Convolutional Neural Networks to Assess Steno-Occlusive Disease Using Cerebrovascular Reactivity

Cerebrovascular Reactivity (CVR) is a provocative test used with Blood oxygenation level-dependent (BOLD) Magnetic Resonance Imaging (MRI) studies, where a vasoactive stimulus is applied and the corresponding changes in the cerebral blood flow (CBF) are measured. The most common clinical application is the assessment of cerebral perfusion insufficiency in patients with steno-occlusive disease (SOD). Globally, millions of people suffer from cerebrovascular diseases, and SOD is the most common cause of ischemic stroke. Therefore, CVR analyses can play a vital role in early diagnosis and guiding clinical treatment. This study develops a convolutional neural network (CNN)-based clinical decision support system to facilitate the screening of SOD patients by discriminating between healthy and unhealthy CVR maps. The networks were trained on a confidential CVR dataset with two classes: 68 healthy control subjects, and 163 SOD patients. This original dataset was distributed in a ratio of 80%-10%-10% for training, validation, and testing, respectively, and image augmentations were applied to the training and validation sets. Additionally, some popular pre-trained networks were imported and customized for the objective classification task to conduct transfer learning experiments. Results indicate that a customized CNN with a double-stacked convolution layer architecture produces the best results, consistent with expert clinical readings.

A dual-center validation of the PIRAMD scoring system for assessing the severity of ischemic Moyamoya disease

Background

Prior Infarcts, Reactivity, and Angiography in Moyamoya Disease (PIRAMD) is a recently proposed imaging-based scoring system that incorporates the severity of disease and its impact on parenchymal hemodynamics in order to better support clinical management and evaluate response to intervention. In particular, PIRAMD may have merit in identifying symptomatic patients that may benefit most from revascularization. Our aim was to validate the PIRAMD scoring system.

Methods

Patients with ischemic Moyamoya disease, who underwent catheter angiographic [modified Suzuki Score (mSS) and collateralization status], morphological MRI and a parenchymal hemodynamic evaluation with blood oxygenation-level dependent cerebrovascular reactivity (BOLD-CVR) at two transatlantic centers, were retrospectively included. The primary outcome was the presence of neurological symptoms. The diagnostic capacity of each PIRAMD feature alone was evaluated, as well as combined and the inter-institutional differences of each parameter were evaluated.

Results

Seventy-two hemispheres of 38 patients were considered for analysis, of which 39 (54%) were classified as symptomatic. The presence of a prior infarct had the highest odds ratio [odds ratio (OR) =24; 95% CI: 6.7-87.2] for having neurological symptoms, followed by impaired CVR (OR =17; 95% CI: 5-62). No inter-institutional differences in the odds ratios or area under the curve (AUC) were found for any study parameter. The PIRAMD score had an AUC of 0.88 (95% CI: 0.80-0.96) with a similar AUC for the PIRAMD grading score.

Conclusions

Our multicentric validation of the recently published PIRAMD scoring system was highly effective in rating the severity of ischemic Moyamoya disease with excellent inter-institutional agreement. Future studies should investigate the prognostic value of this novel imaging-based score in symptomatic patients with Moyamoya disease.

Dependence of resting-state-based cerebrovascular reactivity (CVR) mapping on spatial resolution

Cerebrovascular reactivity (CVR) is typically assessed with a carbon dioxide (CO2) stimulus combined with BOLD fMRI. Recently, resting-state (RS) BOLD fMRI has been shown capable of generating CVR maps, providing a potential for broader CVR applications in neuroimaging studies. However, prior RS-CVR studies have primarily been performed at a spatial resolution of 3-4 mm voxel sizes. It remains unknown whether RS-CVR can also be obtained at high-resolution without major degradation in image quality. In this study, we investigated RS-CVR mapping based on resting-state BOLD MRI across a range of spatial resolutions in a group of healthy subjects, in an effort to examine the feasibility of RS-CVR measurement at high resolution. Comparing the results of RS-CVR with the maps obtained by the conventional CO2-inhalation method, our results suggested that good CVR map quality can be obtained at a voxel size as small as 2 mm isotropic. Our results also showed that, RS-CVR maps revealed resolution-dependent sensitivity. However, even at a high resolution of 2 mm isotropic voxel size, the voxel-wise sensitivity is still greater than that of typical task-evoked fMRI. Scan duration affected the sensitivity of RS-CVR mapping, but had no significant effect on its accuracy. These findings suggest that RS-CVR mapping can be applied at a similar resolution as state-of-the-art fMRI studies, which will broaden the use of CVR mapping in basic science and clinical applications including retrospective analysis of previously collected fMRI data.

Generating dynamic carbon-dioxide traces from respiration-belt recordings: Feasibility using neural networks and application in functional magnetic resonance imaging

INTRODUCTION

In the context of functional magnetic resonance imaging (fMRI), carbon dioxide (CO2) is a well-known vasodilator that has been widely used to monitor and interrogate vascular physiology. Moreover, spontaneous fluctuations in end-tidal carbon dioxide (PETCO2) reflects changes in arterial CO2 and has been demonstrated as the largest physiological noise source for denoising the low-frequency range of the resting-state fMRI (rs-fMRI) signal. However, the majority of rs-fMRI studies do not involve CO2 recordings, and most often only heart rate and respiration are recorded. While the intrinsic link between these latter metrics and CO2 led to suggested possible analytical models, they have not been widely applied.

METHODS

In this proof-of-concept study, we propose a deep-learning (DL) approach to reconstruct CO2 and PETCO2 data from respiration waveforms in the resting state.

RESULTS

We demonstrate that the one-to-one mapping between respiration and CO2 recordings can be well predicted using fully convolutional networks (FCNs), achieving a Pearson correlation coefficient (r) of 0.946 ± 0.056 with the ground truth CO2. Moreover, dynamic PETCO2 can be successfully derived from the predicted CO2, achieving r of 0.512 ± 0.269 with the ground truth. Importantly, the FCN-based methods outperform previously proposed analytical methods. In addition, we provide guidelines for quality assurance of respiration recordings for the purposes of CO2 prediction.

DISCUSSION

Our results demonstrate that dynamic CO2 can be obtained from respiration-volume using neural networks, complementing the still few reports in DL of physiological fMRI signals, and paving the way for further research in DL based bio-signal processing.

Sinusoidal CO2 respiratory challenge for concurrent perfusion and cerebrovascular reactivity MRI

Introduction: Deoxygenation-based dynamic susceptibility contrast (dDSC) has previously leveraged respiratory challenges to modulate blood oxygen content as an endogenous source of contrast alternative to gadolinium injection in perfusion-weighted MRI. This work proposed the use of sinusoidal modulation of end-tidal CO2 pressures (SineCO 2 ), which has previously been used to measure cerebrovascular reactivity, to induce susceptibility-weighted gradient-echo signal loss to measure brain perfusion. Methods: SineCO 2 was performed in 10 healthy volunteers (age 37 ± 11, 60% female), and tracer kinetics model was applied in the frequency domain to calculate cerebral blood flow, cerebral blood volume, mean transit time, and temporal delay. These perfusion estimates were compared against reference techniques, including gadolinium-based DSC, arterial spin labeling, and phase contrast. Results: Our results showed regional agreement between SineCO 2 and the clinical comparators. SineCO 2 was able to generate robust CVR maps in conjunction to baseline perfusion estimates. Discussion: Overall, this work demonstrated feasibility of using sinusoidal CO2 respiratory paradigm to simultaneously acquire both cerebral perfusion and cerebrovascular reactivity maps in one imaging sequence.

Cerebrovascular Reactivity Mapping Using Resting-State Functional MRI in Patients With Gliomas

BACKGROUND

Recently, a data-driven regression analysis method was developed to utilize the resting-state (rs) blood oxygenation level-dependent signal for cerebrovascular reactivity (CVR) mapping (rs-CVR), which was previously optimized by comparing with the CO₂ inhalation-based method in health subjects and patients with neurovascular diseases.

PURPOSE

To investigate the agreement of rs-CVR and the CVR mapping with breath-hold MRI (bh-CVR) in patients with gliomas.

STUDY TYPE

Retrospective.

POPULATION

Twenty-five patients (12 males, 13 females; mean age ± SD, 48 ± 13 years) with gliomas.

FIELD STRENGTH/SEQUENCE

Dynamic T2*-weighted gradient-echo echo-planar imaging during a breath-hold paradigm and during the rs on a 3-T scanner.

ASSESSMENT

rs-CVR with various frequency ranges and resting-state fluctuation amplitude (RSFA) were assessed. The agreement between each rs-based CVR measurement and bh-CVR was determined by voxel-wise correlation and Dice coefficient in the whole brain, gray matter, and the lesion region of interest (ROI).

STATISTICAL TESTS

Voxel-wise Pearson correlation, Dice coefficient, Fisher Z-transformation, repeated-measure analysis of variance and post hoc test with Bonferroni correction, and nonparametric repeated-measure Friedman test and post hoc test with Bonferroni correction were used. Significance was set at P < 0.05.

RESULTS

Compared with bh-CVR, the highest correlations were found at the frequency bands of 0.04-0.08 Hz and 0.02-0.04 Hz for rs-CVR in both whole brain and the lesion ROI. RSFA had significantly lower correlations than did rs-CVR of 0.02-0.04 Hz and a wider frequency range (0-0.1164 Hz). Significantly higher correlations and Dice coefficient were found in normal tissues than in the lesion ROI for all three methods.

DATA CONCLUSION

The optimal frequency ranges for rs-CVR are determined by comparing with bh-CVR in patients with gliomas. The rs-CVR method outperformed the RSFA. Significantly higher correlation and Dice coefficient between rs- and bh-CVR were found in normal tissue than in the lesion.

LEVEL OF EVIDENCE

3 TECHNICAL EFFICACY STAGE: 2.

Higher body mass index is associated with worse hippocampal vasoreactivity to carbon dioxide

Background and objectives

Obesity is a risk factor for cognitive decline. Probable mechanisms involve inflammation and cerebrovascular dysfunction, leading to diminished cerebral blood flow (CBF) and cerebrovascular reactivity (CVR). The hippocampus, crucially involved in memory processing and thus relevant to many types of dementia, poses a challenge in studies of perfusion and CVR, due to its location, small size, and complex shape. We examined the relationships between body mass index (BMI) and hippocampal resting CBF and CVR to carbon dioxide (CVRCO2) in a group of cognitively normal middle-aged and older adults.

Methods

Our study was a retrospective analysis of prospectively collected data. Subjects were enrolled for studies assessing the role of hippocampal hemodynamics as a biomarker for AD among cognitively healthy elderly individuals (age > 50). Participants without cognitive impairment, stroke, and active substance abuse were recruited between January 2008 and November 2017 at the NYU Grossman School of Medicine, former Center for Brain Health. All subjects underwent medical, psychiatric, and neurological assessments, blood tests, and MRI examinations. To estimate CVR, we increased their carbon dioxide levels using a rebreathing protocol. Relationships between BMI and brain measures were tested using linear regression.

Results

Our group (n = 331) consisted of 60.4% women (age 68.8 ± 7.5 years; education 16.8 ± 2.2 years) and 39.6% men (age 70.4 ± 6.4 years; education 16.9 ± 2.4 years). Approximately 22% of them (n = 73) were obese. BMI was inversely associated with CVRCO2 (β = -0.12, unstandardized B = -0.06, 95% CI -0.11, -0.004). A similar relationship was observed after excluding subjects with diabetes and insulin resistance (β = -0.15, unstandardized B = -0.08, 95% CI -0.16, -0.000). In the entire group, BMI was more strongly related to hippocampal CVRCO2 in women (β = -0.20, unstandardized B = -0.08, 95% CI -0.13, -0.02).

Discussion

These findings lend support to the notion that obesity is a risk factor for hippocampal hemodynamic impairment and suggest targeting obesity as an important prevention strategy. Prospective studies assessing the effects of weight loss on brain hemodynamic measures and inflammation are warranted.

Cerebrovascular Reactivity and Neurovascular Coupling in Multiple Sclerosis-A Systematic Review

The inflammatory processes observed in the central nervous system in multiple sclerosis (MS) could damage the endothelium of the cerebral vessels and lead to a dysfunctional regulation of vessel tonus and recruitment, potentially impairing cerebrovascular reactivity (CVR) and neurovascular coupling (NVC). Impaired CVR or NVC correlates with declining brain health and potentially plays a causal role in the development of neurodegenerative disease. Therefore, we examined studies on CVR or NVC in MS patients to evaluate the evidence for impaired cerebrovascular function as a contributing disease mechanism in MS. Twenty-three studies were included (12 examined CVR and 11 examined NVC). Six studies found no difference in CVR response between MS patients and healthy controls. Five studies observed reduced CVR in patients. This discrepancy can be because CVR is mainly affected after a long disease duration and therefore is not observed in all patients. All studies used CO₂ as a vasodilating stimulus. The studies on NVC demonstrated diverse results; hence a conclusion that describes all the published observations is difficult to find. Future studies using quantitative techniques and larger study samples are needed to elucidate the discrepancies in the reported results.

Cerebral Vasoreactivity Changes Over Time in Patients With Different Clinical Manifestations of Cerebral Small Vessel Disease

Objectives: Endothelial dysfunction (ED) has been linked to the pathogenesis of cerebral small vessel disease (SVD). We aimed to assess ED and cerebrovascular reactivity (CVR) in the patients with a diverse manifestation of SVD, with similar and extensive white matter lesions (WMLs, modified Fazekas scale grade ≥2), compared with a control group (CG) without the MRI markers of SVD, matched for age, gender, hypertension, diabetes, and to evaluate the change of CVR following 24 months. Methods: We repeatedly measured the vasomotor reactivity reserve (VMRr) and breath-holding index (BHI) of the middle cerebral artery (MCA) by the transcranial Doppler ultrasound (TCD) techniques in 60 subjects above 60 years with a history of lacunar stroke (LS), vascular dementia (VaD), or parkinsonism (VaP) (20 in each group), and in 20 individuals from a CG. Results: The mean age, frequency of the main vascular risk factors, and sex distribution were similar in the patients with the SVD groups and a CG. The VMRr and the BHI were more severely impaired at baseline (respectively, 56.7 ± 18% and 0.82 ± 0.39) and at follow-up (respectively, 52.3 ± 16.7% and 0.71 ± 0.38) in the patients with SVD regardless of the clinical manifestations (ANOVA, p > 0.1) than in the CG (respectively, baseline VMRr 77.2 ± 15.6%, BHI 1.15 ± 0.47, p < 0.001; follow-up VMRr 74.3 ± 17.6%, BHI 1.11 ± 0.4, p < 0.001). All the assessed CVR measures (VMRr and BHI) significantly decreased over time in the subjects with SVD (Wilcoxon's signed-rank test p = 0.01), but this was not observed in the CG (p > 0.1) and the decrease of CVR measures was not related to the SVD radiological progression (p > 0.1). Conclusions: This study provided evidence that the change in CVR measures is detectable over a 24-month period in patients with different clinical manifestations of SVD. Compared with the patients in CG with similar atherothrombotic risk factors, all the CVR measures (BMRr and BHI) significantly declined over time in the subjects with SVD. The reduction in CVR was not related to the SVD radiological progression.

Cerebrovascular Resistance in Healthy Aging and Mild Cognitive Impairment

Measures of cerebrovascular reactivity (CVR) are used to judge the health of the brain vasculature. In this study, we report the use of several different analyses of blood oxygen dependent (BOLD) fMRI responses to CO₂ to provide a number of metrics of CVR based on the sigmoidal resistance response to CO₂. To assess possible differences in these metrics with age, we compiled atlases reflecting voxel-wise means and standard deviations for four different age ranges and for a group of patients with mild cognitive impairment (MCI) and compared them. Sixty-seven subjects were recruited for this study and scanned at 3T field strength. Of those, 51 healthy control volunteers between the ages of 18–83 were recruited, and 16 (MCI) subjects between the ages of 61–83 were recruited. Testing was carried out using an automated computer-controlled gas blender to induce hypercapnia in a step and ramp paradigm while monitoring end-tidal partial pressures of CO₂. Surprisingly, some resistance sigmoid parameters in the oldest control group were increased compared to the youngest control group. Resistance amplitude maps showed increases in clusters within the temporal cortex, thalamus, corpus callosum and brainstem, and resistance reserve maps showed increases in clusters within the cingulate cortex, frontal gyrus, and corpus callosum. These findings suggest that some aspects of vascular reactivity in parts of the brain are initially maintained with age but then may increase in later years. We found significant reductions in all resistance sigmoid parameters (amplitude, reserve, sensitivity, midpoint, and range) when comparing MCI patients to controls. Additionally, in controls and in MCI patients, amplitude, range, reserve, and sensitivity in white matter (WM) was significantly reduced compared to gray matter (GM). WM midpoints were significantly above those of GM. Our general conclusion is that vascular regulation in terms of cerebral blood flow (CBF) responsiveness to CO₂ is not significantly affected by age, but is reduced in MCI. These changes in cerebrovascular regulation demonstrate the value of resistance metrics for mapping areas of dysregulated blood flow in individuals with MCI. They may also be of value in the investigation of patients with vascular risk factors at risk for developing vascular dementia.

Cerebrovascular-Reactivity Mapping Using MRI: Considerations for Alzheimer's Disease

Alzheimer’s disease (AD) is associated with well-established macrostructural and cellular markers, including localized brain atrophy and deposition of amyloid. However, there is growing recognition of the link between cerebrovascular dysfunction and AD, supported by continuous experimental evidence in the animal and human literature. As a result, neuroimaging studies of AD are increasingly aiming to incorporate vascular measures, exemplified by measures of cerebrovascular reactivity (CVR). CVR is a measure that is rooted in clinical practice, and as non-invasive CVR-mapping techniques become more widely available, routine CVR mapping may open up new avenues of investigation into the development of AD. This review focuses on the use of MRI to map CVR, paying specific attention to recent developments in MRI methodology and on the emerging stimulus-free approaches to CVR mapping. It also summarizes the biological basis for the vascular contribution to AD, and provides critical perspective on the choice of CVR-mapping techniques amongst frail populations.