Estimation of oxygen extraction fraction based on hemodynamic measurements using DSC-MRI Open Access

Cerebral oxygen extraction fraction (OEF) represents the fraction of blood’s oxygen content that is taken up by brain tissue as blood passes through the microcirculation. The normal range of OEF is 0.35–0.45 with relatively uniform values throughout the brain’s gray and white matter (Fan et al., 2020). OEF increases with age and vascular risk burden (Aanerud et al., 2011; Lin et al., 2023) and is altered in a range of pathophysiological conditions, such as acute ischemic stroke (Fan et al., 2018) and neurodegenerative disorders (Butterfield & Halliwell, 2019; Y. Shi et al., 2016). Hence, quantitative measurement of cerebral OEF is a clinically relevant measure and a valuable tool to understand cerebrovascular alterations in pathological conditions.

Positron emission tomography (PET) imaging with [¹⁵O]-labeled tracers is considered the standard reference method for quantitative OEF measurement (Baron & Jones, 2012). Here, [¹⁵O]O₂ is inhaled either as a bolus or continuously for 10 min. The [¹⁵O]O₂ is delivered to the brain by arterial blood and a fraction of the [¹⁵O]O₂ (OEF) is extracted from the capillary bed into the tissue. The extraction of [¹⁵O]O₂ is assumed to occur instantaneously and is modeled by a single-tissue compartment model (Mintun et al., 1984). Quantification of OEF can be estimated from measures of regional cerebral blood flow (CBF) and cerebral blood volume (CBV) as well as of the cerebral metabolic rate of oxygen (CMRO₂) (Fan et al., 2020). This can be achieved by using an intravenous [¹⁵O]H₂O bolus injection to measure CBF, [¹⁵O]CO₂ inhalation or an intravenous injection of carboxyhemoglobin to measure CBV, and [¹⁵O]O₂ gas inhalation to measure CMRO₂, respectively (Baron & Jones, 2012). It has been shown, however, that CBV can be fixed without introducing significant errors in OEF estimation, thereby removing the need for the [¹⁵O]-CO scan (Lammertsma & Jones, 1983; Sasakawa et al., 2011).

Because of the different tracer delivery methods (continuous or bolus delivery), the different image reconstruction techniques, and the different kinetic model analyses available (Fan et al., 2020), variation of the estimations introduced by methodological choices must be considered along with the physiological variation in OEF, when evaluating quantitative PET OEF measurements.

Availability of the [¹⁵O] isotope is a requirement for measuring OEF using PET. As [¹⁵O] has a half-life of only 2 min, an on-site cyclotron is required to produce [¹⁵O]O₂ and [¹⁵O]H₂O. In addition, PET has an intrinsically low image resolution, is expensive, and exposes the patients to radiation (Baron & Jones, 2012). There is, hence, a need for alternative methods to estimate OEF.

Several alternative methods now exist to estimate OEF without the need for radioactive tracers, many of which exploit the sensitivity of magnetic resonance imaging (MRI) to blood flow and blood oxygen levels. These include quantitative susceptibility mapping (QSM) (Rochefort et al., 2010; Zhang et al., 2014), quantitative blood oxygen level-dependent imaging (qBOLD) (He & Yablonskiy, 2007), and dual calibrated functional MRI (Bulte et al., 2012).

Another approach does not include measurements of oxygen per se but calculates tissue oxygen availability based on local hemodynamics from first principles (Jespersen & Østergaard, 2011) and assumes that this availability is coupled to local metabolic demands by neurovascular coupling mechanisms. The hemodynamic parameters of this model, in turn, can be estimated from dynamic susceptibility contrast (DSC) MRI, commonly used in clinical practice (Mouridsen et al., 2014). This approach is based on a biophysical model of oxygen transport in tissue that uses measurements of capillary mean transit time (MTT) and capillary transit time heterogeneity (CTH), the standard deviation of capillary transit times, to model the fraction of oxygen that can be extracted at a fixed tissue oxygen tension (P_tO₂) (Jespersen & Østergaard, 2011). Previously, OEF would be inferred from local CBF and capillary oxygen permeability times surface area (PS product) by what is referred to as the Crone-Renkin or flow-diffusion equation, which assumes that CTH is negligible (Østergaard, 2020). Capillary flows are highly heterogeneous in the resting brain, however, but homogenize during episodes of increased CBF (Kleinfeld et al., 1998; Stefanovic et al., 2007). To account for this heterogeneity and its impact on the capillary bed’s ‘effective’ oxygen PS product, the biophysical model describes the relation between blood’s transit times and oxygen extraction in individual capillaries. The overall OEF is then estimated by integrating over all transit times, weighted by the distribution of capillary transit times.

Thus, the biophysical model can provide insights into the influence of capillaries and local hemodynamic changes on oxygen availability, which are not captured by other MR-based methods such as QSM and Qbold.

The model of oxygen extraction from capillaries with a certain transit time distribution includes several physiological constants, which are assigned using generally accepted literature values or inferred from transit time characteristics measured in rodents. However, comparisons of human OEF values computed from the biophysical model and DSC MRI with quantitative [¹⁵O] PET OEF measurements are thus far casuistic, as they have not been systematically compared in a larger cohort (Østergaard et al., 2015). In this study, we aimed to calibrate and compare the biophysical OEF model presented by Jespersen and Østergaard (2011) utilizing human [¹⁵O] PET and DSC-MRI data in healthy elderly individuals.

Healthy, elderly, cognitively normal individuals aged 55–75 years were recruited by advertisement. Exclusion criteria included: Mini-Mental State Examination score below 25, any significant systemic or psychiatric disease, any history of stroke or other brain damage, any significant vascular problems such as history of acute myocardial infarction or uncontrolled hypertension, and alcohol or drug abuse. Additionally, participants had to have an estimated glomerular filtration rate (eGFR) >60 to safely undergo DSC-MRI. The study was approved by the Central Denmark Region Ethics Committee, and all subjects gave their informed written consent before enrolment in the study.

MRI was performed with a 3T Skyra scanner (Siemens Healthcare, Erlangen, Germany) using a 32-channel head coil. Each participant had a structural T1, T2FLAIR and a perfusion weighted DSC scan. The T1-weighted image was acquired with an MP2RAGE sequence with 0.9 mm isotropic voxels (TR = 6.5 s, TE = 3.46 s, TI₁ = 0.7 s, TI₂ = 2.8 s, 4° and 6° flip angles, acquisition matrix 288 × 288 × 192). T2-FLAIR was acquired with 0.5 mm × 0.5 mm × 1.0 mm voxels (TR = 5 s, TE = 388 ms, TI = 1.8 s, 120° flip angle, acquisition matrix 512 × 512 × 176). The DSC images were acquired with spin echo EPI and 2.5 mm isotropic voxels (110 volumes, TR = 1.56 s, TE = 60 ms, 90° flip angle, acquisition matrix 86 × 86 × 51 m). Gadolinium (Gadobutrol – Gadovist, Bayer HealthCare Pharmaceuticals, GER) contrast medium was administered in a dosage of 0.2 mmol/kg through the antecubital vein in the subjects’ non-dominant arm, followed by 20 mL of saline; both at an injection rate of 5 mL/s.

The T1-weighted images were first denoised and bias-field corrected and then transformed into MNI space where they were skull-stripped. Then, the images were segmented into gray matter, white matter, and cerebrospinal fluid and subsequently classified into specific structures, including frontal, parietal, temporal, and occipital gray and white matter (Aubert-Broche et al., 2013). The T2 FLAIR images were used to identify white matter hyperintensities by segmenting the image into hyperintense and normal-appearing white matter (NAWM) using a region growing algorithm (Schmidt et al., 2012). The perfusion-weighted DSC images were processed using in-house software (Mouridsen et al., 2014). First, the dynamic images were motion and slice-time corrected. Then, the arrival of contrast agent in the brain was detected and the data range was standardized to 60 s after contrast agent arrival to avoid any influence from inter-individual differences in bolus arrival. Automatic selection of the arterial input function (AIF) was then performed within regions of the anterior and middle cerebral arteries. The 10 best AIF voxels were then selected based on AIF curve characteristics and averaged to produce the final AIF (for details, see AIF Selection section in Supplementary Data and Supplementary Fig. 3). Parametric maps of CBF, CBV, MTT, and CTH were calculated with parametric deconvolution (Mouridsen et al., 2014). An overview of the processing steps for DSC-MRI scans is presented in Supplementary Figure 1.

Calculation of OEF based on the hemodynamic measurements made with DSC-MRI (that is MTT and CTH) can be achieved using the framework presented by Jespersen & Østergaard (2011).

Briefly, a three-compartment model (red blood cells, plasma, tissue) is developed to represent the relationship between the blood transit time, $τ$, and the extraction of oxygen from a single capillary, $Q(τ)$⁠.

The flow of oxygen across the capillary membrane is assumed to be proportional to the concentration gradient between plasma oxygen concentration, $Cp$⁠, and the tissue oxygen concentration, $Ct$⁠. The total oxygen concentration, $C$⁠, as a function of the fractional distance along the vessel, $x∈[0,1]$⁠, can then be calculated using the differential equation:

where $τ$ is the capillary transit time, and k is the rate constant of oxygen transfer. By utilizing the Hill equation, a general equation for the oxygen concentration as a function of the normalized capillary distance, given a transit time, $τ$⁠, can be written as:

where $αH$ is Henry’s constant, $P50$ is the oxygen pressure required for half saturation, B is the maximum amount of oxygen bound to hemoglobin, and h is the Hill coefficient. These constants are assigned generally accepted literature values adopted from the original model (Jespersen & Østergaard, 2011). Given a constant tissue oxygen tension, $PtO2=Ct​/​αH$⁠, the differential equation can be numerically solved to yield the oxygen extraction fraction $Q=1−C(x=1)​/​C(x=0)$ of a single capillary as a function of $kτ$⁠.

Given the oxygen extraction fraction of a single capillary with transit time, $τ$⁠, the overall OEF is calculated by integrating over the transit time distribution $h(τ)$⁠, in the capillary bed.

The probability density function of the capillary transit time distribution, $h(τ)$⁠, is parameterized by a gamma variate function with parameters $α$ and $β$⁠.

The values of MTT and CTH are calculated from $h(τ)$ as $MTT=αβ$ and $CTH=αβ$⁠.

Each participant underwent four 3 min dynamic emission recordings in 3D list mode; two after [¹⁵O]O₂ administration followed by two after [¹⁵O]H₂O administration. All four scans were performed within the same session with the participant resting in a supine position. The scanning was performed on a high-resolution research tomograph (CTI/Siemens, Knoxville, TN, USA). Each dynamic scan was binned as 21 time frames (12 × 5 s, 6 × 10 s and 3 × 20 s) immediately after bolus inhalation of [¹⁵O]O₂ (500 MBq or 800 MBq) or an intravenous bolus injection of [¹⁵O]H₂O (500 MBq). All images were reconstructed as a 256 × 256 × 207 matrix with 1.22 mm isotropic voxels. The reconstructed images were corrected for attenuation, radioactive decay, random and scatter events, and detector dead time. The [¹⁵O] radioactivity in arterial blood was measured in the left radial artery using continuous blood sampling with a detector (Allogg AB, Mariefred, Sweden). The blood radioactivity was cross-calibrated with the tomograph and corrected for external delay and dispersion, using a similar approach as Aanerud et al. (2011).

All [¹⁵O]O₂ and [¹⁵O]H₂O scans were processed using Medical Imaging NetCDF (MINC) Toolkit (https://bic-mni.github.io). First, each time frame of the dynamic PET recording was smoothed with an 8 × 8 × 8 mm full width at half maximum Gaussian filter. Then, an average image over all time frames was created and used for coregistration with the corresponding T1 MRI image. Region of interest (ROI) segmentations from the T1 image were then transformed into the PET image space. The segmentation comprised eight ROIs: one for each lobe (including both gray matter and white matter) on both hemispheres. Then, a time activity curve (TAC) was calculated for each ROI and the median value across ROIs was used to create a global TAC. Finally, delay correction was performed using the global TAC and the blood activity curve.

To calculate the parameters of interest (CMRO₂ from [¹⁵O]O₂ PET and CBF from [¹⁵O]H₂O PET), a two-compartment model (single tissue compartment) was used to calculate the unidirectional tracer clearance, K₁ (Blomqvist, 1984; Ohta et al., 1991, 1996), implemented by the Turku PET centre (https://gitlab.utu.fi/vesoik/tpcclib).

During [¹⁵O]O₂ PET, labeled molecular oxygen reaches the capillary via the bloodstream and a fraction of blood’s oxygen content, OEF, diffuses into the tissue where it is metabolized to [¹⁵O]H₂O. The remaining oxygen is cleared from the tissue by the blood flow. The clearance rate, $K1(O2)$, is thus $CBF⋅OEF$⁠. Knowing the oxygen concentration in arterial blood, $[O2]a$⁠, the CMRO₂ can be calculated as:

For [¹⁵O]H₂O PET, radiolabeled water is assumed to equilibrate instantaneously with the tissue compartment. Hence, the [¹⁵O]H₂O concentration in tissue and in the capillary and venous blood is assumed to be in equilibrium and the estimated $K1(H2O)$ equals CBF. An overview of the processing steps for both [¹⁵O]O₂ and [¹⁵O]H₂O scans is presented in Supplementary Figure 2.

The assumption of equilibrium between [¹⁵O]H₂O concentration in tissue and venous blood does not hold true as water diffusion across the blood-brain barrier is not instantaneous (Eichling et al., 1974). Consequently, under normal flow conditions, the CBF as measured by [¹⁵O]H₂O will be systematically underestimated by a factor, $E=QPETF$⁠, where $QPET$ is the CBF measured with [¹⁵O]H₂O, and F is the true CBF. The value of E can be determined by comparing CBF measured with [¹⁵O]H₂O to CBF measured with a freely diffusible tracer, such as [¹¹C]butanol. Herscovitch et al. (1987) measured CBF with both tracers and found that E was relatively uniform throughout the brain with an average value of $E=0.84±0.07$⁠. To account for this systematic underestimation of CBF in the present study, all CBF values from [¹⁵O]H₂O PET were divided by 0.84.

Knowing the CBF and the influx rate of oxygen in [¹⁵O]O₂, $K1(O2)$⁠, the OEF can be determined from equation (5) as:

Voxel-wise $K1(O2)$ and CBF were calculated as the mean of the two [¹⁵O]O₂ and the two [¹⁵O]H₂O PET scans, respectively, for each subject.

The values of P_tO₂ and k in the biophysical OEF model are not known a priori and must be set to calculate OEF (see Eqs. 1 and 2). In the original model, the parameters were set to: P_tO₂ = 25 mmHg and k = 118 s^-1 based on rat studies during forepaw stimulations to yield a resting OEF = 0.3 (Jespersen & Østergaard, 2011). The calculated OEF is dependent on both P_tO₂, k, and the transit time distribution (MTT and CTH). To determine the influence of one parameter on OEF, the remaining parameters must be kept constant. In the following, it is described how a value of P_tO₂ is determined, followed by the calibration of k.

Measurement of P_tO₂ is faced by several methodological challenges and may vary greatly depending on tissue type and physiological state (Swartz et al., 2020). However, assuming tissue-dependent values for P_tO₂ is infeasible due to the intrinsic partial volume effects in DSC-MRI, and hence we chose to use a single value for P_tO₂ for the entire brain. Any selected ‘universal’ P_tO₂ value must be consistent with OEF values observed in normal brain tissue. To find ‘physiological’ P_tO₂ values, we first note that as k tends to infinity, the transit time distribution no longer affects the OEF estimate. In this extreme case, blood and tissue oxygen concentrations equilibrate instantaneously regardless of MTT and CTH—as implicitly assumed in the tracer kinetic analysis of [¹⁵O]O₂ PET data. For infinite k, OEF can therefore be calculated as a function of P_tO₂, and thus, P_tO₂ can be fixed to yield OEF values that are in the range of values reported from normal brain tissue in the literature. Based on a literature review of human PET by Fan et al. (2020), we chose OEF = 0.6 as the upper limit of OEF values for non-ischemic human brain tissue, corresponding to P_tO₂ = 21.8 mmHg (Fig. 1). This is different from the original model in which P_tO₂ was set to 25 mmHg, which yield 0.5 as the upper limit for OEF.

With P_tO₂ fixed, k can now be calibrated based on the hemodynamic measurements (MTT and CTH) to produce a given OEF. Previous reports have calibrated k to yield OEF = 0.3 in NAWM (Eskildsen et al., 2017; Madsen et al., 2023; Mouridsen et al., 2014). By utilizing the PET OEF data in the present study, it is possible to calibrate k in individual subjects according to their measured PET OEF. Like previous reports, this was done for NAWM, as this area is expected to be minimally affected by pathological hemodynamic changes.

Finally, the agreement between DSC-MRI OEF and PET OEF was assessed using correlation analysis and Bland-Altman plots.

A total of 68 healthy elderly individuals (23 males and 45 females, mean age = 64.7 years (std = 5.0)) were recruited and completed both [¹⁵O]-PET and DSC-MRI. All participants underwent [¹⁵O]-PET and DSC-MRI at two different visits. The mean time between the PET and MRI was 211 days [-182; 442] (negative values indicate that the PET scan was acquired before the MRI scan). The influence of this time difference is discussed in section 4.3.

An overview of mean values of PET OEF, MTT, and CTH in NAWM and gray matter is presented in Table 1. The mean calibrated k was found to be k = 68 s^-1 (range: 22; 175). There was a no significant correlation between k and age (p = 0.115, Fig. 2a), and no significant difference in k was observed between males and females (Fig 2b).

In summary, selecting a P_tO₂ value that allows OEF up to 0.6 resulted in an optimal value of P_tO₂ = 21.8 mmHg. Using acquired data to calibrate k to the individual subjects’ mean PET OEF resulted in an optimal k = 68 s^-1.

To evaluate the agreement between OEF estimated using PET and DSC-MRI, respectively, the correlation between the two measures is shown in Figure 3a (NAWM) and Figure 3b (gray matter). The corresponding Bland-Altman plots are shown in Figure 3c (NAWM) and Figure 3d (gray matter). There were statistically significant positive correlations between PET OEF and DSC-MRI OEF in both NAWM (r = 0.312, p = 0.009) and gray matter (r = 0.311, p = 0.009). However, the ranges of OEF values, as determined by DSC-MRI, were smaller than the range of PET based OEF values (Table 1). This is also evident from the Bland-Altman plots, which show a systematic overestimation of DSC-MRI OEF compared to PET OEF for lower OEF values, and a systematic underestimation of DSC-MRI OEF compared to PET OEF for higher OEF values. This is seen in both NAWM (mean difference: 0.0 (95% CI: -0.098; 0.097)) and gray matter (mean difference: -0.038 (95% CI: -0.133; 0.057)). Average DSC-MRI OEF and PET OEF images are presented in Supplementary Figure 4, and detailed comparisons between different brain regions are presented in Supplementary Table 1 and Supplementary Figures 5-7.

In this study, the biophysical model of OEF estimation based on hemodynamic measurements using DSC-MRI (Jespersen & Østergaard, 2011) was calibrated and compared to quantitative OEF measurements obtained with [¹⁵O]-PET. The parameters of the biophysical model were set to P_tO₂ = 21.8 mmHg and k = 68 s^-1, based on individual PET OEF and hemodynamic measurements (MTT and CTH) in NAWM. A moderate correlation was found between the PET OEF and DSC-MRI OEF in both NAWM and gray matter (p < 0.01). However, the range of OEF values from the biophysical model was smaller than the PET OEF, causing overestimation of lower OEF values and underestimation of higher OEF values compared to PET.

The results demonstrate that the biophysical OEF model can provide OEF estimates in relatively good agreement with the PET OEF, which is considered the standard reference OEF measurement; however, quantitative measurements may differ between models, especially in more extreme cases.

The smaller range of OEF estimates produced by DSC-MRI might reflect the assumptions of constant P_tO₂, saturation, oxygen-carrying capacity, and oxygen affinity in the biophysical model. Individual variations in these parameters would change the OEF in ways that are not accounted for when keeping these parameters constant in the biophysical OEF model. For example, higher OEF values are expected to be accompanied by slightly lower P_tO₂ and a higher oxygen concentration gradient between the blood and the tissue, and vice versa for lower OEF values. Omitting this shift in P_tO₂ would cause a systematic underestimation of higher OEF values and overestimation of lower OEF values by the DSC-MRI approach and thus contribute to the systematic differences between DSC-MRI and PET OEF estimates that we observed. The oxygen extraction model by Jespersen & Østergaard does not explicitly incorporate oxygen metabolism. A subsequent model of oxygen extraction, which explicitly includes oxygen metabolism and relaxes the assumption of constant P_tO₂, predicts OEF to vary over a larger range across various physiological conditions than the model used in this study, consistent with the discussion above (Angleys et al., 2014). While this model’s predictions are likely to correspond better with the PET OEF data, it also includes additional parameters, adding complexity to its parametrization.

Given the model assumptions, special caution should be exercised if applying it to DSC-MRI data acquired in subjects with conditions that might affect model parameters. This includes diseases such as acute stroke, where the P_tO₂ is dramatically reduced in the infarct region (H. Shi & Liu, 2007), or in sickle cell disease, where the oxygen-carrying capacity and oxygen affinity is reduced (Farrell et al., 2010; Kavanagh et al., 2022). Although the model parameters can, in principle, be set to reflect the physiological changes observed in diseases, further validation of the biophysical OEF model by PET in such diseases, as well as in different age groups, is warranted.

The one-tissue compartment model used in the [¹⁵O]H₂O PET analysis assumes instant equilibration of the PET tracers between the tissue and blood compartments. However, it has been established that [¹⁵O]H₂O extraction is limited by capillary transit time (Eichling et al., 1974) and therefore does not reach equilibrium under normal flow conditions. This causes a systematic underestimation of CBF of approximately 20% (Herscovitch et al., 1987). The fraction of complete equilibrium achieved by [¹⁵O]H₂O has been estimated to reach 84% under normal physiological conditions and to be relatively uniform throughout the brain (Raichle et al., 1983). Like other freely diffusible substances, the [¹⁵O]H₂O extraction fraction is influenced by both CBF and CTH (Jespersen & Østergaard, 2011). The [¹⁵O]H₂O extraction and, in turn, the mean fraction of complete equilibrium would therefore be less than 84% in a capillary network showing increased cerebral flow or capillary flow heterogeneity (increased CTH relative to MTT), and the correcting factor accounting for this incomplete equilibrium, E = 0.84, would thus need to be adjusted accordingly. Gray matter blood flow is approximately three times larger than that of white matter (Helenius et al., 2003; Østergaard et al., 1998) and hence, the gray matter CBF might be underestimated, resulting in OEF overestimation in the PET model (see Eq. 6). The bias caused by CBF underestimation in the PET model could contribute to the discrepancies observed between the PET- and MRI-based OEF estimates. However, further experimental data are needed to validate this hypothesis.

Individual, average NAWM MTT, CTH, and PET OEF values were used in the calibration of the biophysical model. However, as seen in Figure 3 and Supplementary Figure 6, the OEF estimates produced by the biophysical model are lower in gray matter compared to NAWM. This was not observed for PET OEF estimates, which is similar for gray matter and NAWM. The difference in DSC-MRI OEF estimates between gray matter and NAWM originates from the larger flow (lower MTT) in gray matter (mean MTT = 3.40) compared to NAWM (mean MTT = 4.52), resulting in lower OEF estimates given the same model parameters. Ideally, the biophysical model should be calibrated for each tissue type. However, there are several drawbacks to tissue-specific calibration, including partial volume effects, increased model complexity, and the risk of model overfitting. Consequently, we selected a single set of model parameters for the entire brain. Accordingly, a comparison of DSC-MRI OEF estimates is feasible within the same tissue type, but not reliable between different tissue types. The intersubject variation in k (Fig. 2) reflects that different subjects require different k value for the absolute OEF values to correspond between the biophysical model and PET. However, given the differences in scanning dates, model assumptions, and noise levels between the methods, a perfect correspondence is not appropriate and would cause overfitting of the biophysical model. Hence, despite the differences between the estimates, a single k value for all subjects was selected to produce a more general model.

The PET and MRI scans were acquired on different days with a mean time between scans of 213 days. OEF has been shown to increase with age (Aanerud et al., 2011); however, the age-effect is expected to be negligible compared to day-to-day variations, which have been shown to be up to 10% for OEF (Bremmer et al., 2011). These variations would also affect the agreement between the two methods. However, the day-to-day variations are expected to be random and hence cannot explain the OEF bias observed in the MRI model. Future studies comparing the different methods would benefit from utilizing a hybrid PET/MRI scanner to provide simultaneously acquisition, which would limit the effects of day-to-day variations in CBF and OEF. Additionally, both PET and DSC-MRI have relatively low resolution, making both image modalities susceptible to partial volume effects. Finally, participants with reduced kidney function (eGFR <60) were excluded for DSC-MRI, hence the model is not suitable for every patient population.

In summary, we have calibrated the model parameters of the biophysical OEF model based on hemodynamic measurements using DSC-MRI. We found a moderate correlation between OEF estimations from the biophysical OEF model and OEF estimated from [¹⁵O]-PET in healthy elderly subjects, which is considered the standard reference OEF measurement. Due to a smaller range of OEF estimates by the biophysical model, we saw a systematic underestimation of higher OEF values and a systematic overestimation of lower OEF values compared to PET OEF. Despite this, the biophysical OEF model provides a valuable tool to assess how hemodynamic changes can affect overall oxygen availability and could potentially provide insights into the effects of macrovascular and microvascular dysfunction, respectively, in brain disorders.

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

M.K.T. and L.Ø. conceived and designed the research; M.K.T. performed experiments; L.S.M. analyzed the data; L.S.M., H.A., I.K.M., D.J.B., S.F.E., and L.Ø. interpreted the results of experiments; L.S.M. prepared the figures; L.S.M. drafted the manuscript; L.S.M., M.K.T., H.A., I.K.M., D.J.B., S.F.E., and L.Ø. edited and revised the manuscript; and L.S.M., M.K.T., H.A., I.K.M., D.J.B., S.F.E., and L.Ø. approved the final version of the manuscript.

This study was funded by the Lundbeck Foundation (grant no. R310-2018-3455).

L.Ø. is a scientific advisory board (SAB) member and minority shareholder in Cercare Medical A/S. The authors declare no other conflict of interest.

The authors thank Dora Grauballe, Michael Geneser, and Joel Aanerud from Aarhus University Hospital for their help with MRI and PET imaging and reviewing.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00562.