uses O₂ at a rate of 35 mL/min/kg[30]. This can be mathematically demonstrated by the following equation: cerebral metabolic rate of oxygen = CBF × arteriovenous difference of oxygen[31].

AUTOREGULATION OF CBF

Danish scientist and professor Mogens Fog (1904-1990) in 1938 directly studied the pial vessels in cats and their response to various stimuli. He observed that a decrease in blood pressure led to immediate vasoconstriction and secondary dilation of the pial vessels, and an increase in blood pressure caused immediate vasodilation and secondary vasoconstriction[32]. In 1959, Lassen[33] illustrated the well-known cerebral autoregulation curve and conceptualized a wide mean arterial pressure (MAP) range (approximately 60-150 mmHg) in which CBF remains constant. Lassen’s illustration of cerebral autoregulation continues to be popularized in high-impact research articles and educational textbooks (Figure 1).

Figure 1 Lassen’s cerebral autoregulatory curve. Cerebral blood flow is normally reported to remain constant over a large mean arterial pressure range of 50-150 mmHg. This range is significantly shifted or narrower in patients with acute brain injury. The ischemic critical perfusion threshold is defined at 30 mL/100 g/min. This figure was generated using BioRender (Agreement No: SE269MFWGB).

However, Lassen’s autoregulation curve has recently been under scrutiny due to its inaccuracy and heterogeneity between individuals[34]. Furthermore, Lassen’s original research on cerebral autoregulation is criticized due to its small sample size, single data point analysis, and insufficient reporting on error margins. However, Lassen’s pioneering work provided a foundational framework for understanding cerebral autoregulation. Czosnyka et al[35] demonstrated this autoregulation curve by performing a daily TCD analysis of CBF in patients with traumatic brain injury (TBI). They demonstrated both a lower and an upper limit of autoregulation and a plateau, as initially shown by Lassen. However, contrary to the wide range claimed by Lassen, the curve experimentally shown here was only approximately 40 mmHg. This relatively small plateau in the autoregulation curve is believed to be attributed to the acute brain injury suffered by these patients.

The cornerstone principle of cerebral autoregulation states that CBF remains constant despite fluctuations in systemic blood pressure, ensuring that the delicate and complex neural network receives constant nourishment and oxygen supply. This intricate mechanism is orchestrated by a symphony of myogenic, chemical, neuronal, and metabolic responses that dynamically allow cerebral arteriole vasodilation and constriction. Kontos et al[36] demonstrated both cerebral vasodilation and constriction in response to fluctuation of systemic blood pressure. Interestingly, stimulation of the vagus nerve allowing increased baroreceptor efferent activity and a subsequent drop in MAP was met with a phase shift of approximately 10 s in the corresponding increase in the diameter of the pial arteriolar vessels. In a subsequent experiment, successive infusions of ATP led to oscillatory episodes of systemic hypotension met with a corresponding oscillatory but phase-shifted response in the pial arteriolar diameter.

MAP is the average pressure in the arteries of a patient during one cardiac cycle. It is a function of cardiac output (CO) and systemic vascular resistance (SVR): MAP = CO × SVR + central venous pressure. However, since central venous pressure is usually much lower than MAP (and often near zero in healthy individuals), it is often omitted in the calculation, simplifying it to: MAP = CO × SVR. MAP is also commonly estimated using systolic and diastolic blood pressure (DBP): MAP ≈ DBP + 1/3 (systolic blood pressure - DBP).

ICP is the pressure exerted by fluids (such as CSF) and tissues within the skull. An elevated ICP can reduce CBF. The cerebral perfusion pressure (CPP) is the net pressure gradient that drives oxygen and nutrients to the cerebral tissues. It is the pressure available to ensure blood flow through the cerebral vasculature. Since blood must flow from the systemic circulation (represented by MAP) into the intracranial space (where the pressure is ICP), the effective pressure driving this flow (CPP) is the difference between these two pressures: CPP = MAP - ICP.

Myogenic/endothelial response

The myogenic response is a fundamental component of autoregulation, where vascular smooth muscle cells in the arterioles respond to changes in intravascular pressure. Elevated pressure induces vasoconstriction, reducing blood flow, while decreased pressure triggers vasodilation, increasing blood flow[37]. This inherent mechanism is vital to prevent overperfusion or underperfusion of the brain, thus safeguarding neural integrity. Nitric oxide (NO), endothelium-derived hyperpolarizing factor, and prostacyclin are endothelial factors and potent vasodilators. Endothelin 1, thromboxane A2, and angiotensin II are endothelial factors that allow for vasoconstriction of the arterioles[38,39]. NO is synthesized by endothelial NO synthase. NO diffuses to smooth muscle cells in the vascular wall and activates guanylate cyclase, leading to increased cyclic guanosine monophosphate and vasodilation, thus modulating CBF. Prostacyclin is another important vasodilator synthesized by endothelial cells. It activates adenylate cyclase in smooth muscle cells, leading to increased cyclic AMP levels and subsequent relaxation of the vascular smooth muscle. Endothelium-derived hyperpolarizing factor hyperpolarizes vascular smooth muscle cells, leading to relaxation and vasodilation, although its exact identity and mechanism of action are still being researched[38]. Endothelin 1 is a potent vasoconstrictor produced by endothelial cells. It acts mainly through the ET-A receptor on vascular smooth muscle cells, leading to increased intracellular calcium and smooth muscle contraction[40]. Thromboxane A2 synthesized by endothelial cells acts as a vasoconstrictor by promoting the mobilization of intracellular calcium in vascular smooth muscle cells[41]. Angiotensin II produced through the renin-angiotensin system induces vasoconstriction by activating the angiotensin I receptor in vascular smooth muscle cells. Coordinated action of vasodilators and vasoconstrictors, in response to changes in transmural pressure, ensures the maintenance of CBF within a certain range of perfusion pressures, protecting the brain from hypoperfusive and hyperperfusive events[42,43]. Additionally, magnesium may play an important role in maintaining endothelial homeostasis within the cerebral vasculature and is implicated in cerebral arteriopathies such as reversible cerebral vasoconstriction syndrome and posterior reversible encephalopathy syndrome[44].

Chemical/metabolic response

Chemoregulation, another crucial aspect, is primarily driven by changes in the concentration of carbon dioxide, hydrogen ions, and oxygen within the cerebral vascular smooth muscle. Carbon dioxide has a profound vasodilatory effect on cerebral blood vessels. Elevated levels of carbon dioxide in the blood, or hypercapnia, lead to decreased blood pH (respiratory acidosis) and prompt dilation of the cerebral arterioles, which in turn increases CBF to meet the increased metabolic demand. This mechanism is particularly important during states of increased cerebral activity or altered respiratory states, where increased carbon dioxide production or retention requires increased blood supply to the brain[45]. On the contrary, decreased levels of carbon dioxide or hypocapnia induce vasoconstriction of the cerebral vessels, reducing CBF, which can lead to decreased oxygen supply to brain tissue.

The general rule of thumb is that for every 1 mmHg change in arterial carbon dioxide tension (PaCO₂), CBF changes by approximately 1-2 mL/100 g/min. This relationship is usually linear within the physiological range of PaCO₂ (approximately 20-80 mmHg). The mathematical expression for this relationship is often given as: ΔCBF = k × ΔPaCO₂. Where: ΔCBF is the change in CBF (in mL/100g/min), ΔPaCO₂ is the change in PaCO₂ (in mmHg), and k is a constant that represents the sensitivity of CBF to changes in PaCO₂, which typically ranges between 1-2 mL/100g/min per mmHg. It is important to note that this relationship can be altered in certain pathological conditions, and the response of CBF to PaCO₂ can be blunted in patients with chronic hypercapnia or significant cerebrovascular disease.

Oxygen, on the other hand, exerts the opposite effect on the cerebral vasculature compared to carbon dioxide[46]. Hypoxia, or low oxygen levels, induces vasodilation of cerebral blood vessels to improve the oxygen supply to brain tissue. This is a crucial adaptive response to ensure the viability of neural cells during states of decreased oxygen availability. On the contrary, hyperoxia or elevated oxygen levels induce vasoconstriction of the cerebral blood vessels, leading to a reduction in CBF. This mechanism prevents hyperperfusion of brain tissue and contributes to maintaining optimal oxygen delivery to the brain. The interaction between oxygen and carbon dioxide in cerebral autoregulation is a delicate and dynamic balance, ensuring optimal blood supply, nutrient delivery, and the removal of brain waste[47].

The relationship between arterial oxygen tension (PaO₂) and CBF is less straightforward than the relationship between PaCO₂ and CBF. However, there is a general understanding of how changes in PaO₂ affect CBF, although the specific quantitative relationship can vary between individuals and under different physiological and pathological conditions. In general, CBF remains relatively stable over a wide range of normal PaO₂ levels. However, when PaO₂ falls below a critical threshold (usually around 50-60 mmHg), CBF begins to increase significantly as a compensatory mechanism to maintain adequate oxygen delivery to the brain[48]. This relationship is generally nonlinear and can be represented as follows: CBF ∝ [1/(PaO₂ × n)]. Here, n is a constant that determines the sensitivity of CBF to changes in PaO₂. This equation suggests that as PaO₂ decreases, CBF increases, but the rate of change in CBF accelerates as PaO₂ drops below the critical threshold. The exact value of n can vary and is typically determined empirically. In general, CBF remains relatively stable over a wide range of normal PaO₂ levels. However, when PaO₂ falls below a critical threshold (usually around 50-60 mmHg), CBF begins to increase significantly as a compensatory mechanism to maintain adequate oxygen delivery to the brain[49].

Disruption in balance due to alterations in arterial blood gases can affect brain metabolism and function. Understanding the interaction between oxygen and carbon dioxide in the modulation of cerebral vasculature is crucial to managing conditions that alter CBF and pressure dynamics, such as TBI, stroke, and various respiratory and metabolic disorders. The careful monitoring and manipulation of arterial carbon dioxide and oxygen levels can be therapeutic strategies in conditions where cerebral autoregulation is impaired[50].

Neuronal response

This regulation, often termed neurovascular coupling, involves a sophisticated interplay between neurons, astrocytes, and blood vessels. Neurogenic regulation involves the intricate network of perivascular nerves and local release of neurotransmitters, mainly norepinephrine and acetylcholine, controlling vessel diameter. This modulates cerebral blood vessel tone, contributing to the maintenance of optimal CBF and responding to dynamic synaptic activity in different brain regions. Vasodilatory peptides include NO and acetylcholine. Vasoconstrictive peptides include the neuropeptide Y and serotonin[51,52].

The anterior circulation, primarily served by the carotid arteries, is generally more adept at autoregulating CBF. This robustness in autoregulation ensures a stable blood supply to the cerebral hemispheres even with variations in blood pressure[53]. On the contrary, the posterior circulation, which includes the vertebrobasilar arteries that supply the brainstem, cerebellum, and occipital lobes, shows a comparatively weaker autoregulatory response. This difference makes the posterior regions more susceptible to ischemic events under hypotensive conditions[54]. A contributing factor to these regional disparities is the variation in sympathetic nerve density, which plays a crucial role in modulation of vascular tone. The anterior cerebral regions are characterized by a higher density of sympathetic innervation compared to the posterior regions[55]. This difference in sympathetic nerve distribution may partly explain the more effective autoregulatory responses observed in the anterior circulation. Understanding these regional variations in cerebral autoregulation is crucial, especially in disease processes such as reversible cerebral vasoconstriction syndrome and posterior reversible encephalopathy syndrome spectrum disorders[56].

Pressure reactivity index

The pressure reactivity index (PRx) is an advanced neuromonitoring parameter crucial to understanding cerebral autoregulation, particularly in the context of TBI and other neurocritical care conditions. PRx quantitatively assesses the state of cerebrovascular autoregulation by correlating ICP with arterial blood pressure (ABP). It is calculated as the moving Pearson’s correlation coefficient between slow waves of ICP and ABP over a specified time window, typically ranging from 30 s to several minutes. A positive PRx indicates impaired autoregulation, where ICP passively follows changes in ABP, while a negative or zero value suggests intact autoregulation, with appropriately constricting or dilating cerebral vessels in response to changes in ABP[57]. The utility of PRx lies in its ability to provide continuous real-time assessment of the cerebrovascular response to physiological and pathological stimuli. It has been particularly valuable in guiding therapeutic strategies in patients with TBI, such as optimizing CPP. Studies have demonstrated a correlation between PRx and outcomes in TBI, with higher PRx values associated with worse outcomes[58]. This correlation suggests that PRx can serve as a prognostic tool and a guide for individualized patient management.

Additionally, PRx has been used in research investigating the pathophysiology of cerebral autoregulation and its alteration in various clinical conditions. For example, studies have explored the impact of factors such as carbon dioxide levels, body temperature, and metabolic changes on PRx and cerebral autoregulation[59]. The insights gained from these studies contribute to a deeper understanding of the complex dynamics of CBF regulation and have implications for the development of targeted therapies in neurocritical care.

CEREBRAL ENERGETICS

After conceptualizing CBF, autoregulation, and their pressure flow relationships to better understand neurophysics, it is imperative to move from the arterioles through the capillary beds and into the cell. Cerebral energetics encompasses the study of the energy metabolism of brain, a critical aspect for its function and response to injury. As mentioned previously, the brain, although it only represents about 2% of body weight, accounts for approximately 20% of the body’s total energy expenditure. This high demand is primarily due to the energy-intensive processes of synaptic transmission and maintenance of ionic gradients across neuronal membranes[60]. The primary energy substrate for the brain is glucose, metabolized through aerobic glycolysis to produce ATP, the main energy currency of cells. The dependence of the brain on glucose is so significant that alterations in glucose metabolism can have profound effects on neural function and are implicated in various neuropathologies, including neurodegenerative diseases and brain injuries[61].

Oxygen is crucial for the aerobic metabolism of glucose in the brain. The tight coupling between CBF and metabolic demand ensures a constant supply of oxygen, adapting to the spatial and temporal changes of neuronal activity[62]. This relationship is fundamental in cerebral autoregulation and is affected in pathologies where metabolic demand or blood flow are altered. Recent studies in neuroenergetics have begun to elucidate the links between energy metabolism, brain function, and neurological diseases. For example, in conditions such as ischemic stroke or TBI, disruptions in energy metabolism can lead to neuronal dysfunction and death[63]. Mitochondrial dysfunction is a key feature in many neurodegenerative diseases, leading to energy deficits and neuronal degeneration. Research focusing on improving mitochondrial function or protecting mitochondria from damage holds promise for therapeutic interventions in a variety of neurological conditions[64].

In a healthy brain, there is a tight coupling between neuronal activity and CBF, ensuring an adequate supply of glucose and oxygen to active neurons. This is essential for maintaining ionic gradients and neurotransmitter cycling, which are energy-intensive processes[65]. During acute brain injury, such as TBI or stroke, normal neurometabolic coupling can be disrupted. This disruption leads to neurometabolic disconnection, a state in which neuronal energy demand and the supply of essential substrates (oxygen and glucose) are mismatched[66].

Neurometabolic coupling in brain injury can result from several factors: (1) Reduced CBF leading to ischemia and insufficient oxygen/glucose supply; (2) Mitochondrial dysfunction that affects ATP production; (3) Increased energy demand due to cellular repair processes or exacerbated by excitotoxicity; and (4) Disruption of the blood-brain barrier, leading to altered substrate delivery. The consequences of neurometabolic decoupling include energy deficits, lactate accumulation (due to anaerobic metabolism), cellular edema, oxidative stress, and eventually neuronal death[67]. In conditions like TBI, the area surrounding the injury (penumbra) is particularly vulnerable to neurometabolic decoupling. This area can undergo secondary injury processes exacerbated by the metabolic mismatch. For example, after TBI there is cerebral hyperglycolysis where brain cells consume glucose at a rate higher than normal, even in the presence of adequate oxygen supply. This is believed to be a response to the increased energy demands of the injured brain for processes such as membrane repair, ion homeostasis, and inflammation[67]. Therapeutic strategies often aim to restore the balance between energy supply and demand, such as ensuring adequate cerebral perfusion, using neuroprotective agents to mitigate excitotoxicity and supporting mitochondrial function[68]. Both CMD and PET imaging provide further insight into cerebral energetics and will be explored in more detail in this manuscript.

INVASIVE NEUROMONITORING TOOLS

The next section serves to provide a general overview of the various invasive and noninvasive neuromonitoring tools, indications, methodology, and physiological basis in the treatment of neurocritical patients and acute brain injury (Figure 2).

Figure 2 General overview of the various invasive and noninvasive neuromonitoring tools, indications, methodology, and physiological basis in the treatment of neurocritical patients and acute brain injury. EEG: Electroencephalography; ICP_wf: Intracranial pressure wave-form. This figure was generated using BioRender (Agreement No: SE269MFWGB).

ICP monitoring & wave-form analysis

External ventricular drains (EVDs) are a critical tool in neurocritical care to manage ICP and hydrocephalus. EVDs provide a direct method of measuring ICP and allow drainage of CSF to relieve elevated pressure within the cranial vault. This method is especially crucial in patients with acute brain injury, where rapid changes in ICP can have significant impacts on the patient’s prognosis[69]. The placement of an EVD involves the insertion of a catheter into the lateral ventricles, usually guided by neuroimaging to ensure accurate positioning. The catheter is connected to an external transducer that provides continuous ICP readings, thus facilitating real-time monitoring and management[70]. Monitoring ICP through EVDs is vital in patients with TBI, stroke, or other conditions that lead to increased ICP. By allowing for the timely detection and intervention of elevated ICP, EVDs play a significant role in preventing secondary brain injury and improving patient outcomes[71].

While EVDs are an invaluable tool, they are not without risks. Complications can include infection, bleeding, and catheter misplacement. Strict protocols for insertion and maintenance are essential to minimize these risks[72]. In the context of MMM, EVDs provide crucial data that can be integrated with other monitoring methods, such as CPP and oxygenation measurements of brain tissue. This integration improves the understanding of the patient’s neurological status and guides therapeutic interventions[73]. Additionally, the EVD provides critical information on CSF hydrodynamics via ICP wave-form (ICP_wf) analysis.

ICP_wf are essential for the detailed assessment and management of patients in neurocritical care. These waveforms, often referred to as Lundberg waves, are categorized into three main types: A, B, and C waves. A waves, also known as plateau waves, are characterized by sudden and dramatic increases in ICP to levels as high as 50-100 mmHg, lasting for 5-20 min. They are often indicative of significantly compromised cerebral autoregulation and are associated with severe intracranial pathology[74]. B waves are smaller rhythmic oscillations of ICP, occurring every 1-2 min and are often associated with unstable CBF, especially in patients with TBI or hydrocephalus[75]. The C waves are even smaller and more frequent fluctuations, usually seen in normal ICP monitoring, reflecting normal physiological variations in intracranial dynamics[76].

Continuous monitoring and analysis of these ICP_wf provide critical information on cerebral compliance, blood flow dynamics, and overall neurological status of the patient, guiding therapeutic decision-making in neurocritical care settings[77]. The SYNAPSE-ICU study demonstrated that continuous ICP monitoring in patients with acute brain injury led to increased therapeutic vigilance, decreased mortality, and improved neurologic outcomes[78]. Additionally, thresholds for increased ICP have fluctuated throughout the years. In 2000, the Brain Trauma Foundation guidelines suggested a normal ICP ranging from 20-25 mmHg. This was changed to ≤ 20 mmHg in 2007. More recently in 2016 the guidelines were refined to an ICP ≤ 22 mmHg that showed improved morality and favorable outcomes[57].

Güiza et al[79] evaluated the impact of increased ICP on neurological outcomes in adult and pediatric patients with TBI. The researchers analyzed minute-by-minute ICP and blood pressure data from 261 adults and 99 children. ICP above 20 mmHg lasting more than 37 min in adults and more than 8 min in children correlated with poorer outcomes. The study also established that the cumulative burden of ICP over time is an independent predictor of mortality, in conjunction with known baseline risk factors for severe TBI. Furthermore, it highlighted that impaired cerebrovascular autoregulation limits the ability to tolerate elevated ICP significantly. When cerebral perfusion pressure falls below 50 mmHg, any duration of elevated ICP is associated with worse outcomes. This study underscored that secondary injury in pediatric TBI occurs at lower ICP thresholds and emphasizes the importance of maintaining cerebral perfusion pressure above 50 mmHg, especially in cases of severe TBI. Therefore, ICP should not be perceived as a dichotomous value but rather as a dynamic value individualized to each patient with considerations of the integrity of cerebral autoregulation.

Brain tissue oxygenation monitoring

Continuous monitoring of brain tissue oxygenation (PbtO₂) represents a pivotal advance in neurocritical care, offering localized, real-time information on cerebral oxygenation status. This technique involves inserting a probe into the cerebral tissue, typically the white matter of the frontal lobe, to directly measure the partial pressure of oxygen in the brain tissue. PbtO₂ monitoring is particularly crucial in the management of patients with conditions such as TBI and stroke, where cerebral oxygenation is at risk[80]. The primary utility of the technique lies in detecting cerebral hypoxia, a state of inadequate oxygen supply to brain tissues, which can lead to cellular dysfunction and death if not promptly managed[81]. Normal PbtO₂ values typically range between 20-40 mmHg, with values below 10 mmHg indicating critical hypoxia. Brain tissue hypoxia has been defined as PbtO₂ < 15 mmHg by some researchers[82,83] and as a PbtO₂ < 20 mmHg by others[84-86].

Through continuous monitoring, PbtO₂ allows the neurointensivist to tailor therapeutic interventions, including optimizing CPP and adjusting ventilation strategies. It is often employed alongside other neuromonitoring methods, such as ICP monitoring and CMD, to provide a comprehensive view of the state of the brain[87]. Despite its benefits, PbtO₂ monitoring measures oxygenation in a limited brain area and carries risks such as infection and bleeding because it is invasive. However, advances in sensor technology are enhancing its accuracy and application, paving the way for more personalized treatment approaches in neurocritical care[88].

The Brain Oxygen Optimization in Severe Traumatic Brain Injury Phase (BOOST)-1 and BOOST-2 studies were significant in advancing our understanding of brain tissue oxygenation in the treatment of severe TBI. BOOST-1, an observational study, established the feasibility of monitoring PbtO₂ in patients with severe TBI, highlighting a correlation between lower levels of PbtO₂ and poorer outcomes. This study underscored the potential importance of maintaining adequate oxygenation of brain tissue in the management of TBI[89].

Based on this, the BOOST-2 study, a randomized clinical trial, further evaluated PbtO₂-directed therapy. It showed that a management protocol focused on maintaining PbtO₂ above a threshold was feasible and safe, suggesting potential improvements in neurological outcomes, although it did not conclusively establish this benefit, indicating the need for more trials[90]. In response, the ongoing BOOST-3 trial aims to provide more definitive evidence. This large-scale study examines whether the management of patients with severe TBI using PbtO₂ monitoring, in conjunction with ICP monitoring, leads to better outcomes than relying solely on ICP monitoring. The results of BOOST-3 are eagerly awaited, as they have the potential to significantly influence the standard of care in the management of severe TBI.

SjvO₂

SjvO₂, an integral monitoring tool in neurocritical care, offers crucial insights into global cerebral oxygenation status by measuring oxygen saturation in the venous blood that drains from the brain. This technique involves the insertion of a catheter into the internal jugular vein to continuously monitor the SjvO₂. SjvO₂ acts as an indirect indicator of the balance between cerebral oxygen supply and demand. Typically, SjvO₂ values range from 55%-75%, with values below 55% suggesting potential cerebral ischemia or hypoxia indicating that the metabolic demands of the brain are not adequately met.

This monitoring is especially relevant in patients with TBI, subarachnoid hemorrhage (SAH), or severe ischemia, as it helps detect critical changes in cerebral oxygenation and guides therapeutic decisions such as ventilator adjustments and perfusion management. The catheter is usually placed in the dominant jugular vein, determined by either imaging or clinical assessment, to ensure accurate measurement of cerebral venous outflow[91].

Although SjvO₂ provides continuous, global information on cerebral oxygenation, it does have limitations, including its inability to offer regional data and its susceptibility to inaccuracies due to patient positioning or extracranial contamination. Despite these limitations, its role in dynamically assessing cerebral oxygenation makes it a valuable tool in neurocritical care. Current research efforts aim to enhance its precision and integrate it with other neuromonitoring methods, enriching overall patient care strategies[92].

TDF and regional cerebral blood flow

TDF offers a distinctive approach in neurocritical care for assessing regional cerebral blood flow (rCBF), especially crucial in patients with TBI or cerebrovascular disease. This technique involves placing a specialized probe in brain tissue, which measures blood flow according to the principle of thermal diffusion. The probe consists of a heater and a temperature sensor; the heat dissipation of the heater is affected by the blood flow in adjacent brain tissues, thus providing a measure of rCBF. TDF is valued for its ability to provide continuous, real-time monitoring of localized CBF, allowing clinicians to detect and respond to changes in cerebral perfusion immediately. Normal rCBF values range from 20 to 30 mL/100 g/min, and deviations from this range can indicate pathological conditions. For example, reduced rCBF can signal ischemia, while excessively high values can indicate hyperemia, both of which require immediate medical attention[93]. The ability of TDF to continuously monitor rCBF helps guide therapeutic interventions such as optimizing cerebral perfusion pressure and managing ICP. Despite its benefits, TDF does have limitations, including its invasiveness and the potential for measurement error due to probe placement or tissue changes around the probe. However, TDF remains a valuable tool in nuanced management of CBF dynamics in neurocritical patients[94].

CMD

CMD is a sophisticated neuromonitoring technique that is increasingly being used in neurocritical care to assess the metabolic state of brain tissue at the cellular level. This technique involves inserting a small catheter into the brain parenchyma, typically in the cerebral cortex. The catheter is perfused with a physiological solution, allowing the collection of extracellular fluid, which reflects the metabolic processes that occur within brain tissue. The primary indication for CMD is in patients with TBI, SAH, or large hemispheric strokes, where it helps detect metabolic disturbances that can lead to secondary brain injury[95]. The placement of the microdialysis catheter is typically guided by neuroimaging or during surgery, with the aim of positioning it within or near the area of interest, such as a penumbral zone during stroke or an area at risk in TBI. The collected samples are analyzed for various biochemical markers that provide information on cellular metabolism, such as glucose, lactate, pyruvate, and glutamate levels.

One of the key outcome measures in CMD is the lactate/pyruvate ratio (LPR). A normal LPR is typically below 25. An elevated LPR indicates a change to anaerobic metabolism, often due to ischemia or hypoxia, and is associated with poor outcomes in patients with brain injury[96]. Elevated LPR can guide interventions aimed at improving cerebral perfusion and oxygenation. On the contrary, low LPR values, particularly when accompanied by low glucose levels, may indicate hyperglycolysis, a condition seen in the early stages after injury[97]. CMD provides real-time information on the biochemical environment of the brain, allowing for tailored treatment of patients with acute brain injuries. Data from this technique help guide therapeutic decisions, such as optimizing cerebral perfusion pressure, adjusting glucose and oxygen delivery, and mitigating excitotoxicity. Despite its invasiveness and the need for specialized equipment and expertise, CMD is a powerful tool in the arsenal of neurocritical care, offering unique insights into brain metabolism and guiding patient-specific treatment strategies (Table 1).

Table 1 Various proposed cerebral microdialysis substrates.

Function	Substrates
ATP usage	Adenosine, hypoxanthine, inosine, K+
Membrane dysfunction	Glycerol
Reactive oxygen species	Xanthine, uric acid, glutathione, cysteine
Nitric oxide formation	Nitrite, nitrate, citrulline, arginine
Brain swelling	Potassium, taurine
Inflammation	IL-1, IL-6, GFAP, TNF
Blood-brain barrier	Alanine, leucine, and valine
Neurotransmitters and amino acids	Serine, 5-HIAA, GABA, Glycine

5-HIAA: 5-hydroxyindoleacetic acid; GABA: Gamma-aminobutyric acid; GFAP: Glial fibrillary acidic protein; IL-1: Interleukin 1; IL-6: Interleukin 6; TNF: Tumor necrosis factor.

Retrodialysis is a unique application of CMD in which substances are administered to the brain parenchyma through the microdialysis catheter, allowing for localized therapeutic interventions or metabolic studies. One specific application of retrodialysis is the focal perfusion of succinate, a key intermediate in the tricarboxylic acid cycle. Succinate administration via retrodialysis can serve as a tool to assess mitochondrial function and cellular metabolism within the brain. This approach is based on the premise that the administration of succinate locally can improve mitochondrial respiration, especially in areas of compromised metabolism or mitochondrial dysfunction. Stovell et al[98] aimed to examine the impact of succinate on brain energy metabolism in patients with acute TBI. The study carried out in 8 patients with TBI showed that succinate perfusing by microdialysis increased extracellular pyruvate levels and reduced the LPR, indicating improved cellular chemistry. Interestingly, a significant correlation was found between the decrease in the LPR and the increase in phosphocreatine/ATP ratio in individual patients, suggesting that succinate may enhance brain energy metabolism in patients with TBI experiencing mitochondrial dysfunction. This study supports the connection between LPR and brain energy state and the potential of succinate in treating TBI patients.

The study by Khellaf et al[99] focused on patients with TBI who exhibited impaired energy metabolism, as indicated by elevated LPR and was identified in 73% of the 33 patients with TBI under MMM. The researchers administered disodium 2,3-¹³C2 succinate via retrodialysis, observing a decrease in LPR by 12% and an increase in brain glucose by 17%. Nuclear magnetic resonance spectroscopy of the microdialysates confirmed that the administered succinate was metabolized intracellularly through the tricarboxylic acid cycle. This study, using a tiered management protocol focusing on LPR and integrating various neurocritical care monitoring methods, demonstrated that succinate administration can improve energy metabolism in patients with TBI with CMD.

Electrocorticography and dEEG

In the context of neurocritical care within the ICU, electrocorticography (ECoG) and dEEG play a pivotal role in comprehensive monitoring and treatment of patients with severe brain injuries and those at risk for neurological complications, such as seizures or epilepsy. ECoG in the ICU setting is primarily used in cases where precise monitoring of cortical electrical activity is essential, such as in patients with TBI or after neurosurgery. This technique, which involves placing electrodes directly on the surface of the brain, provides high-resolution data on cortical electrical activity. In the acute care setting, ECoG can be instrumental in detecting subtle seizures that are not apparent on scalp EEG, thus guiding antiepileptic therapy. In addition, it can monitor cortical function in real time, providing valuable information for decision making in cases where cerebral edema or ischemia is concerned[100].

dEEG, involving the insertion of electrodes into brain tissue, is crucial for monitoring patients with deep brain injuries or those who have undergone surgeries involving subcortical structures. In the ICU, dEEG is often used in patients who are at risk of seizures originating in deep cerebral regions, which are not easily captured by surface EEG. This technique is particularly valuable for differentiating epileptic activity from other forms of abnormal brain activity, such as posttraumatic or postsurgical changes. The precise location of the seizures provided by dEEG is crucial in guiding the administration of targeted antiepileptic drugs and in making decisions about potential surgical interventions in cases of refractory epilepsy[101].

Connolly et al[102] obtained simultaneous EEG and ICP data in 2 patients undergoing clinical burst suppression. They identified a vasodilatory index to assess neurovascular coupling in patients with acute brain injury. They found that the electrical activity captured by EEG was correlated with transient increases in ICP, and the rise and fall of ICP were correlated with both the burst and suppression seen on EEG. They estimated a parabolic relationship where the ICP peaked 8 s after the EEG burst and then normalized approximately 15 s later. Furthermore, the presence of a correlating vasodilatory index demonstrates that the increase in ICP is mainly driven by changes in cerebrovascular hemodynamics.

NONINVASIVE NEUROMONITORING TOOLS

TCD

A key application of TCD is the calculation of the Lindegaard ratio, an index used to differentiate between cerebral vasospasm and hyperemia. The Lindegaard ratio is calculated by dividing the mean blood flow velocity in the middle cerebral artery by the mean velocity (MV) in the extracranial internal carotid artery. A ratio greater than 3 is indicative of vasospasm. This distinction is crucial, especially in patients after SAH, as it guides therapeutic decisions and interventions to manage vasospasm and prevent ischemic complications[103].

TCD also allows for an indirect estimate of CBF. Although TCD does not directly measure CBF volume, blood flow velocity can serve as a surrogate marker. By assessing changes in flow velocities in response to various physiological and pharmacological challenges, TCD can provide information on cerebrovascular reactivity and autoregulation. For example, techniques such as breath holding or carbon dioxide challenge tests can be used to assess cerebrovascular reactivity, where an increase in carbon dioxide levels leads to vasodilation and increased blood flow velocity. This response can be quantified and used to assess the integrity of cerebral autoregulation mechanisms[104].

The pulsatility index (PI) obtained from TCD ultrasound is a valuable parameter that reflects the resistance to blood flow in the cerebral vasculature. The PI is calculated using the formula: PI = [peak systolic velocity - end diastolic velocity (EDV)]/MV. The Bellinger PI formula is another way to express resistance in cerebral vessels and is calculated as follows: Bellinger PI = [2 × (MV − EDV)]/(peak systolic velocity + EDV). Both formulas are used to assess cerebral hemodynamics, with higher PI values generally indicating increased cerebral vascular resistance, which can be seen in conditions such as cerebral vasospasm or increased ICP.

In SAH, elevated PI values often indicate increased cerebral vascular resistance, which can be due to factors such as cerebral vasospasm or elevated ICP. Monitoring PI in patients with SAH is particularly useful for the early detection of vasospasm, a common and serious complication that can lead to delayed cerebral ischemia. A rise in PI may precede the clinical symptoms of vasospasm, providing a window for early intervention. In addition, PI can help to assess the efficacy of therapeutic interventions aimed at mitigating vasospasm and managing cerebral hemodynamics. Elevated PI of approximately 1.5 has been associated with vasospasm in SAH, and normal PI ranges from 0.5-0.6[105].

Cerebral perfusion pressure has also been estimated from TCD measurements by the following equation[106]: CPP = mean flow velocity/( mean flow velocity - EDV) × (MAP -diastolic blood pressure).

Kim et al[107] created a mean ICP_wf from thousands of individual waveforms. A similar impairment in cerebral hemodynamics has been observed in non-acute brain injury within the ICU. For example, patients with sepsis have been shown to have an elevated median blood flow velocity and PI in early sepsis without an impairment in cerebral autoregulation in early sepsis. However, in late sepsis impaired cerebral autoregulation has been observed.

NIRS & cerebral oximetry

NIRS uses the near-infrared region of the electromagnetic spectrum (approximately 500-800 nm) to determine the properties of the brain. An infrared beam passes through a substance and is then interpreted by a sensory probe, following Beer-Lambert’s law. A = [(log)_10 I_0]/I_i = alc. Where: A = absorbance, a = molar absorptivity, I = path length of the sample, and c = concentration of compound in solution.

The primary advantage of NIRS is its noninvasiveness, ease of use, and the ability to provide real-time monitoring without the need for patient sedation or transport. However, its limitations include sensitivity to extracranial contamination (such as scalp and skull blood flow), limited penetration depth (restricting its assessment to the cortical surface), and variability in readings based on individual anatomy and probe placement. An important application of NIRS is in assessing cerebral autoregulation and the ability of the brain to maintain stable CBF despite changes in systemic blood pressure. NIRS can generate various measurements including regional cerebral oxygen saturation (rSO₂), regional cerebral blood flow index, fractional tissue oxygen extraction, tissue oxygenation index, and relative total tissue hemoglobin concentration. Oxygen metabolism within the neuron can also be estimated by measuring the oxidation state of cytochrome c oxidase[108].

Time domain and frequency domain analyses are critical in evaluating cerebral autoregulation and blood flow dynamics. The cerebral oximetry index (COx), derived from the correlation of rSO₂ with ABP, utilizes time domain analysis to observe how rSO₂ changes over time with fluctuations in blood pressure and frequency domain analysis to examine spectral components of these signals. Positive COx indicates impaired autoregulation, where cerebral oxygenation passively follows changes in systemic pressure[58]. The tissue oxygen reactivity index (TOx), which correlates with ABP, functions similarly, providing information on the stability of cerebral oxygenation. The hemoglobin volume index (HVx) uses the relative total tissue hemoglobin concentration of the NIRS and correlates it with ABP to assess changes in cerebral blood volume. Positive HVx suggests impaired autoregulation indicated by a passive relationship between cerebral blood volume and blood pressure.

Lastly, the brain flow autoregulation index associates the regional cerebral blood flow index with ABP, helping to understand how CBF, estimated by NIRS, responds to systemic pressure variations. A high correlation in brain flow autoregulation index implies a direct influence of blood pressure on CBF, indicating a loss of autoregulatory capacity. A normal rSO₂ is from 55%-80%, and < 50% is concerning for cerebral ischemia[109]. Both TOx and COx are the most widely used indices for assessing cerebral autoregulation. Elevated TOx/COx values have been associated with an increased risk of delirium and mortality[110,111]. These analyses are instrumental in neurocritical care for conditions such as TBI, stroke, and SAH, providing essential information for patient-specific treatment (Table 2)[112].

Table 2 Near infrared spectroscopy-derived indices and clinical utility for the neurointensivist.

NIRS derived indices	Utility
COx = correlation coefficient (rSO2, MAP)	COx: This index is derived by correlating rSO2 with MAP. It is used to assess cerebral autoregulation, with a positive correlation indicating impaired autoregulation
TOx = correlation coefficient (TOI, MAP)	TOx: Similar to COx, TOx is calculated by correlating the TOI with MAP. It provides information on the balance between oxygen supply and demand at the tissue level
HVx = correlation coefficient (rTHb, MAP)	HVx: HVx correlates the rTHb, as measured by NIRS, with MAP. It is used to evaluate cerebral blood volume changes in response to blood pressure fluctuations.
BFAx = correlation coefficient (rCBFi, MAP)	BFAx: BFAx correlates the rCBFi with MAP. A positive correlation indicates that cerebral blood flow is more passive to pressure, suggesting impaired autoregulation
CMRO2 = cerebral blood flow × (arterial O2 content - venous O2 content)	Estimation of the CMRO2: Although the direct calculation of CMRO2 using NIRS is complex, NIRS data combined with other parameters can estimate CMRO2. This provides information on the metabolic state of the brain

BFAx: Blood flow autoregulation index; COx: Cerebral oximetry index; CMRO₂: Cerebral metabolic rate of oxygen consumption; HVx: Hemoglobin volume index; MAP: Mean arterial pressure; NIRS: Near infrared spectroscopy; O₂: Oxygen; rCBFi: Regional cerebral blood flow index; rSO₂: Regional cerebral oxygen saturation; rTHb: Relative total tissue hemoglobin concentration; TOI: Tissue oxygenation index; TOx: Tissue oxygenation index.

Damian et al[113] explored the effectiveness of NIRS in monitoring intracranial oxygenation in patients with complete middle cerebral artery stroke and associated brain swelling. The study involved 24 patients, for whom NIRS optodes were placed on both frontal lobes to measure rSO₂ at frequent intervals. The findings indicated that while the absolute values of rSO₂ varied significantly between patients and did not directly correlate with clinical data, the differential in rSO₂ between the infarcted and contralateral hemispheres provided valuable information. This differential was typically higher on the infarct side in most cases. In particular, the differential in rSO₂ decreased with brain swelling and disappeared in patients with herniation but increased markedly after successful decompressive hemicraniectomy. The study by Kurth et al[114] focused on determining hypoxic-ischemic thresholds for cerebral oxygen saturation (SCO₂) in neonates (piglets) using NIRS. In 60 anesthetized piglets, the study correlated SCO₂ levels with changes in EEG, brain ATP, and lactate concentrations and compared them with CBF and sagittal sinus oxygen saturation. Key findings included identifying SCO₂ thresholds for increased lactate, minor and major EEG changes, and decreased ATP, which were 44%, 42%, 37%, and 33%, respectively. These thresholds were significantly lower than the baseline SCO₂ of 68%, indicating a substantial buffer zone between normal function and dysfunction.

A major limitation to the utilization of NIRS is scattering and reflectance. The beam of light passes through multiple layers including the skin, tissue, and bone. For typical source-to-detection distances of 4 cm, approximately 0.3 cm of brain cortex is measured[115]. Furthermore, in neurosurgical patients or those with TBI, CSF, hygroma, and blood products including iron may further confound the clinical utility of NIRS. Interbiological variability between patients, including variations in arterial-venous filling, also hinders the use of ‘normal’ baseline values. Rather, using NIRS indices as a ‘trend monitor’ for changes may be of greater clinical utility.

ONSD

Measurement of the ONSD has emerged as a noninvasive technique for estimating ICP in neurocritical care. The ONSD assessment is based on the principle that the optic nerve sheath, being contiguous with the subarachnoid space, reflects changes in the CSF pressure. An increase in ICP leads to a corresponding increase in sheath diameter. ONSD is typically measured using transorbital sonography, a bedside ultrasound technique[116]. The ultrasound probe is placed on the closed eyelid and measurements are taken 3 mm behind the globe where the optic nerve is perpendicular to the probe.

ONSD measurement is particularly useful in settings where invasive ICP monitoring is not feasible or in patients with contraindications to invasive methods[117]. It has been used effectively in the treatment of conditions such as TBI, stroke, and idiopathic intracranial hypertension. Several studies have validated the correlation between ONSD measurements and invasive ICP readings. The sensitivity and specificity of ONSD in detecting elevated ICP have been reported to be high, making it a reliable surrogate marker. However, individual patient variability and technique dependence are factors that require careful interpretation of ONSD measurements[118].

Chang et al[119] explored the correlation between PI, ONSD, and ICP in patients with TBI after surgery. The study, which involved 68 patients with TBI, used TCD ultrasound to analyze the relationship between these parameters. The key findings included a significant correlation between ONSD and ICP, especially when ONSD was ≥ 5 mm. In addition, a strong correlation was observed between PI and ICP, particularly on days 6 and 7 after surgery. The study also evaluated the effectiveness of PI and ONSD in predicting intracranial hypertension, finding that the combination of PI ≥ 1.2 mm and ONSD < 5 mm provided the most accurate prediction, with an area under the curve (AUC) of 0.943. These results underscore the utility of PI and ONSD as noninvasive methods for assessing and predicting elevated ICP in postoperative patients with TBI.

In their systematic review and meta-analysis, Aletreby et al[120] evaluated the diagnostic accuracy of ONSD ultrasound as a noninvasive estimator of raised ICP comparing it against standard invasive ICP measurements. The review included 18 prospective studies, with 16 studies encompassing 619 patients primarily analyzed. In particular, the analysis revealed a correlation coefficient of 0.7 between ICP and ONSD, suggesting a strong relationship. However, meta-regression did not identify any significant covariates, and subgroup analyses on severe TBI and parenchymal ICP indicated no heterogeneity. The conclusion drawn from the study was that while ONSD ultrasound is a valuable tool in assessing ICP, it should not replace invasive methods where feasible and appropriate. The high diagnostic precision of ONSD suggests its utility as an adjunct in the evaluation of ICP, especially in scenarios where invasive monitoring may not be immediately available or feasible.

TMD

TMD measurement is an emerging noninvasive technique for estimating ICP. The underlying principle is based on the anatomical and physiological connection between the intracranial compartment and the perilymphatic fluid of the inner ear. Changes in ICP are transmitted to the perilymph and can affect the position and mobility of the tympanic membrane[121]. TMD is measured using specialized equipment that detects and quantifies the movement of the tympanic membrane in response to sound stimuli. Typically, a probe is inserted into the external auditory canal, creating a hermetic seal. Low-frequency sound waves are introduced, and the resulting movements of the tympanic membrane are recorded. The displacement patterns are then analyzed to infer ICP levels, with specific displacement characteristics correlating with increased ICP[122].

Evensen et al[123] conducted a study to determine whether the tympanic membrane pressure (TMP) waveform could noninvasively estimate the ICP waveform. The study involved 28 people who underwent invasive ICP measurements for clinical reasons, such as surveillance after SAH or diagnostics for CSF circulation disorders. The researchers established a transfer function estimate between invasive ICP and noninvasive TMP signals for each participant, with the aim of evaluating the potential of the method. The validation of this method involved comparing the mean wave amplitude (MWA) computed from the estimated and invasively measured ICP_wf. The study found that patient specific noninvasive ICP signals could satisfactorily predict MWA in 4 of the 28 individuals (14%). In these 4 patients, the differences between the original and estimated MWAs were less than 1.0 mmHg in more than 50% of the observations and less than 0.5 mmHg in more than 20% of the observations. Furthermore, the study revealed that the cochlear aqueduct functions as a physical low-pass filter, influencing the transmission of pressure signals. This finding is significant as it suggests the feasibility of using TMP waveforms to estimate ICP_wf noninvasively in certain patients, although the effectiveness of this method varies among individuals. Research implies the potential for TMP waveform analysis in noninvasive ICP monitoring but also highlights the need for further research to improve its accuracy and applicability[123].

PET

PET is a highly sophisticated imaging modality that plays a pivotal role in neurocritical care by providing functional and metabolic information on brain pathology. PET works by detecting gamma rays emitted indirectly by a positron-emitting radionuclide (tracer), which is introduced into the body on a biologically active molecule[124]. This technique is particularly valuable for assessing cerebral metabolism, blood flow, and receptor binding, offering a unique window into the biochemical processes of the brain. Typically, a tracer such as 18F-fluorodeoxyglucose, which is absorbed by active brain tissue, is used. The distribution of the tracer, imaged by PET, correlates with regional cerebral glucose metabolism, offering insights into neuronal viability and function. The main advantage of PET is its ability to provide detailed information about the functional state of the brain, which cannot be obtained by other imaging modalities[125]. However, its limitations include high cost, limited availability, and the need for radioactive tracers. Moreover, the interpretation of PET data requires sophisticated analytical techniques and a thorough understanding of cerebral pathophysiology[126].

Veenith et al[127] investigated tissue hypoxia and ischemia in patients with TBI using combined oxygen 15-labeled PET and fluorine 18-labeled fluoromisonidazole (18F-FMISO) PET imaging. The study involved 10 patients with TBI and two control cohorts of 10 healthy volunteers each, undergoing 18F-FMISO and oxygen 15-labeled PET scans. The research assessed the burden and distribution of macrovascular and microvascular ischemia in early TBI. The main findings were that compared to controls, TBI patients had significantly higher median ischemic brain volume and hypoxic brain volume. The spatial distributions of these pathophysiological tissue classes within the injured and normal-appearing brain regions were poorly matched. The hypoxic brain volume compartment showed similar CBF, cerebral blood volume, and cerebral metabolic rate of oxygen to the ischemic brain volume compartment but had a lower oxygen extraction fraction, indicating a more severe injury. Monitoring of brain tissue oximetry suggested that increased 18F-FMISO trapping is likely to occur at oxygen tensions of 15 mmHg or lower. The study concluded that tissue hypoxia in TBI is not limited to regions with structural abnormalities and can occur without conventional macrovascular ischemia. This finding, consistent with microvascular ischemia, highlights the need for novel neuroprotective strategies targeting these physiological changes[127].

HRV analysis

HRV is an emerging tool in neurocritical care, providing valuable insight into the function of the autonomic nervous system. HRV refers to the physiological phenomenon of variation in the time interval between heartbeats, measured by the variation in the beat-to-beat interval. It reflects the ability of the heart to respond to a variety of physiological and environmental stimuli. In patients with TBI stroke and other neurological conditions, HRV analysis can help assess autonomic dysfunction, which is often a secondary complication. HRV metrics are associated with the severity of brain injury and can predict outcomes. For example, reduced HRV has been associated with a poor prognosis in patients with TBI, indicating a higher degree of autonomic dysregulation[128].

Megjhani et al[129] conducted a study to evaluate the use of HRV measures to detect neurocardiogenic injury (NCI) in patients with aneurysmal SAH. The study included 326 consecutive admissions, with 56 (17.2%) developing NCI, which was defined by abnormal wall movement with ventricular dysfunction or elevated cardiac troponin I levels without signs of coronary artery insufficiency. HRV measures, which encompassed time and frequency domains, as well as nonlinear techniques like detrended fluctuation analysis, were calculated for the first 48 h. Using multilevel longitudinal linear regression, the study examined the relationship between HRV measures and NCI and evaluated differences between groups at baseline and over time. The results showed a decrease in vagal activity in subjects with NCI, indicated by a higher low/high frequency ratio, suggesting a change in sympathovagal balance towards sympathetic activity. All time-domain HRV measures were lower in patients with SAH with NCI.

Using an ensemble machine learning approach, these HRV measures were transformed into a classification tool that exhibited good discrimination capabilities with an area under the receiver operating characteristic curve (AUC-ROC) of 0.82, an area under the precision recall curve of 0.75, and a correct classification rate of 0.81. The study concluded that HRV measures are significantly associated with NCI, and a machine learning model using HRV-derived characteristics can effectively classify SAH patients who develop NCI. This suggests that HRV analysis, enhanced by machine learning techniques, could be a valuable tool in the early detection of NCI in patients with SAH[129].

Medical management of malignant catatonia and toxic leukoencephalopathy with diffuse white matter disease has also been associated with decreased HRV parameters, including standard deviation of normal-to-normal intervals, percentage of successive normal-to-normal R-R intervals that differ by more than 50 ms, root mean square of successive RR intervals, and cardiac entropy[130].

Srichawla et al[130] conducted a systematic review to assess the effectiveness of HRV biofeedback in treating autonomic nervous system dysregulation following TBI. The review found that the participants underwent an average of 11 HRV biofeedback sessions (ranging from 1 to 40 sessions). The results indicated that HRV biofeedback was associated with improved HRV in patients with TBI. Furthermore, there was a positive correlation between increased HRV after biofeedback and recovery from TBI. Improvements in cognitive and emotional functioning, as well as physical symptoms such as headaches, dizziness, and sleep problems, were noted.

MMM AND MACHINE LEARNING

Integration of MMM with machine learning in the ICU represents a significant advance in neurocritical care. This fusion utilizes comprehensive data obtained from various monitoring modalities, such as ICP, cerebral oximetry, EEG, and brain tissue oxygenation, to improve patient care through predictive analytics and a decision support system. Machine learning algorithms can analyze vast and complex datasets from MMM, identifying patterns and correlations that may be imperceptible to human analysis. This approach enables the prediction of critical events, such as secondary brain injury, cerebral ischemia, or imminent seizures, well before they become clinically apparent. For example, adequate utilization of MMM in conjunction with advanced machine learning techniques may allow for the identification of herniation syndromes hours before they occur. Algorithms such as neural networks, support vector machines, and decision trees have been applied to neuromonitoring data to predict outcomes and guide therapeutic interventions[131,132].

Tas et al[133] examined the application of MMM in adult patients with various forms of acute brain injury, including TBI, SAH, intracerebral hemorrhage, acute ischemic stroke, and hypoxic ischemic brain injury after cardiac arrest. The review, covering studies from 2015 to 2022, identified 112 MMM studies, highlighting the predominant combination of ICP monitoring with brain tissue oxygenation PbtO₂ in most studies. Their analysis revealed that MMM is mostly applied to cases of TBI and SAH, with a median sample size of 36 patients in the studies, many of which took over 5 years to enroll participants. The studies were classified into observational (68 studies) and interventional (44 studies), with the interventions further subdivided into systemic, cerebral, and guided by MMM. A significant majority (82%) of MMM-guided intervention studies incorporated clinical outcomes as endpoints, and 78% of these studies demonstrated a significant improvement in outcomes with MMM-guided interventions. Major limitations of this review include record screening by a single author and not being conducted according to accepted PRISMA guidelines as a systematic review.

Schweingruber et al[134] conducted a study using recurrent machine learning models to predict significant increases in ICP in patients with invasive ICP monitoring. The study trained models using an institutional cohort (ICP-ICU) of 1346 patients and validated them externally on two publicly available data sets: medical information mart for intensive care (MIMIC, 998 patients) and the Collaborative Research Database of the emergency intensive care unit (eICU, 1634 patients). The models were designed to forecast increases in ICP of 22 mmHg over periods longer than 2 h in the coming hours. The study evaluated predictive performance at different time intervals up to 24 h before the critical phase, observing a decrease in performance over longer intervals. However, even at 24 h, the models demonstrated robust AUC-ROC (ICP-ICU: 0.826; MIMIC: 0.836; eICU: 0.779), which were higher at shorter intervals (1-h AUC-ROC: ICP-ICU: 0.982; MIMIC: 0.965; eICU: 0.941). The machine learning models, based on long-short-term memory (LSTM), operated on sparse hourly data and were capable of handling variable input lengths and data missingness. The study also applied gradient-based feature importance analysis to reveal the decision-making processes of the model, improving its clinical interpretability. The findings indicated that recurrent machine learning models, particularly those based on LSTM, could be an effective tool to predict increases in ICP, offering high translational potential in neurocritical care settings. The study involved predicting changes in ICP over time, which is inherently sequential data. Recurrent models like LSTM are designed to process time-series data by maintaining a memory of previous inputs. This is crucial for accurately predicting future events based on past trends. They can handle inconsistencies in data collection, which is common in clinical settings. Given their complexity, there is a risk of overfitting to the training data, which can reduce the performance of the model on new, unseen data.

The MASTER-TBI project is a multicenter longitudinal cohort study that employs a novel hybrid cloud platform and data science techniques to collect and analyze data from moderate-severe TBI (m-sTBI) patients undergoing ICP monitoring and ICM+^® (Cambridge Enterprise) neuromonitoring in three Australian trauma ICUs. The collaborative employs a hybrid cloud platform for data storage and management. Hybrid cloud platforms combine on-premises infrastructure (or private clouds) with public clouds, allowing for flexible, scalable, and more secure data handling. This is particularly useful in healthcare settings where large volumes of sensitive data are generated and need to be processed efficiently while complying with data protection regulations. Data captured for analysis include pathophysiological events, surgical interventions, duration of hospital and ICU stay, patient discharge status, and Glasgow Outcome Score Extended at 6 mo, where available. The study highlighted the development of data science-informed systems and techniques within the MASTER-TBI collaborative, with the objective of maximizing the utility of high-resolution m-sTBI patient neuromonitoring data. The systems developed are described as innovative and world class, with the potential to significantly improve the care and outcomes of patients with m-sTBI[135].

CONCLUSION

The integration of neurophysical principles including cerebral hemodynamics and energetics, MMM, and advanced data science analyses using machine learning algorithms provides a promising intersection in the field of neurocritical care. Further advances are on the horizon including triple-lumen intracranial monitors assessing ICP, brain oxygenation, and temperature, as well as simultaneous PbtO₂-EEG, NIRS-EEG, and retrodialysis. Utilizing these tools alongside advanced machine learning models will aim to improve predictive modeling within the ICU. For example, estimating intracranial hypertension, associated herniation syndromes, and delayed cerebral ischemia in SAH hours or days before they occur. In addition, a better understanding of the underlying cerebral physiology of acute brain injury will provide additional avenues for therapeutics. Thus, the role of the neurointensivist is changing from neuroprognisticator to neuroprotector. However, it is crucial to continue to advance our techniques and methodologies to further enhance our understanding of cerebral physiology and patient care in the neurological ICU.