Patients were scanned on a 3 T Skyra VE11 (Siemens, Erlangen, Germany) scanner with a 32-channel head coil. An axial 2D T2*-weighted gradient echoplanar sequence (50 slices with interleaved acquisition) planned on the ACPC line plus 20° anticlockwise on a sagittal image was used to acquire the BOLD data, with voxel size 2.5 × 2.5 × 2.5 mm³, repetition time (TR)/TE 1800/30 m, flip angle 80°, bandwidth 2168 Hz/Px, field of view (FOV) 220 × 220 mm². During the echo-planar imaging (EPI) sequence acquisition, 160 volumes were acquired corresponding to a duration of circa 5 minutes. Acquisition parameters of T1-CE, FLAIR, T2 as per standard brain tumor protocol have been described in previously published article (Stumpo et al., 2021). For a subgroup of patients, clinical-diagnostic work-up required execution of a gadolinium (Dotarem) enhanced DSC-MRI GRE-EPI, FOV 100 × 100 mm², resolution 128 × 128, TR/TE 2040/36, FA 90°.

A custom-built computer controlled gas blender (RepirAct™ Gen 4, Thornhill Medical, Toronto, Canada) was used to precisely control partial pressure of end-tidal O₂ (P_ETO₂) and CO₂ (P_ETCO₂) with the prospective gas targeting algorithm based on (Slessarev et al., 2007). BOLD signal changes were induced by a rapid double hypoxic stimulus as previously proposed by and Sayin et al. (Poublanc et al., 2021; Sayin et al., 2022; Sayin et al., 2023a) The programmed P_ETO₂ stimulus pattern was 4-min and 20 s long and consisted of a 60 s baseline P_ETO₂ of 95 mmHg (normoxia), a step decrease in P_ETO₂ to 40 mmHg (hypoxia) for 60 s, a return to normoxia for 20–40 s, a second step decrease in P_ETO₂ to 40 mmHg for 60 s, followed by a return to normoxia for 40-60 s (see also Figure 1).

The acquired transient hypoxia induced-dOHb images were first slice-time corrected and volume re-registered for motion correction within the time-series. Next the BOLD images were realigned to their respective axial anatomical T1-weighted images using Analysis of Functional Neuroimaging (AFNI) software Version 22.0.04‘Hadrian’. (National Institutes of Health, Bethesda, Maryland) (Cox, 1996; Cox and Hyde, 1997). Using Statistical Parameter Mapping (SPM 12, Wellcome Trust Centre for Neuroimaging, Institute of Neurology, University College London; http://www.fil.ion.ucl.ac.uk/spm), automated segmentation of the T1-weighted image yielded grey and white matter, cerebrospinal fluid, skull and skin probability maps. T2, FLAIR, and T1-CE volumes were co-registered and resliced to the high-resolution T1 image acquired in the same session as the BOLD.

For quality control, we calculated the goodness of fit (Rsquared) of the percentage BOLD signal change and the P_ETO₂ time series was regressed using a linear least square fitting to the BOLD time series on a voxel-per-voxel basis.

Perfusion measures using transient hypoxia induced-dOHb were calculated with a conventional analysis using an arterial input function (AIF) chosen over the middle cerebral artery and a deconvolution-based model as previously described by (Poublanc et al., 2021). First, an AIF was chosen over a voxel in the middle cerebral artery. The signal was smoothed voxel-wise using an adaptive mean filtering of width 7 mm. The tracer kinetics relationship was applied in the signal domain and rCBV and MTT metrics were determined using a least square fitting procedure, with MTT bound between 1 and 8 s. The model uses a mono-exponential residue function. Using the central volume theorem, rCBF was then calculated as rCBF = rCBV/MTT (Østergaard, 2005) and scaled by 30 arbitrary units (a.u.).

Region of interest masks (contrast enhancement, necrosis and edema) were automatically segmented using Oncohabitats Software (Juan-Albarracín et al., 2018) for glioblastomas and manually segmented for the other tumor types. Automatically segmented tumor region-of-interest were visually inspected for accuracy and required no subsequent manual correction.

MRI volumes were analyzed using MATLAB 2019 (The MathWorks, Inc., Natick, United States). rCBV, rCBF and MTT values were calculated in the grey and white matter after masking out the whole tumor (including edema) and for each of the aforementioned tumor ROIs. Two additional ROI were defined: tumor, i.e., contrast-enhancement + necrosis and whole lesion, including surrounding edema; and the corresponding volumes in the contralateral hemisphere identified using proprietary Matlab programme. Perfusion metrics were overlayed onto their respective anatomical images using SPM software and qualitatively compared to the co-registered T1CE and FLAIR to identify perfusion patterns in tumor ROIs (contrast enhancement, edema, necrosis).

In a subgroup of patients, a DSC-MRI was acquired during clinical workup for diagnostic purposes. The acquired sequences were analyzed (including leakage correction) using Olea Software version 3.0 to calculate rCBV, rCBF, MTT maps. The perfusion maps were exported, co-registered and resliced to the high-resolution T1 anatomical images and qualitatively compared with the resting perfusion measures obtained with transient hypoxia induced-dOHb. Average rCBV, rCBF and MTT were calculated in the same above-mentioned ROI and contralateral flipped masks and compared with transient hypoxia induced-dOHb estimated values.

Eleven patients provided informed consent for participation in the study. Of these, nine patients completed the imaging protocol and were included in the final study population, reasons of dropout of the two excluded patients being in one patient interruption of study due to dyspnea during the respiratory protocol, in the other discomfort and claustrophobia during MRI acquisition. One patient underwent the imaging protocol twice, the second being after tumor resection. Mean age was 56 (SD 13.8; 3 female subjects). The baseline characteristics are provided in Table 1.

The study population included patients with different types of brain tumors: 1 metastasis, 1 meningioma, 1 astrocytoma WHO grade 2, 1 pleomorphic xanthoastrocytoma WHO grade 3, 1 oligodendroglioma WHO grade 3, 4 glioblastomas (1 patient was also scanned post-operatively). The histopathological diagnosis was based on the 2021 WHO classification (Louis et al., 2021). Patient 3 was excluded from the following analysis due to excessively noisy BOLD signal, caused by suboptimal stimulus during the examination. Four of the included patients received during diagnostic work-up a standard gadolinium perfusion MRI, with one of them receiving it at another center before referral.

An illustrative BOLD signal change and linear fit plot is shown in Figure 1 and is reported for each patient in Supplementary Figure S1.

Transient hypoxia induced-dOHb resting perfusion measures in the abovementioned regions-of-interest were calculated for each patient as previously reported (Poublanc et al., 2021). The relative measurements are reported in Table 2.

Figure 2 shows an illustrative case, i.e., patient 2, 58 y. o. male with a left frontobasal glioblastoma. The perfusion maps show high CBV, CBF in the contrast-enhanced lesion regions, while these metrics are very low in the central necrotic area with high agreement with the morphological high-resolution sequences. MTT is mostly increased in peritumoral edematous tissue. Figure 3 displays perfusion patterns in different tumor types. Qualitative image analysis shows perfusion to be higher in grey matter than in white matter and increased CBV and CBF were observed in contrast-enhancing lesion of metastases (Figure 3A), oligodendroglioma WHO grade 3 (Figure 3B), glioblastoma (Figure 3C) and meningioma (Figure 3E) with respect to healthy tissue. On the contrary, lower CBV and CBF were observed in the astrocytoma WHO grade 2 (Figure 3D). MTT was shown to be higher in areas of edema identified by T2 FLAIR (Figures 3A, B) as well as in the astrocytoma WHO grade 2 dense cellular tissue (Figure 3D).

In the three patients who underwent a pre-operative DSC-MRI in our institution, this was compared with the transient hypoxia induced-dOHb perfusion maps for visual qualitative examination and rCBV, rCBF and MTT were calculated in each of the two techniques. The transient hypoxia induced-dOHb BOLD perfusion and corresponding gadolinium DSC-MRI are shown for three patients (2 glioblastomas and 1 astrocytoma WHO grade 2) in Figure 4. The values of rCBV, rCBF, MTT obtained by gadolinium-based DSC-MRI and analyzed with Olea-Software were compared to the transient hypoxia induced-dOHb estimated values in grey matter, white matter, identified tumor ROIs and flipped masks (Figure 5). In particular, the plots show congruent measurements in the different ROIs and contralateral flipped masks between the two techniques with similar trends.

Based on the availability of 2 transient hypoxia induced-dOHb perfusion scans for patient 4, a multifocal glioblastoma with lesions in left cuneus and precuneus, follow-up imaging from pre-to post-operative is shown and compared with gadolinium DSC (Figure 6). Transient hypoxia induced-dOHb BOLD perfusion maps display good spatial agreement with gadolinium-based DSC perfusion also in the longitudinal follow-up of a glioblastoma patient.

Our preliminary study shows good feasibility of transient controlled hemoglobin desaturation by means of a standardized hypoxic stimulus to obtain resting perfusion parameters in a population of patients with brain tumor. At the single patient level, we found the calculated transient hypoxia induced-dOHb perfusion maps to be in good agreement with intrinsic tumor characteristics that would be expected in a GBCA perfusion, i.e., increased CBV and CBF in highly perfused tumors such as brain metastasis, glioblastomas, oligodendroglioma and meningioma compared to healthy tissue; decreased CBV, CBF in astrocytoma grade 2 as well as increased MTT in edematous or dense cellular tissue. For the subgroup of patients for whom a DSC-MRI was available, the relative ROI-level patterns estimated using transient hypoxia induced-dOHb well reflect the measurement obtained in standard gadolinium perfusion. Moreover, the longitudinal assessment in one patient shows good qualitative agreement with the relative DSC-MRI. The investigated technique, which exploits the BOLD-related drop in MRI signal induced by a bolus of paramagnetic deoxyhemoglobin as a possible alternative for gadolinium contrast, has been recently introduced in seminal publications by (Vu et al., 2021; Poublanc et al., 2021; Sayin et al., 2022) in healthy subjects as well as in patients with steno-occlusive cerebrovascular disease (Sayin et al., 2023a).

Perfusion assessment is clinically relevant in brain tumor diagnosis and follow-up, as increased perfusion correlates with tumor’s aggressiveness and can be used in the differential diagnosis of cerebral lesions as well as to distinguish post-treatment changes from tumor recurrence/progression (Guida et al., 2022; Stumpo et al., 2022). However, in a proportion of brain tumors the blood brain barrier is either disrupted or dysfunctional (e.g., glioblastomas, metastasis) (Arvanitis et al., 2020; Guan et al., 2021), making contrast leakage in the extravascular extracellular space a significant confounder to reliable and reproducible measurements (Shiroishi et al., 2015; Leu et al., 2017; Boxerman et al., 2020). For this reason, GBCA perfusion techniques require appropriate correction and rely most commonly on a loading dose of contrast agent administered prior to the acquisition to minimize T1 changes during first pass, and/or on mathematic correction algorithms (Leu et al., 2017; Boxerman et al., 2020).

Previous literature has focused on BOLD-contrast for functional MRI and task-based pre-surgical mapping (as neural activation induces changes in regional blood flow) (Glover, 2011; Zacà et al., 2014), or resting-state fMRI for functional architecture inference (Lee et al., 2013; Stoecklein et al., 2020), as well as characterization of cerebrovascular reactivity during vasodilatory stimulus (Fierstra et al., 2013; Fierstra et al., 2016; Fierstra et al., 2018; Fisher et al., 2018; Muscas et al., 2020; Sebök et al., 2020; Sebök et al., 2021). Only recently has the possibility to induce a reproducible and standardized transient bolus of deoxygenated blood (while controlling for competing PCO₂ effects) - in combination with the high temporal resolution of EPI sequence - become focus of attention in the setting of perfusion quantification (Poublanc et al., 2021; Vu et al., 2021; Sayin et al., 2022; Sebök et al., 2023a). BOLD imaging contrast is well suited for this technique as it derives from the different physical properties of hemoglobin in its saturated and desaturated state which result in a diamagnetic versus a paramagnetic signal response, respectively (Buxton, 2013; Gauthier and Fan, 2019). Perfusion patterns observed in our study are in agreement to what is reported in the literature for different tumor histologies (Tamrazi et al., 2016; Guida et al., 2022; Stumpo et al., 2022) and display, even if with some spatial differences, remarkably good qualitative agreement with DSC-MRI. To which extent the spatial as well as quantitative differences observed are determined by intrinsic tumor characteristics exhibiting differential response to a contrast medium versus an endogenous deoxyhemoglobin bolus remains to be further investigated in future studies. Nevertheless, development and validation of the transient hypoxia induced-dOHb technique for perfusion measurement in patients with brain tumors would present several benefits with respect to gadolinium contrast. We expect that endogenous deoxyhemoglobin, as opposed to GBCA, remains completely intravascular even in cases of BBB leakage. As such it could be even more sensitive to certain vascular properties of some tumors with respect to gadolinium, despite caution need to be taken due to the described magnetic distortion generated by deoxyhemoglobin also in the extravascular space. (Buxton, 2013). The potential advantages of this technique include avoidance of an exogenous contrast agent with its connected drawbacks, i.e., potential of allergic reactions (Gracia Bara et al., 2022), known accumulation of gadolinium in the brain (Gulani et al., 2017), difficult handling in nephropathic patients due to concerns of nephrotoxicity (Weinreb et al., 2021) and a higher repeatability of the scan during treatment follow-up. These benefits are also shared from arterial spin labeling (ASL) MRI, which in recent years has been extensively investigated in the setting of cerebral perfusion assessment and also in patients with brain tumor. Despite promising results, this technique lagged somewhat behind in the clinical implementation. (Cebeci et al., 2014; Alsop et al., 2015; Haller et al., 2016; Alsaedi et al., 2018).

On the other hand, possible drawbacks include susceptibility artifacts that limit the study of tumors with hemorrhagic component (e.g., melanoma metastasis), reduced contrast to noise ratio and higher predisposition to movement artifacts. Optimization of pre-processing steps (smoothening, etc.) should be also pursued to find the optimal trade-off between decreasing noise and maintaining a fair spatial specificity useful in brain tumor assessment. In order to characterize the relative effect of such confounders, as well as differences and similarities between traditional DSC-MRI versus transient hypoxia induced-dOHb, future validation studies should ideally adopt the same acquisition parameters.

Further, blood desaturation for investigation of brain tumor vascularity could exploit analysis technique different than bolus tracking method such as carpet plot analysis (Bhogal et al., 2022) or transfer function analysis (Sayin et al., 2023b), with these having already provided encouraging results in healthy subjects. Moving away from resting perfusion assessment, other approaches may be employed to investigate intrinsic tissue features by dissecting specific BOLD signal changes characteristics during gas control. A preliminary study demonstrated the feasibility of such technique, with a small cohort of glioblastoma patients exhibiting unique tissue response patterns during hypoxic, hyperoxic and hypercapnic stimuli (Stumpo et al., 2021). For this reason, a longitudinal cohort study after optimization of the hypoxic stimulus is currently ongoing. The potential to better characterize tumoral tissue and peritumoral tissue infiltration with such approach relies not only on the observed magnitude of signal change during evoked stimulus, but also the complementary information derived from a refined analysis including other parameters such as goodness-of-fit, contrast-to-noise ratio and temporal lag. In fact, these variables allow further categorization of dynamic functional tissue characteristics that may not be appreciated in static conditions by traditional perfusion analysis. Moreover, advanced data-driven analysis methods such as time-series clustering could also be exploited to identify subgroups of voxels based on stimulus-evoked hemodynamic patterns to correlate with physiological versus pathological vasculature and underlying histological and functional properties of the tissue studied (Baudelet and Gallez, 2003).