Systematic Review and Meta-Analysis of Prehospital Machine Learning Scores as Screening Tools for Early Detection of Large Vessel Occlusion in Patients With Suspected Stroke.

BACKGROUND: Enhanced detection of large vessel occlusion (LVO) through machine learning (ML) for acute ischemic stroke appears promising. This systematic review explored the capabilities of ML models compared with prehospital stroke scales for LVO prediction. METHODS AND RESULTS: Six bibliographic databases were searched from inception until October 10, 2023. Meta-analyses pooled the model performance using area under the curve (AUC), sensitivity, specificity, and summary receiver operating characteristic curve. Of 1544 studies screened, 8 retrospective studies were eligible, including 32 prehospital stroke scales and 21 ML models. Of the 9 prehospital scales meta-analyzed, the Rapid Arterial Occlusion Evaluation had the highest pooled AUC (0.82 [95% CI, 0.79-0.84]). Support Vector Machine achieved the highest AUC of 9 ML models included (pooled AUC, 0.89 [95% CI, 0.88-0.89]). Six prehospital stroke scales and 10 ML models were eligible for summary receiver operating characteristic analysis. Pooled sensitivity and specificity for any prehospital stroke scale were 0.72 (95% CI, 0.68-0.75) and 0.77 (95% CI, 0.72-0.81), respectively; summary receiver operating characteristic curve AUC was 0.80 (95% CI, 0.76-0.83). Pooled sensitivity for any ML model for LVO was 0.73 (95% CI, 0.64-0.79), specificity was 0.85 (95% CI, 0.80-0.89), and summary receiver operating characteristic curve AUC was 0.87 (95% CI, 0.83-0.89). CONCLUSIONS: Both prehospital stroke scales and ML models demonstrated varying accuracies in predicting LVO. Despite ML potential for improved LVO detection in the prehospital setting, application remains limited by the absence of prospective external validation, limited sample sizes, and lack of real-world performance data in a prehospital setting.

Linking Vascular Structure and Function: Image-Based Virtual Populations of the Retina.

Purpose: This study explored the relationship among microvascular parameters as delineated by optical coherence tomography angiography (OCTA) and retinal perfusion. Here, we introduce a versatile framework to examine the interplay between the retinal vascular structure and function by generating virtual vasculatures from central retinal vessels to macular capillaries. Also, we have developed a hemodynamics model that evaluates the associations between vascular morphology and retinal perfusion. Methods: The generation of the vasculature is based on the distribution of four clinical parameters pertaining to the dimension and blood pressure of the central retinal vessels, constructive constrained optimization, and Voronoi diagrams. Arterial and venous trees are generated in the temporal retina and connected through three layers of capillaries at different depths in the macula. The correlations between total retinal blood flow and macular flow fraction and vascular morphology are derived as Spearman rank coefficients, and uncertainty from input parameters is quantified. Results: A virtual cohort of 200 healthy vasculatures was generated. Means and standard deviations for retinal blood flow and macular flow fraction were 20.80 ± 7.86 µL/min and 15.04% ± 5.42%, respectively. Retinal blood flow was correlated with vessel area density, vessel diameter index, fractal dimension, and vessel caliber index. The macular flow fraction was not correlated with any morphological metrics. Conclusions: The proposed framework is able to reproduce vascular networks in the macula that are morphologically and functionally similar to real vasculature. The framework provides quantitative insights into how macular perfusion can be affected by changes in vascular morphology delineated on OCTA.

Predicting acute and long-term mortality in a cohort of pulmonary embolism patients using machine learning.

BACKGROUND: Pulmonary embolism (PE) is a severe condition that causes significant mortality and morbidity. Due to its acute nature, scores have been developed to stratify patients at high risk of 30-day mortality. Here we develop a machine-learning based score to predict 30-day, 90-day, and 365-day mortality in PE patients. METHODS: The Birmingham and Black Country Venous Thromboembolism registry (BBC-VTE) of 2183 venous thromboembolism patients is used. Random forests were trained on a 70% training cohort and tested against 30% held-out set. The outcomes of interest were 30-day, 90-day, and 365-day mortality. These were compared to the pulmonary embolism severity index (PESI) and simplified pulmonary embolism severity index (sPESI). Shapley values were used to determine important predictors. Oral anticoagulation at discharge was also investigated as a predictor of mortality. RESULTS: The machine learning risk score predicted 30-day mortality with AUC 0.71 [95% CI: 0.63 - 0.78] compared to the sPESI AUC of 0.65 [95% CI: 0.57 - 0.73] and PESI AUC of 0.64 [95% CI: 0.56 - 0.72]. 90-day mortality and 365-day mortality were predicted with an AUC of 0.74 and 0.73 respectively. High counts of neutrophils, white blood cell counts, and c-reactive protein and low counts of haemoglobin were important for 30-day mortality prediction but progressively lost importance with time. Older age was an important predictor of high risk throughout. CONCLUSION: Machine learning algorithms have improved on standard clinical risk stratification for PE patients. External cohort validation is required before incorporation into clinical workflows.

Regional Cerebral Blood Flow Changes in Healthy Ageing and Alzheimer's Disease: A Narrative Review.

BACKGROUND: Cerebral blood flow is known to decline with increasing age and is a potential biomarker to distinguish between healthy and unhealthy ageing, where healthy ageing is defined as an absence of comorbidities in senescence. This review aims to synthesize evidence of cerebral blood flow changes over multiple brain regions, for use as a clinical reference or for in silico modelling. SUMMARY: The search identified 1,087 studies, of which 33 met the inclusion criteria to map the difference in cerebral blood flow reduction between healthy ageing and Alzheimer's disease. Analysis was also performed on the effect of imaging modality and brain region functionality as potential confounding factors. KEY MESSAGES: No significant difference was found between the specific functionality of a brain region and cerebral blood flow in healthy ageing (p = 0.65) or Alzheimer's disease (p = 0.42). Arterial spin labelling MRI imaging was shown to measure statistically larger decreases in flow in both healthy ageing (p = 0.0001) and Alzheimer's disease (p = 0.0465). Cerebral blood flow was shown to decrease 0.3-0.5% per year in healthy ageing, which increased to a decline of 2-5% per year in Alzheimer's disease. There was large variability both between and within individual brain regions, and this variability increased greatly in Alzheimer's disease. Future studies would add value by taking more cerebral blood flow measurements during Alzheimer's disease progression and by investigating ageing with comorbidities such as hypertension.

Pulsatile tympanic membrane displacement is associated with cognitive score in healthy subjects.

To test the hypothesis that pulsing of intracranial pressure has an association with cognition, we measured cognitive score and pulsing of the tympanic membrane in 290 healthy subjects. This hypothesis was formed on the assumptions that large intracranial pressure pulses impair cognitive performance and tympanic membrane pulses reflect intracranial pressure pulses. 290 healthy subjects, aged 20-80 years, completed the Montreal Cognitive Assessment Test. Spontaneous tympanic membrane displacement during a heart cycle was measured from both ears in the sitting and supine position. We applied multiple linear regression, correcting for age, heart rate, and height, to test for an association between cognitive score and spontaneous tympanic membrane displacement. Significance was set at P < 0.0125 (Bonferroni correction.) A significant association was seen in the left supine position (p = 0.0076.) The association was not significant in the right ear supine (p = 0.28) or in either ear while sitting. Sub-domains of the cognitive assessment revealed that executive function, language and memory have been primarily responsible for this association. In conclusion, we have found that spontaneous pulses of the tympanic membrane are associated with cognitive performance and believe this reflects an association between cognitive performance and intracranial pressure pulses.

Mathematical modelling of haemorrhagic transformation within a multiscale microvasculature network.

Objective.Haemorrhagic transformation (HT) is one of the most common complications after ischaemic stroke, caused by damage to the blood-brain barrier (BBB) that could be the result of stroke progression or a complication of stroke treatment with reperfusion therapy. The aim of this study is to develop further a previous simple HT mathematical model into an enlarged multiscale microvasculature model in order to investigate the effects of HT on the surrounding tissue and vasculature. In addition, this study investigates the relationship between tissue displacement and vascular geometry.Approach.By modelling tissue displacement, capillary compression, hydraulic conductivity in tissue and vascular permeability, we establish a mathematical model to describe the change of intracranial pressure (ICP) surrounding the damaged vascular bed after HT onset, applied to a 3D multiscale microvasculature. The use of a voxel-scale model then enables us to compare our HT simulation with available clinical imaging data for perfusion and cerebral blood volume (CBV) in the multiscale microvasculature network.Main results. We showed that the haematoma diameter and the maximum tissue displacement are approximately proportional to the diameter of the breakdown vessel. Based on the voxel-scale model, we found that perfusion reduces by approximately13-17%andCBVreduces by around20-25%after HT onset due to the effect of capillary compression caused by increased interstitial pressure. The results are in good agreement with the limited experimental data.Significance. This model, by enabling us to bridge the gap between the microvascular scale and clinically measurable parameters, providing a foundation for more detailed validation and understanding of HT in patients.

The Birmingham and Black Country cohort of Venous Thromboembolism (BBC-VTE) registry: Rationale, design and preliminary results.

The Birmingham Black Country Venous Thromboembolism registry (BBC-VTE) is a multi-ethnic cohort of patients who suffered a first episode of venous thromboembolism (VTE) and were admitted to various hospital sites across the West Midlands and Black Country regions in the United Kingdom. The BBC-VTE registry is a retrospective, observational cohort study which aims to collect data on outcomes including mortality, bleeding and VTE recurrence in this patient cohort. In addition, the comprehensive, structured data collected will allow us to conduct machine learning analyses for risk prediction in such patients and also to compare to previously derived mortality scores such as the PESI and the simplified PESI (sPESI). Our registry included 2183 patients admitted to hospital between the years 2012-14 and 2016-18 with a first episode of VTE and the mean follow up was 36 months. The cohort was ethnically diverse with 72.5% white Caucasian, 8.2% Asian (including South Asian), 6.7% black, and 11.7% of unknown/other ethnicity. Of those admitted during the collection period 56% had PE, 40% had DVT, with the rest presenting with both PE and DVT. Around 7% of patients went on to develop a bleeding episode and 36% died (all-cause mortality). Of the deaths, 10% of patients died within 30-days of admission (30-day mortality), with 16% dying within 90 days. In summary, this study investigates real-world outcomes of patients after the first index VTE event and attempts to bridge the gap in evidence for contemporary data in this population which will allow to construct more accurate risk prediction tools and management decisions.

The Ageing Brain: Investigating the Role of Age in Changes to the Human Cerebral Microvasculature With an in silico Model.

Ageing causes extensive structural changes to the human cerebral microvasculature, which have a significant effect on capillary bed perfusion and oxygen transport. Current models of brain capillary networks in the literature focus on healthy adult brains and do not capture the effects of ageing, which is critical when studying neurodegenerative diseases. This study builds upon a statistically accurate model of the human cerebral microvasculature based on ex-vivo morphological data. This model is adapted for "healthy" ageing using in-vivo measurements from mice at three distinct age groups-young, middle-aged, and old. From this new model, blood and molecular exchange parameters are calculated such as permeability and surface-area-to-volume ratio, and compared across the three age groups. The ability to alter the model vessel-by-vessel is used to create a continuous gradient of ageing. It was found that surface-area-to-volume ratio reduced in old age by 6% and permeability by 24% from middle-age to old age, and variability within the networks also increased with age. The ageing gradient indicated a threshold in the ageing process around 75 years old, after which small changes have an amplified effect on blood flow properties. This gradient enables comparison of studies measuring cerebral properties at discrete points in time. The response of middle aged and old aged capillary beds to micro-emboli showed a lower robustness of the old age capillary bed to vessel occlusion. As the brain ages, there is thus increased vulnerability of the microvasculature-with a "tipping point" beyond which further remodeling of the microvasculature has exaggerated effects on the brain. When developing in-silico models of the brain, age is a very important consideration to accurately assess risk factors for cognitive decline and isolate early biomarkers of microvascular health.

Modelling the effects of cerebral microthrombi on tissue oxygenation and cell death.

Thrombectomy, the mechanical removal of a clot, is the most common way to treat ischaemic stroke with large vessel occlusions. However, perfusion cannot always be restored after such an intervention. It has been hypothesised that the absence of reperfusion is at least partially due to the clot fragments that block the downstream vessels. In this paper, we present a new way of quantifying the effects of cerebral microthrombi on oxygen transport to tissue in terms of hypoxia and ischaemia. The oxygen transport was simulated with the Green's function method on physiologically representative microvascular cubes, which was found independent of both microvascular geometry and length scale. The microthrombi occlusions were then simulated in the microvasculature, which were extravasated over time with a new thrombus extravasation model. The tissue hypoxic fraction was fitted as a sigmoidal function of vessel blockage fraction, which was then taken to be a function of time after the formation of microthrombi occlusions. A novel hypoxia-based 3-state cell death model was finally proposed to simulate the hypoxic tissue damage over time. Using the cell death model, the impact of a certain degree of microthrombi occlusions on tissue viability and microinfarct volume can be predicted over time. Quantifying the impact of microthrombi on oxygen transport and tissue death will play an important role in full brain models of ischaemic stroke and thrombectomy.

Modelling the impact of clot fragmentation on the microcirculation after thrombectomy.

Many ischaemic stroke patients who have a mechanical removal of their clot (thrombectomy) do not get reperfusion of tissue despite the thrombus being removed. One hypothesis for this 'no-reperfusion' phenomenon is micro-emboli fragmenting off the large clot during thrombectomy and occluding smaller blood vessels downstream of the clot location. This is impossible to observe in-vivo and so we here develop an in-silico model based on in-vitro experiments to model the effect of micro-emboli on brain tissue. Through in-vitro experiments we obtain, under a variety of clot consistencies and thrombectomy techniques, micro-emboli distributions post-thrombectomy. Blood flow through the microcirculation is modelled for statistically accurate voxels of brain microvasculature including penetrating arterioles and capillary beds. A novel micro-emboli algorithm, informed by the experimental data, is used to simulate the impact of micro-emboli successively entering the penetrating arterioles and the capillary bed. Scaled-up blood flow parameters-permeability and coupling coefficients-are calculated under various conditions. We find that capillary beds are more susceptible to occlusions than the penetrating arterioles with a 4x greater drop in permeability per volume of vessel occluded. Individual microvascular geometries determine robustness to micro-emboli. Hard clot fragmentation leads to larger micro-emboli and larger drops in blood flow for a given number of micro-emboli. Thrombectomy technique has a large impact on clot fragmentation and hence occlusions in the microvasculature. As such, in-silico modelling of mechanical thrombectomy predicts that clot specific factors, interventional technique, and microvascular geometry strongly influence reperfusion of the brain. Micro-emboli are likely contributory to the phenomenon of no-reperfusion following successful removal of a major clot.

Coupling one-dimensional arterial blood flow to three-dimensional tissue perfusion models for in silico trials of acute ischaemic stroke.

An acute ischaemic stroke is due to the sudden blockage of an intracranial blood vessel by an embolized thrombus. In the context of setting up in silico trials for the treatment of acute ischaemic stroke, the effect of a stroke on perfusion and metabolism of brain tissue should be modelled to predict final infarcted brain tissue. This requires coupling of blood flow and tissue perfusion models. A one-dimensional intracranial blood flow model and a method to couple this to a brain tissue perfusion model for patient-specific simulations is presented. Image-based patient-specific data on the anatomy of the circle of Willis are combined with literature data and models for vessel anatomy not visible in the images, to create an extended model for each patient from the larger vessels down to the pial surface. The coupling between arterial blood flow and tissue perfusion occurs at the pial surface through the estimation of perfusion territories. The coupling method is able to accurately estimate perfusion territories. Finally, we argue that blood flow can be approximated as steady-state flow at the interface between arterial blood flow and tissue perfusion to reduce the cost of organ-scale simulations.

Dynamic Changes in Microvascular Flow Conductivity and Perfusion After Myocardial Infarction Shown by Image-Based Modeling.

Background Microcirculation is a decisive factor in tissue reperfusion inadequacy following myocardial infarction ( MI ). Nonetheless, experimental assessment of blood flow in microcirculation remains a bottleneck. We sought to model blood flow properties in coronary microcirculation at different time points after MI and to compare them with healthy conditions to obtain insights into alterations in cardiac tissue perfusion. Methods and Results We developed an image-based modeling framework that permitted feeding a continuum flow model with anatomical data previously obtained from the pig coronary microvasculature to calculate physiologically meaningful permeability tensors. The tensors encompassed the microvascular conductivity and were also used to estimate the arteriole-venule drop in pressure and myocardial blood flow. Our results indicate that the tensors increased in a bimodal pattern at infarcted areas on days 1 and 7 after MI while a nonphysiological decrease in arteriole-venule drop in pressure was observed; contrary, the tensors and the arteriole-venule drop in pressure on day 3 after MI , and in remote areas, were closer to values for healthy tissue. Myocardial blood flow calculated using the condition-dependent arteriole-venule drop in pressure decreased in infarcted areas. Last, we simulated specific modes of vascular remodeling, such as vasodilation, vasoconstriction, or pruning, and quantified their distinct impact on microvascular conductivity. Conclusions Our study unravels time- and region-dependent alterations of tissue perfusion related to the structural changes occurring in the coronary microvasculature due to MI . It also paves the way for conducting simulations in new therapeutic interventions in MI and for image-based microvascular modeling by applying continuum flow models in other biomedical scenarios.

Quantifying the contribution of intracranial pressure and arterial blood pressure to spontaneous tympanic membrane displacement.

OBJECTIVE: Although previous studies have shown associations between patient symptoms/outcomes and the spontaneous tympanic membrane displacement (spTMD) pulse amplitude, the contribution of the underlying intracranial pressure (ICP) signal to the spTMD pulse remains largely unknown. We have assessed the relative contributions of ICP and arterial blood pressure (ABP) on spTMD at different frequencies in order to determine whether spTMD contains information about the ICP above and beyond that contained in the ABP. APPROACH: Eleven patients, who all had invasive ICP and ABP measurements in situ, were recruited from our intensive care unit. Their spTMD was recorded and the power spectral densities of the three signals, as well as coherences between the signals, were calculated in the range 0.1-5 Hz. Simple and multiple coherences, coupled with statistical tests using surrogate data, were carried out to quantify the relative contributions of ABP and ICP to spTMD. MAIN RESULTS: Most power of the signals was found to predominate at respiration rate, heart rate, and their harmonics, with little outside of these frequencies. Analysis of the simple coherences found a slight preference for ICP transmission, beyond that from ABP, to the spTMD at lower frequencies (7/11 patients at respiration, 7/10 patients at respiration 1st harmonic) which is reversed at the higher frequencies (2/11 patients at heart rate and its 1st harmonic). Both ICP and ABP were found to independently contribute to the spTMD. The multiple coherence reinforced that ICP is preferentially being transmitted at respiration and respiration 1st harmonic. SIGNIFICANCE: Both ABP and ICP contribute independently to the spTMD signal, with most power occurring at clear physiological frequencies-respiration and harmonics and heart rate and harmonics. There is information shared between the ICP and spTMD that is not present in ABP. This analysis has indicated that lower frequencies appear to favour ICP as the driver for spTMD.

Investigating the effects of a penetrating vessel occlusion with a multi-scale microvasculature model of the human cerebral cortex.

The effect of the microvasculature on observed clinical parameters, such as cerebral blood flow, is poorly understood. This is partly due to the gap between the vessels that can be individually imaged in humans and the microvasculature, meaning that mathematical models are required to understand the role of the microvasculature. As a result, a multi-scale model based on morphological data was developed here that is able to model large regions of the human microvasculature. From this model, a clear layering of flow (and 1-dimensional depth profiles) was observed within a voxel, with the flow in the microvasculature being driven predominantly by the geometry of the penetrating vessels. It also appears that the pressure and flow are decoupled, both in healthy vasculatures and in those where occlusions have occurred, again due to the topology of the penetrating vessels shunting flow between them. Occlusion of a penetrating arteriole resulted in a very high degree of overlap of blood pressure drop with experimentally observed cell death. However, drops in blood flow were far more widespread, providing additional support for the theory that pericyte controlled regulation on the capillary scale likely plays a large part in the perfusion of tissue post-occlusion.

A statistical model of the penetrating arterioles and venules in the human cerebral cortex.

OBJECTIVE: Models of the cerebral microvasculature are required at many different scales in order to understand the effects of microvascular topology on CBF. There are, however, no data-driven models at the arteriolar/venular scale. In this paper, we develop a data-driven algorithm based on available data to generate statistically accurate penetrating arterioles and venules. METHODS: A novel order-based density-filling algorithm is developed based on the statistical data including bifurcating angles, LDRs, and area ratios. Three thousand simulations are presented, and the results validated against morphological data. These are combined with a previous capillary network in order to calculate full vascular network parameters. RESULTS: Statistically accurate penetrating trees were successfully generated. All properties provided a good fit to experimental data. The k exponent had a median of 2.5 and an interquartile range of 1.75-3.7. CBF showed a standard deviation ranging from ±18% to ±34% of the mean, depending on the penetrating vessel diameter. CONCLUSIONS: Small CBF variations indicate that the topology of the penetrating vessels plays only a small part in the large regional variations of CBF seen in the brain. These results open up the possibility of efficient oxygen and blood flow simulations at MRI voxel scales which can be directly validated against MRI data.

Multi-scale homogenization of blood flow in 3-dimensional human cerebral microvascular networks.

The microvasculature plays a crucial role in the perfusion of blood through cerebral tissue. Current models of the cerebral microvasculature are discrete, and hence only able to model the perfusion over small voxel sizes before becoming computationally prohibitive. Larger models are required to provide comparisons and validation against imaging data. In this work, multi-scale homogenization methods were employed to develop continuum models of blood flow in a capillary network model of the human cortex. Homogenization of the local scale blood flow equations produced an averaged form of Darcy׳s law, with the permeability tensor encapsulating the capillary bed topology. A statistically accurate network model of the human cortex microvasculature was adapted to impose periodicity, and the elements of the permeability tensor calculated over a range of voxel sizes. The permeability tensor was found to converge to an effective permeability as voxel size increased. This converged permeability tensor was isotropic, reflecting the mesh-like structure of the cerebral microvasculature, with off-diagonal terms normally distributed about zero. A representative elementary volume of 375µm, with a standard deviation of 4.5% from the effective permeability, was determined. Using the converged permeability values, the cerebral blood flow was calculated to be around 55mLmin(-1)100g(-1), which is in very close agreement with experimental values. These results open up the possibility of future multi-scale modeling of the cerebral vascular network.