Cerebrovascular Reactivity Mapping in Brain Tumors Based on a Breath‐Hold Task Using Arterial Spin Labeling

ABSTRACT

Hemodynamic measurements such as cerebral blood flow (CBF) and cerebrovascular reactivity (CVR) can provide useful information for the diagnosis and characterization of brain tumors. Previous work showed that arterial spin labeling (ASL) in combination with vasoactive stimulation enabled simultaneous non‐invasive evaluation of both parameters, however this approach had not been previously tested in tumors. The aim of this work was to investigate the application of this technique, using a pseudo‐continuous ASL (PCASL) sequence combined with breath‐holding at 3 T, to measure CBF and CVR in high‐grade gliomas and metastatic lesions, and to explore differences across tumoral‐peritumoral regions and tumor types. To that end, 27 patients with brain tumor were studied. Baseline CBF and CVR were measured in tumor, edema, and gray matter (GM) volumes‐of‐interest (VOIs). Peritumoral ipsilateral ring‐shaped VOIs were also generated and mirrored to the contralateral hemisphere. Differences in baseline CBF and CVR were evaluated between contralateral and ipsilateral GM, contralateral and ipsilateral peritumoral rings, and among VOIs and tumor types. CBF in the tumor was higher in grade 4 gliomas than metastases. In grade 4 gliomas, edema had lower CBF than the tumor and contralateral GM. CVR values were different between grade 3 and grade 4 gliomas, and between grade 4 and metastases. CVR values in the tumor were lower compared to the contralateral GM. Differences in CVR between contralateral and ipsilateral‐ring VOIs were also found in grade 4 gliomas, presumably suggesting tumor infiltration within the peritumoral tissue. A cut‐off value for CVR of 27.9%‐signal‐change is suggested to differentiate between grade 3 and grade 4 gliomas (specificity = 83.3%, sensitivity = 70.6%). In conclusion, CBF and CVR mapping with ASL offered insights into the perilesional environment that could help to detect infiltrative disease, particularly in grade 4 gliomas. CVR emerged as a potential biomarker to differentiate between grade 3 and grade 4 gliomas.

ASL‐based CVR mapping offered insights into the infiltrative perilesional environment, particularly in grade 4 gliomas, emerging as a potential biomarker to differentiate between grade 3 and grade 4 gliomas. A patient with glioblastoma WHO grade 4, with a lesion in the right temporoparietal region with ring enhancement, necrotic center (T1‐postcontrast), and moderate surrounding edema (T2‐FLAIR). CBF and CVR maps revealed high CBF within the contrast enhancing lesion (arrow) and impaired CVR that exceeded the enhancing lesion (dotted line).

Abbreviations

ANOVA

analysis of variance

ART

aligned rank transform

ASL

arterial spin labeling

AUC

area under the curve

BH

breath‐hold

BOLD

blood oxygen level dependent

CBF

cerebral blood flow

ce

contrast enhanced

CSF

cerebrospinal fluid

CVR

cerebrovascular reactivity

EGFR

epidermal growth factor receptor

FA

flip angle

FLAIR

fluid attenuated inversion recovery

FOV

field‐of‐view

GLM

generalized linear model

GM

gray matter

IDH

isocitrate dehydrogenase

IDH1

isoform 1 of the isocitrate dehydrogenase enzyme

MPRAGE

magnetization prepared rapid gradient echo

nCE

non‐contrast enhanced

NVU

neurovascular uncoupling

PCASL

pseudo‐continuous arterial spin labeling

PSC

percentage of signal change

RARE

rapid imaging with refocused echoes

ROC

receiver operating characteristics

SPM

statistical parametric mapping

TE

echo‐time

TERT

telomerase reverse transcriptase promoter

TOF

time‐of‐flight

TR

repetition‐time

VOI

volume‐of‐interest

WHO

World Health Organization

WM

white matter

1. Introduction

Tumors that arise within the brain can be classified as either primary or metastatic, based on the cell type from which the tumor originates. While metastatic tumors are the most common malignant brain tumors in adults [1], gliomas represent the most frequent among the primary malignant tumors [2], with devastating prognosis. Their differentiation has important clinical implications, as it influences patient outcomes and treatment management.

Even though histopathological analysis is the current reference standard for diagnosis and grading of gliomas, following the updated 2021 World Health Organization (WHO) classification [3], the ability to noninvasively differentiate gliomas from brain metastases or to predict primary tumor histology in the preoperative evaluation is important.

Preoperative MRI is an indispensable tool in the diagnostic process offering insights into the tumor location, size, extent and blood–brain‐barrier integrity. However, conventional protocols, including pre and postcontrast T1 and T2‐weighted images along with Fluid Attenuated Inversion Recovery (FLAIR), are insufficient to accurately provide a diagnostic assessment [4], since different types of tumors can display similar imaging features. Perfusion imaging by different methods has proven to be useful in this context. Specifically, the finding of hyperperfused areas beyond the area of contrast enhancement argues in favor of a malignant glioma rather than metastatic disease [5, 6, 7]. Moreover, conventional MRI has limitations in accurately delineating glioma boundaries, as it has been established that infiltrative margins usually extend beyond the contrast‐enhanced region on T1‐weighted MRI, making it difficult to differentiate tumoral tissue from the peritumoral edema and normal tissue [8].

Thus, challenges in brain tumor imaging persist, not only to establish biomarkers for accurate tumor diagnosis and grading, but also to precisely delineate the infiltrative glioma region for an accurate complete resection. Hemodynamic brain tumor imaging is an emerging field, resulting from numerous technical advances in the last years [9], that comprises techniques for imaging perfusion and cerebrovascular reactivity (CVR). In normal brain tissue, autoregulation and neurovascular coupling allow vessels to contract or dilate in response to physiologic stimuli. This ability, known as CVR, is compromised in brain diseases that provoke hemodynamic impairment and perfusion changes [10, 11], such as brain tumors. CVR has been proposed as a useful biomarker for differentiating tumor tissue from healthy tissue during the presurgical evaluation of brain tumors, and for detecting neurovascular uncoupling (NVU) in gliomas [12, 13, 14] even in lower grade gliomas in the absence of blood–brain‐barrier rupture [15, 16]. Its clinical potential encompasses not only the characterization of gliomas [17, 18, 19, 20], but also the identification of infiltrative glioma components, as this vascular impairment is believed to extend beyond the contrast‐enhancing tumor boundaries [18, 20].

CVR can be measured by applying a vasoactive challenge (vasodilatory or vasoconstrictive), to induce the vascular response, in combination with an imaging technique sensitive to cerebral blood flow (CBF) changes. As vasoactive stimuli, vasodilatory agents are preferred, which include acetazolamide [21], or mixed gas administration with carbon dioxide [17, 18, 20]. Voluntary breath‐holding challenges with apneas have the advantage of being non‐invasive and easy to implement [12, 22], and have demonstrated a measurable increase in CBF [10, 23]. The most used MRI technique for this purpose is Blood Oxygen Level Dependent (BOLD) imaging, with a T2 or T2*‐weighted sequence [11, 20, 22]. However, the BOLD signal does not provide a direct measurement of CBF, instead the BOLD response to a vasoactive stimulus such as breath‐hold is due to a combination of CBF and cerebral blood volume changes. Recent studies have tested the potential of Arterial Spin Labeling (ASL) MRI to measure CVR, with promising results [23, 24, 25]. ASL is a non‐invasive technique that quantifies CBF without the use of contrast media [26]. In ASL, the inflowing blood water is magnetically inverted at the level of the cervical arteries, and a post‐labeling delay is set to allow the labeled water proton spins to reach the brain. A perfusion‐weighted image is obtained after subtraction of the labeled images from a control condition, achieving absolute CBF measurements by using a mathematical quantification model [27].

Thus, the aims of this study were: to evaluate CBF and CVR using ASL with a breath‐holding paradigm in a cohort of patients with brain tumor, which to the best of our knowledge had not been attempted before with this technique; to determine differences in CBF and CVR values in distinct tumor areas, including the tumor itself and adjacent brain tissue; and to study the potential use of CBF and CVR to differentiate among metastases, grade 3 and grade 4 gliomas.

2. Materials and Methods

The research protocol was approved by the Ethics Research Committee of our institution. All patients signed a written informed consent prior to participating in the study. Thirty patients referred for presurgical evaluation of brain tumors, primary or metastatic, were prospectively enrolled, from May 2021 to May 2023. Inclusion criteria for the current study were presumed high‐grade gliomas (grades 3 and 4) and solitary metastatic brain tumor cases where a differential imaging diagnosis from primary tumors (without considering clinical criteria) was challenging. Exclusion criteria were standard MRI contraindications (claustrophobia, non‐MRI compatible devices), age lower than 18 years and the inability to complete the breath‐hold paradigm during the MRI session.

2.1. Histopathologic Analysis

Surgically resected specimens were analyzed by a neuropathologist (JIE, 19 years of experience). Three groups were defined on the basis of histopathology: metastasis, grade 3 gliomas and grade 4 gliomas, taking into account data on isocitrate dehydrogenase (IDH) mutation status and other molecular parameters, according to the updated 2021 WHO classification [3]. This classification was used as the ground truth for posterior analyses.

For clarity, in the present study glioma grades have been denoted using Arabic numerals following the updated 2021 WHO classification nomenclature [3], while Roman numerals have been used when referring to studies in the literature using previous WHO classifications [28, 29].

2.2. MRI Protocol

MRI scanning was performed before surgery. Images were acquired on a 3 T MRI scanner (MAGNETOM Skyra, Siemens Healthineers AG, Forchheim, Germany) with a 32‐channel head coil. For this study, a pseudo‐continuous ASL (PCASL) sequence with a breath‐holding paradigm was added to the conventional preoperative protocol used for surgery planning. The imaging sequences used in this work were: T1‐weighted anatomical 3D Magnetization Prepared Rapid Gradient Echo Imaging (MPRAGE) sequence, before and after gadolinium contrast administration; T2‐FLAIR weighted sequence; time‐of‐flight (TOF) angiography sequence, to position the PCASL labeling plane at the carotid and vertebral arteries, using as reference the sagittal and coronal angiograms; and the PCASL sequence (labeling time = 1.2 s, post‐labeling delay = 1.4 s). The PCASL labeling plane was positioned above the carotid bifurcation, within the vertebral arteries V2 segment, and as perpendicular as possible to the carotid and vertebral arteries. The PCASL sequence had background suppression and single‐shot 3D Rapid Imaging with Refocused Echoes (RARE) stack‐of‐spirals readout with through plane acceleration (1D‐GRAPPA) and imaging parameters: excitation flip angle (FA) = 90, echo‐time (TE) = 10.33 ms, repetition‐time (TR) = 3 s, isotropic resolution = 3.75 mm, in‐plane field‐of‐view (FOV) = 240 mm × 240 mm, and in‐plane reconstructed matrix = 64 × 64, as previously described [30]. An M0 baseline image was acquired with the same parameters. Total PCASL acquisition time was 12 min and 30 s, during which 125 label‐control pairs were acquired. Detailed parameter specifications for these sequences can be found in Supplementary Material (Supplementary Tables).

2.3. Breath‐Hold Task

During ASL acquisition, patients were asked to complete a breath‐hold task following pre‐recorded audio cues triggered by the start of the MRI sequence. The audio file was played using PsychoPy2 (University of Nottingham, www.psychopy.org) over the MRI sound system. The breath‐hold task started with a period of 120 s of free‐breathing followed by 10 breath‐holding cycles at end‐expiration of 21 s interleaved with normal breathing periods of 42 s. The breath‐hold task implementation was adapted from a previous work [23]. The subject performance was monitored by means of respiratory bellows, using the MRI Physiological Measurement Unit, and data were saved for later analysis. A graphic representation of the breath‐hold task is depicted in Supplementary Figure S1.

2.4. Anatomical Imaging Processing and Brain Mask Generation

Data processing was carried out in Matlab R2021b (Mathworks, MA, United States) using Statistical Parametric Mapping (SPM) toolbox (v12, Welcome Trust Center for Neuroimaging, University College London, United Kingdom) with in‐house built scripts. Anatomical images were resliced to an isotropic resolution of 2 mm. T1‐weighted post‐contrast and T2‐FLAIR weighted images were co‐registered to T1‐weighted pre‐contrast image.

Gray matter (GM) and white matter (WM) probability maps were generated from the anatomical pre‐contrast T1‐weighted image using the standard segmentation tool in SPM. First, a brain mask was generated adding the GM and WM probability maps, which was used to limit subsequent analyses to brain tissue, excluding the cerebrospinal fluid (CSF). Then, the GM probability map was binarized with a threshold of 0.90 to obtain a GM mask. The cerebellum and the tumor lesion were manually removed from the GM mask. Finally, left and right hemispheres were manually separated within the brain and GM masks to obtain contralateral and ipsilateral masks relative to the lesion.

2.5. ASL Data Processing for CBF and CVR Estimation

ASL and M0 images were realigned and co‐registered to the anatomical T1‐weighted pre‐contrast image for motion correction. The extent of head motion during the breath‐holding task was evaluated for each subject (see Section 3 of Supplementary material). Registration was checked and manually corrected if necessary. ASL images were sync‐interpolated to double the time resolution [31]. Perfusion weighted images with a temporal resolution of 3 s were obtained by the subtraction of interpolated label and control images. CBF maps were generated from the perfusion weighted images using the single‐compartment quantification model [27] with a T1 value of arterial blood (1650 ms) [32, 33]. The GM mask was used to compute the CBF time series subsequently employed in the delay analysis. No smoothing was performed in this analysis.

A baseline mean CBF map was calculated by averaging the CBF images obtained during the initial period of normal breathing, according to the respiratory trace. To assess the impact of arterial transit time artifacts in the generated CBF maps, the spatial coefficient of variation of the CBF signal measured in the contralateral GM was evaluated, according to the method proposed by Mutsaerts et al. [34] (see section 4 of Supplementary Material).

An individual breath‐hold regressor was built for each patient adapted to the actual breath‐hold durations based on the recorded respiratory trace as previously described [23]. This regressor was later modelled as a ramp, with an amplitude from 0 to 1 (see Figure 1A,B).

FIGURE 1.

Post‐processing steps. (A) The respiratory signal was used to create an adapted breath‐hold regressor for each patient. (B) This regressor was later transformed into a ramp pattern. (C) Mean GM CBF time series. The green box represents the initial period of normal breathing. (D) Cross‐correlation function obtained by cross‐correlating the mean CBF in GM and the ramp pattern regressor delayed from 0 to 30 s. The arrow indicates the maximum cross‐correlation value, which was used to determine the optimal time delay. (E) Ramp regressors before (discontinuous line) and after (continuous line) applying the delay and (F) delayed ramp regressor and mean gray matter CBF signal. Data are shown for only the first 350 s of the complete task.

The CBF response to breath‐holding was delayed with respect to the recorded respiratory trace. The time delay was calculated for each subject by cross‐correlation between the mean CBF in GM and the generated breath‐hold ramp pattern. To that end, the breath‐hold ramp pattern was delayed from 0 s to 30 s, in intervals of 3 s (the CBF signal time resolution) to generate multiple delayed regressors. The delay that yielded the highest correlation was used to adjust the breath‐hold regressor during the subsequent analysis.

CVR was determined from the CBF images acquired during the breath‐hold task using a Generalized Linear Model (GLM) containing the breath‐hold delayed ramp regressor, the motion parameters (Mot, obtained during the realignment of ASL images) and a constant offset (β₀), thus fitting the voxel‐wise signal to the Equation 1:

$$y\left(t\right)={\beta }_0+{\beta }_{1}x\left(t\right)+{\beta }_m\mathit{Mot}\left(t\right)$$ (1)

Where y(t) represents the voxel‐wise CBF signal at time t, $x\left(t\right)$ represents the breath‐hold delayed ramp regressor generated for each subject, ${\beta }_m$ are the β parameters scaling the motion regressors, and ${\beta }_1$ returned the voxel CVR value. CVR was expressed as percentage of signal change (PSC), determined as the obtained CVR (${\beta }_1$) divided by the constant offset (${\beta }_0$) times one hundred [35].

A graphical representation of the post‐processing steps is depicted in Figure 1A–F.

2.6. Tumor Masks, Edema Mask and Peritumoral Ring Volumes‐of‐Interest Determination

Tumor and edema volumes‐of‐interest (VOIs) were manually drawn by a neuroradiologist (MC‐I, 4 years of experience in neuroimaging) based on pre‐ and post‐contrast T1‐weighted, and T2‐FLAIR weighted images. Three different masks were delineated: contrast enhanced (ce) tumor depicting only contrast‐enhancing lesion; non‐contrast enhanced (nce) tumor depicting areas of solid tumor without contrast‐enhancement; and edema depicting areas of possible perilesional edema. The CE and nCE tumor masks were combined to generate a whole tumor mask. Necrotic components were not included.

Three‐dimensional concentric VOIs of 6 mm (3 pixels) thickness (i.e., ring VOIs) were created using an in‐house script in Matlab for peritumoral tissue evaluation. Four individualized ring VOIs of 0–6 mm, 6–12 mm, 12–18 mm, and 18–24 mm were generated from the whole tumor mask border outwards, as shown in Figure 2 and Figure 3. The tumor VOI was also mirrored on the contralateral hemisphere to the lesion, and identical contralateral ring VOIs were created for comparison. These ring VOIs were restricted to the ipsilateral or contralateral hemisphere. Voxels outside the detected brain mask were removed from the generated ring VOIs, to avoid measurements in the cerebrospinal fluid.

FIGURE 2.

Anatomical brain MRI including T1 weighted pre and post contrast, and T2‐FLAIR, and computed CBF and CVR maps for a representative patient with a grade 3 oligodendroglioma. The drawn VOIs are represented in the anatomical brain MRI. Abbreviations: BH, breath‐hold; CBF, cerebral blood flow; CVR, cerebrovascular reactivity; VOI: volume‐of‐interest. Images are presented in radiological orientation.

FIGURE 3.

Anatomical brain MRI including T1 weighted pre and post contrast, and T2‐FLAIR, and computed CBF and CVR maps for two representative patients with grade 4 glioblastoma. The drawn VOIs are represented in the anatomical brain MRI. Abbreviations: BH, breath‐hold; CBF, cerebral blood flow; CVR, cerebrovascular reactivity; VOI: volume‐of‐interest. Images are presented in radiological orientation.

2.7. CBF and CVR Measurements

Ipsilateral and contralateral GM masks, whole tumor and edema VOIs, as well as ipsilateral and contralateral ring VOIs, were overlaid on the CBF and CVR maps to obtain mean values in those regions.

2.8. Statistical Analysis

The statistical analysis was performed in RStudio 2022.02.3 (Boston, MA URL http://www.rstudio.com/). Due to the low number of patients in each group, all analyses were conducted using non‐parametric tests, except for the factorial Analysis of Variance (ANOVA) that were employed after transforming the data using the Aligned Rank Transform (ART) [36].

Baseline differences in CBF and CVR between contralateral and ipsilateral GM were evaluated using Wilcoxon signed‐rank tests. Differences in baseline CBF and CVR among the different VOI‐regions (tumor, edema, and contralateral GM) and tumor types (metastasis, grade 3 and grade 4) were assessed with a two‐factor ANOVA for repeated measurements that included the factors VOI‐region and tumor type and their interaction.

For CVR data, the interaction effect was not significant (see Results), so pairwise post‐hoc comparisons across VOIs and across tumor types were conducted with Wilcoxon rank‐sum tests (for tumor types, independent measures) and Wilcoxon signed‐rank test (for VOIs, repeated measures), with Bonferroni correction. For CBF data, the interaction term was significant, thus comparisons across VOIs within each tumor type were conducted using Friedman test for repeated measures, followed by post‐hoc tests using Wilcoxon signed‐rank test with Bonferroni correction. In addition, comparisons across tumor types within each VOI were conducted using Kruskal–Wallis test for independent measures, followed by post‐hoc tests using Wilcoxon rank‐sum test with Bonferroni correction.

For the peritumoral ring VOI analyses, multiple Wilcoxon signed‐rank tests were conducted to compare ipsilateral and contralateral values (one for each ring VOI). No correction method was applied to the p‐values due to the exploratory nature of the analysis.

Spearman's rank correlation coefficient was computed to assess the relationship between baseline CBF and CVR, for the tumor VOI.

Receiver operating characteristics (ROC) curves were generated to evaluate the capacity of CBF and CVR values to discriminate between grade 3 and grade 4 gliomas and the optimal quantitative threshold for classification was determined using the Youden Index. The area under the curve (AUC), sensitivity and specificity at the optimal threshold were computed. A DeLong's test was performed to assess statistically significant differences between the correlated ROC curves.

For all the analyses, a significance level of p < 0.05 was used.

3. Results

3.1. Patient Characteristics

Thirty patients met the inclusion criteria. Three patients (10%) were not able to complete the breath‐holding task and were excluded from the analysis. The breath‐holding task was well tolerated by the remaining subjects, and no adverse effects were recorded. Finally, a total of 27 patient datasets were analyzed (mean age 56 ± 11 years, 17 males). A complete characterization of the patients is presented in Table 1. Among them, 23 patients presented diffuse gliomas (grade 4, n = 17, and grade 3, n = 6), and four had solitary metastasis (three of pulmonary origin and one from ovarian cancer), confirmed by histopathological analysis [3]. Patients with glioma were studied before any treatment. Within the metastasis group, two individuals were treatment‐naïve and presented the cerebral lesion in the staging brain MRI, having exclusively received corticosteroids. The remaining two patients, one with ovarian cancer and the other with lung adenocarcinoma, were diagnosed several years earlier during their oncological condition. The patient with lung adenocarcinoma had previously undergone systemic treatment with Cisplatin + Vinorelbine, followed by Darvalumab. The primary treatment approach for the brain lesion involved radiosurgery (stereotactic fractionated radiotherapy). The patient with ovarian cancer had undergone a treatment regimen consisting of Paclitaxel + Carboplatin (6 cycles) + Bevacizumab.

TABLE 1.

Demographic information, tumor characterization, and histopathological diagnosis (2021 WHO classification [3], including the most important molecular information available) of patients included in the analysis.

Patient number	Age (years)	Gender	Pathology	Molecular information	Tumor location	Tumor volume (mm3)
Metastasis (n = 4)
1	59	F	Primary ovarian	Ovarian serous carcinoma	R parietal	1456
2	61	F	Primary lung	Lung adenocarcinoma	L parietal	3544
3	48	F	Primary lung	Lung adenocarcinoma	L parietal‐occipital	14,960
4	57	M	Primary lung	Lung adenocarcinoma	L parietal	1056
Grade 3 (n = 6)
5	47	M	Astrocytoma	IDH1 mutation	L frontal	23,944
6	54	M	Astrocytoma a	IDH1‐wildtype, no EGFR amplification, no TERT mutation	L frontal	2624
7	81	F	Oligodendroglioma	IDH1 mutation/1p19q codeletion	R temporal	20,080
8	58	F	Oligodendroglioma	IDH1 mutation/1p19q codeletion	R frontal	8816
9	31	F	Astrocytoma	IDH1 mutation	L frontal	27,088
10	64	M	Astrocytoma	IDH1 mutation	R occipital	17,176
Grade 4 (n = 17)
11	48	M	Astrocytoma a	IDH1‐wildtype, EGFR amplification	L frontal	8416
12	68	F	Glioblastoma	IDH1‐wildtype	R parietal	27,792
13	62	M	Glioblastoma	IDH1‐wildtype	R frontal	15,336
14	52	F	Astrocytoma	IDH1‐wildtype	R occipital	8776
15	51	M	Glioblastoma	IDH1‐wildtype	L thalamic‐temporal	22,360
16	50	F	Glioblastoma	IDH1‐wildtype	R temporal–parietal	19,712
17	70	M	Glioblastoma	IDH1‐wildtype	R temporal‐occipital	19,536
18	48	M	Glioblastoma a	IDH1‐wildtype, EGFR amplification	R‐L parietal	36,848
19	56	M	Glioblastoma	IDH1‐wildtype	R frontal	10,480
20	46	M	Gliosarcoma	IDH1‐wildtype	R parietal	6888
21	48	M	Glioblastoma	IDH1‐wildtype	R parietal‐occipital	6728
22	77	F	Glioblastoma	IDH1‐wildtype	L temporal–parietal	30,672
23	50	M	Glioblastoma a	IDH1‐wildtype, EGFR amplification	R temporal	18,232
24	52	M	Glioblastoma	IDH1‐wildtype	R frontal	15,944
25	41	M	Glioblastoma	IDH1‐wildtype	L basal ganglia	9192
26	74	M	Glioblastoma	IDH1‐wildtype	R temporal	21,568
27	53	M	Glioblastoma	IDH1‐wildtype	R frontal‐temporal	42,784

Note: The volume was calculated from the tumor volume‐of‐interest (VOI).

Abbreviations: EGFR, epidermal growth factor receptor; F, female; IDH1, isoform 1 of the isocitrate dehydrogenase enzyme; L, left; M, male; R, right; TERT, telomerase reverse transcriptase promoter; WHO, World Health Organization.

^a Multi‐lesion or butterfly configuration.

Three of the patients with glioma had more than one lesion within the same hemisphere, so only the largest lesion was analyzed. Two patients were not included in the ring VOI assessment due to bilateral tumor lesions which prevented the comparison with the contralateral side: one patient had a tumor crossing the midline with a butterfly configuration, and the other patient had a multicentric tumor with two lesions one in each hemisphere.

Obtained anatomical images, graphical representation of used VOIs, and reconstructed CBF and CVR maps from a representative subject with grade 3 glioma are included in Figure 2 and two representative subjects with grade 4 gliomas are presented in Figure 3.

3.2. Baseline Cerebral Blood Flow

Baseline CBF was obtained averaging a mean of 38 images (± 6 images) acquired during normal breathing. Baseline CBF values are reported in Table 2 and estimated median differences between tumor types and VOI‐regions are included in Supplementary Material (Supplementary Tables).

TABLE 2.

Mean and median baseline CBF and CVR values.

	All (n = 27)	Metastasis (n = 4)	Grade 3 (n = 6)	Grade 4 (n = 17)
Baseline CBF (mL/100 g/min)
Tumor	49.3 ± 35.0 36.5 (22.9, 62.9)	20.8 ± 6.1 20.3 (18.1, 23.1)	40.4 ± 34.5 21.6 (18.7, 54.4)	59.2 ± 35.6 42.4 (34.4, 65.6)
Edema	27.5 ± 19.1 23.3 (16.4, 29.3)	18.4 ± 4.5 18.1 (14.8, 21.7)	24.2 ± 18.6 19.6 (13.2, 24.3)	30.7 ± 21.1 28.0 (21.5, 30.9)
Contralateral GM	41.4 ± 8.6 41.7 (35.4, 45.6)	43.3 ± 4.7 43.1 (41.6, 44.9)	36.4 ± 8.3 36.1 (30.1, 43.7)	42.7 ± 9.1 41.7 (36.1, 45.8)
Ipsilateral GM	41.3 ± 8.6 40.9 (36.7, 45.5)	44.6 ± 5.4 43.4 (41.0, 47.0)	37.7 ± 8 39.6 (31.1, 43.1)	41.9 ± 9.3 39.8 (35.8, 45.6)
CVR (PSC)
Tumor	29.8 ± 19.3 27.9 (14.6, 43.2)	41.5 ± 8.4 43.2 (38.7, 46.0)	42.9 ± 25.5 42.8 (29.3, 64.3)	22.4 ± 15.4 24.2 (10.6, 29.9)
Edema	37.0 ± 20.9 33.1 (22.1, 49.2)	43.8 ± 7.9 46.7 (41.6, 48.9)	46.5 ± 20 49.2 (29.6, 57.7)	32.0 ± 22.4 29.2 (20.1, 41.6)
Contralateral GM	38.7 ± 14.4 37.0 (25.2, 52.7)	50.0 ± 5.3 49.2 (47.0, 52.2)	46.0 ± 17.0 52.7 (31.8, 55.3)	33.5 ± 12.7 29.9 (23.7, 41.6)
Ipsilateral GM	35.9 ± 13.4 37.1 (26.1, 46.2)	44.7 ± 2.3 45.4 (44.1, 46.0)	38.6 ± 11.7 37.6 (29.3, 46.4)	32.8 ± 14.7 27.6 (23.1, 39.5)

Notes: Data are shown as mean ± standard deviation (in gray) and median (quartile 1 and quartile 3) in black.

Abbreviations: CBF, cerebral blood flow; CVR, cerebrovascular reactivity; GM, gray matter; PSC, percentage of signal change.

There was no difference in baseline CBF between the contralateral and ipsilateral GM (p = 0.78) in the whole cohort, as illustrated in Supplementary Figure S2. Baseline CBF values for the different histological tumor types and the different VOI‐regions (tumor, edema, and contralateral GM) are graphically depicted in the boxplots of Figure 4A. The ANOVA results showed a significant interaction between the VOI‐regions and the tumor type (p = 0.01). Subsequent comparisons across VOIs performed within each tumor type revealed that edema had significant lower CBF values than the tumor and the contralateral GM only in the grade 4 glioma cohort (p = 0.012). Comparisons across tumor types performed within each tissue type revealed that the tumor VOI had higher CBF values in grade 4 than in metastases (p = 0.004), but no other significant differences were observed between grade 3 and grade 4 (p = 0.26), or between grade 3 and metastases (p = 1). Also, no significant CBF differences were observed between tumor types for the edema VOI (p = 0.12) or contralateral GM VOI (p = 0.37).

FIGURE 4.

(A) Baseline CBF (mL/100 g/min) values for the three tumor types (metastasis, grade 3 and grade 4) measured in the different VOI‐regions (tumor, edema, and contralateral GM). (B) CVR (as PSC) for the three tumor types measured in the same VOIs. The asterisk (*) denotes significant differences.

3.3. Cerebrovascular Reactivity

During the breath‐hold challenge, the delay between the respiratory trace and the recorded CBF response in the GM was of 4 ± 1 images (12 ± 3 s).

CVR values are reported in Table 2 and graphically depicted in Figure 4B. Estimated median differences between tumor types and VOI‐regions are included in Supplementary Material (Supplementary Tables). There was a statistically significant difference in CVR between the contralateral and the ipsilateral GM (p = 0.03), with lower values in the ipsilateral hemisphere to the tumor location, see Supplementary Figure S2. The ANOVA results showed no significant interaction between VOI‐regions and tumor type (p = 0.85) but yielded significant main effects of tumor type (p = 0.03) and VOI‐regions (p = 0.004). Subsequent comparisons across tumor types showed that grade 4 gliomas had significantly lower CVR values than metastases and grade 3 tumors, while no significant differences were found between metastases and grade 3 (p = 1.00). Across the different VOI‐regions, statistically significant differences were found between the tumor and the contralateral GM (p = 0.016), with tumor presenting lower CVR values than the contralateral GM.

A significant strong negative correlation (spearman rho = −0.71, p < 0.001) was found between baseline CBF and CVR values in the tumor VOI. See Figure 5.

FIGURE 5.

Correlation between baseline CBF in mL/100 g/min and CVR expressed as PSC, measured in the tumor VOI. Spearman's rho is reported with its p‐value.

3.4. Ring VOIs Assessment

Figure 6A–C shows CBF and CVR evaluation results performed in the perilesional ring VOIs to assess perfusion and cerebrovascular response beyond the tumor margins.

FIGURE 6.

Ring VOI assessment. Values in metastasis (A), grade 3 (B) and grade 4 (C) groups of baseline CBF (mL/100 g/min) (left) and CVR (PSC) (right) obtained for each VOI. Red line represents the values from the VOIs at the ipsilateral side of the tumor, and blue line represents the values from the VOIs at the contralateral side of the tumor. Only statistically significant differences are reported with the corresponding p values.

In the metastasis group, CBF values in the tumor VOI were lower than in the contralateral‐mirrored VOI. Conversely, grade 3 and grade 4 gliomas presented higher CBF than their contralateral‐mirrored VOI, although differences were only significant in grade 4 gliomas (p = 0.04). CBF values did not show significant differences between the ipsilateral and contralateral side in the perilesional rings in any group. Nonetheless the following patterns were observed: CBF values in the first 6 mm‐ring in metastases were slightly lower than in the contralateral homologous region, whereas in grade 3 and 4 gliomas the first 6 mm VOI had slightly higher values than the contralateral homologous VOI. For subsequent ring VOIs extending further from the lesion, values were similar in both hemispheres.

Different patterns were observed in the CVR measurements. In the three tumor types, CVR values in the ipsilateral hemisphere were reduced compared to the contralateral side and did not reach the contralateral values (except at 24 mm in the metastasis group). Significant differences were found between ipsilateral and contralateral sides for all the ring VOIs in grade 4 gliomas (Figure 6C).

The visual analysis of CVR graphs for each tumor type revealed a different behavior in the peritumoral regions. In general, CVR values in grade 4 gliomas were lower than those presented for metastases and grade 3. In grade 4 gliomas, CVR gradually increased until 18 mm from the tumor, reaching a plateau. In grade 3, in contrast, CVR values were similar across the ipsilateral ring VOIs. In metastases, a slight difference was observed between tumor and the first 6 mm VOI, with similar values in the rest of the rings.

3.5. ROC Curves

Results from the ROC analysis to differentiate between grade 3 and grade 4 gliomas are presented in Figure 7. The AUC for tumor CBF was 74.5% (CI 95%: 42.2%, 100%), while the AUC for the tumor CVR was 77.5% (CI 95%: 51%, 100%). Even though there were no statistically significant differences between both curves, only the AUC for CVR was significantly different from 0.5. The Youden Index cut‐off value for CVR was 27.9 (specificity: 83.3%, sensitivity: 70.6%), suggesting that CVR could be a better biomarker to differentiate between grade 3 and grade 4 gliomas.

FIGURE 7.

ROC curve for CVR, to differentiate between grade 3 and grade 4 gliomas. The cut‐off value calculated by the Youden Index is shown, as well as corresponding sensitivity and specificity values. Abbreviation: AUC: area under the curve.

4. Discussion

This study demonstrated that PCASL could be used to simultaneously measure CBF and CVR in a cohort of patients with brain tumor, without the necessity for external gas manipulation. Instead, we employed a noninvasive breath‐holding task as hypercapnic stimulus, rendering the methodology easily implementable.

4.1. CBF Measurements

In our cohort, CBF values did not show significant differences between grade 3 and grade 4 gliomas, aligning with previous findings from several small‐scale studies, which have showed limited efficacy of CBF in distinguishing between glioma grades III and IV [37, 38, 39]. Additionally, a comprehensive systematic review by Alsaedi et al. [39] found that while CBF can differentiate low‐grade from high‐grade gliomas, its discriminative ability diminishes when comparing grade II versus grade III, or grade III versus grade IV gliomas.

Tumoral CBF values in metastases were lower than in gliomas, although the differences were only significant between metastases and grade 4 gliomas. Regarding this finding, discordant results were found in the literature. Lin et al. [5] reported no differences in CBF between gliomas and metastases within the enhancing tumor lesion, in contrast to the differences found in the peritumoral edema. Conversely, other studies found higher CBF values in high‐grade gliomas compared to metastases [6, 40]. The variable nature of metastases, depending on their origin, can potentially confound the results: metastases from lung, colon, breast, ovarian and gastric cancer tend to be hypovascular; while those originating from melanoma, neuroendocrine tumors, renal cell, and thyroid carcinoma are typically hypervascular [41, 42]. In our cohort, metastases exhibited lower tumoral CBF values than those on the contralateral‐mirrored side (although the difference was not statistically significant), in agreement with their origin, that corresponded to the hypovascular category. However, CBF values could also be influenced by prior systemic treatments. Notably, one patient with ovarian carcinoma was undergoing antiangiogenic therapy (Bevacizumab) during the presurgical assessment.

Related to the peritumoral evaluation of CBF, we found reduced baseline CBF values in edema compared to the tumor and the contralateral GM, with differences only statistically significant for grade 4 tumors. This is in line with prior work that has found lower CBF values in edema in a cohort of patients with brain metastases [43]. To delve deeper into peritumoral perfusion, we conducted a more detailed evaluation within several VOI rings to examine perfusion beyond the tumor edges. Despite the lack of statistically significant differences, mean CBF values within the first 6 mm ring did not reach the values found in the contralateral ring in any of the groups. In metastases, CBF values were lower than those in the contralateral homologous areas. Conversely, grade 4 and grade 3 gliomas showed slightly higher CBF values within the first 6 mm ring compared to the contralateral homologous regions, while the rest of the rings demonstrated similar values as the contralateral side. Lin et al. [5] introduced the concept of the CBF “gradient” within the peritumoral edema, referring to the disparity in CBF values between the vicinity of the enhancing tumor and the normal‐appearing WM, a pattern consistent with our findings.

4.2. CVR Measurements

A reduction in CVR within the ipsilateral GM to the tumor was found with respect to the contralateral GM, in agreement with previous work [20]. This is likely due to the global impairment in autoregulatory hemodynamic mechanisms, that is known to occur on a widespread scale in diffuse gliomas [17], even in areas remotely located from the visible contrast‐enhancing lesion [16].

In line with these findings, the perilesional analysis using expanding ring VOIs originating from the tumor showed decreased CVR values in comparison to the contralateral flipped VOIs within all three tumor types (although differences were significant only in grade 4 gliomas), and they did not reach the contralateral values, except at 24 mm in the metastasis group. Muscas et al. [18] previously examined peritumoral CVR within 3 mm‐VOIs, noting marked CVR impairment within the lesion and up to 21 mm outside the contrast‐enhancing lesion. Similarly, these hemodynamic alterations were most pronounced closest to the contrast‐enhancing tumor lesion. Sebök et al. [20], in a similar analysis of the peritumoral region with 6 mm ring VOIs, observed a trend toward CVR normalization beginning at 18 mm distance from the visible glioma lesion. Both studies highlighted impaired CVR beyond the standard MRI‐defined tumor border, suggesting active tumor infiltration in the peritumoral tissue, in agreement with our findings.

The tumor VOI consistently exhibited significantly lower CVR values in comparison to the contralateral GM for all tumor types, also in agreement with previous investigations [17, 18, 19, 20]. Notably, these findings were consistent despite the different methods employed, as previous studies used BOLD‐functional MRI acquisition with controlled carbon dioxide‐inhalation stimulus. Furthermore, our study revealed significant differences in CVR across the different tumor types. More specifically, grade 4 gliomas exhibited lower CVR values in comparison to both grade 3 gliomas and metastases. This suggests that CVR could be a potentially useful biomarker for their presurgical differentiation, which was confirmed by the ROC analysis findings that differentiated between grade 3 and grade 4 gliomas.

4.3. CBF and CVR Correlation

Remarkably, we found a significant inverse correlation between the tumor perfusion and CVR metrics in our cohort of high‐grade gliomas and metastases, similar to the results of Pillai et al. [15] in grade 4 tumors. This correlation suggests that the increase in CBF is associated to a marked decline in CVR. This decreased autoregulatory capacity of arterioles to dilate or constrict in response to vasoactive stimuli is likely due to the presence of immature and unorganized neoangiogenic vessels that exhibit hyperperfusion and it is likely to account for NVU in high‐grade tumors. Moreover, the regionally decreased CVR even in the peritumoral region could serve as a biomarker for potential NVU [14, 15].

4.4. Other Considerations

The labeling time and post‐labeling delay used in the PCASL sequence (1.2 s and 1.4 s) were shorter than those recommended by the consensus paper (1.8 s) [26]. Following the consensus recommendation would have required a minimum TR of 4 s, meaning that only one label‐control pair could be acquired every 8 s. Reducing the labeling time and post‐labeling delay allowed running the PCASL sequence with a TR of 3 s. Combined with the interpolation step, this approach enabled the perfusion signal to be sampled seven times within the 21 s breath‐hold duration. By reducing the labeling time, an SNR penalty was expected, however previous work in healthy subjects showed that the perfusion SNR reduction was less than 15% [23]. Reducing the pulse labeling delay on the other hand, was also done to increase SNR, although it compromised CBF quantification for those voxels where arterial transit time (ATT) was longer than the PLD, but 1.4 s was expected to be long enough for the labeled spins to reach most gray matter voxels, as in healthy gray matter ATT can vary between 500 and 1500 ms, while in tumors ATT is expected to be shorter [44, 45, 46]. In case of studies in elderly patients (> 70 years old), the use of a longer PLD should be considered, as in this population, the short PLD could have a larger effect on the CBF and CVR measurements [47].

Tumor and edema CBF and CVR values were not normalized to the contralateral GM, as previous studies have shown whole brain CVR impairment in patients with brain tumor compared to healthy controls, thus the contralateral hemisphere cannot be considered healthy.

The IDH molecular status in gliomas, widely used after the WHO 2016 classification [29], was not used to categorize patients because (i) the IDH mutation versus IDH‐wildtype categories would closely matched to tumor grades 3 and 4, respectively (as presented in Table 1); and (ii) the final grading category includes other molecular markers known to confer a worst prognosis after the WHO 2021 classification [3] such as CDKN2A deletion or epidermal growth factor receptor (EGFR) amplification.

4.5. Limitations

The results of this study are limited by the unbalanced and relatively small sample size, as a result of which some of the statistical tests could be underpowered. Future multi‐center studies with larger cohorts of patients are needed to validate these results.

End‐tidal CO₂ measures were not available to evaluate the CO₂ increase induced by breath‐holding.

Metastatic brain tumors were included in this study, in spite of the small group size, as a preliminary exploration of vascular reactivity within this tumor group, although it must be taken into account that previous treatments could confound the evaluation of CBF, and eventually, the brain's response in terms of vasodilation or constriction.

5. Conclusion

We have successfully demonstrated that PCASL can be used for CVR mapping with a breath‐hold task in patients with tumor. CBF and CVR can provide complementary information that could be potentially useful for tumor grading and for assessing the peritumoral extent in gliomas. Specifically, CVR impairment was observed to a greater extent within the tumor but was also present in the adjacent tissue, with different patterns according to tumor types. Grade 4 gliomas exhibited greater CVR impairment in comparison to both grade 3 gliomas and metastases, which could be potentially useful for their presurgical differentiation.

Conflicts of Interest

M.V. works for Siemens Healthineers, Madrid, Spain. All other authors declare that they have no conflicts of interest.

Supporting information

Figure S1. Breath‐hold task. (A) Ideal breathing pattern expected from patients according to the instructions provided in the audio file. (B) Actual respiratory signal in a representative subject. Abbreviations: BH: breath‐hold, NB: normal breathing.

Figure S2. Boxplots of baseline CBF (A) and CVR (B) in Contralateral and Ipsilateral GM. Significant differences in CVR were found (p = 0.03).

Table S1. Imaging parameters.

Table S2. CBF and CVR differences: (A) Between tumor types within each VOI‐region; (B) Between VOI regions within each tumor type.

Table S3. Motion parameters obtained from the ASL images after the realignment step.

Figure S3. Coefficient of variation (CoV) of the baseline CBF map measured in the contralateral gray matter expressed in percentage. (A) Histogram of the computed CoV for all patients, (B) Boxplots of the CoV categorized per tumor type.

Figure S4. Baseline CBF maps from two representative patients, one with low and one with high coefficient of variation (COV). On the right, the segmentation of the contralateral gray matter is represented overlaid in red.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.