Cell and tissue viability is closely connected to energy balance and thus to the oxygen metabolism. Oxygen-17-MRI (¹⁷O-MRI) is capable of directly measuring the localized H₂¹⁷O-concentration in-vivo (nat. abundance 0.037%) which also is the turnover product of oxidative phosphorylation. In combination with inhalation of ¹⁷O₂ a localized cerebral metabolic rate of oxygen consumption (CMRO₂) map is feasible¹. However, the ¹⁷O-nucleus experiences extremely fast transverse relaxation due to its electrical quadrupole interaction (I=5/2). Additionally, the in-vivo signal is reduced by 10⁵ compared to protons (¹H). This requires pulse-sequences that enable ultra-short echo-times and high SNR-efficiency such as 3D density adapted radial (3D-DAPR)² or twisted projection imaging³ to achieve nominal resolutions of ≲(7mm)³. Spherical acquisition schemes and T₂^*-relaxation additionally enlarge the full-widths-at-half-maximums (FWHM) of the point-spread-functions (PSF). Thus, strong partial-volume (PV) effects reduce the accuracy of quantitative ¹⁷O-MRI. Recently, a partial volume correction (PVC) algorithm⁴ was already successfully applied^5,6 to non-proton MRI. This algorithm relies on accurate knowledge of relaxation properties of considered compartments. In this study, an iterative PVC approach for improved T₂^*-determination was applied to simulations, phantom and in-vivo data. Subsequently, determined relaxation properties were used for PVC-¹⁷O-MRI.

First, the T₂^*-determination capability was tested in phantom simulations/measurements. Three cylindrical tubes, two filled with 5%-agar+ 0.9%-NaCl solution to alter T₂^*-properties and one filled with 0%-agar+ 0.9%-NaCl (r=1.5cm) solution where placed within a cylindrical reservoir (0.4%-NaCl solution). The same configuration was simulated. Known relaxation properties were attributed to the individual compartments: tubes with 5%-agar/ 0%-agar T₂^*=2ms/6ms, T₁=5ms/5ms.

Phantom simulations and measurements: A 3D-DAPR-sequence was applied for simulation and data acquisition (Fig.1). All data were reconstructed with a SNR-enhancing Hamming filter (FOV: 156x156x156mm³) and image data were B₁-corrected with a phase-sensitive method⁷. Imaging was conducted on a 7T MR-system (Magnetom 7T, Siemens AG) using a custom-built quadrature ¹⁷O/¹H-head-coil. Tube structures were segmented manually from a proton 3D-GRE-sequence (TR/TE=8.1ms/4.88ms, Θ=10°, (1mm)³). PVC was performed for each echo-data set individually, first without any T₂^*-consideration for PSF simulation. A mono-exponential function was fitted to data of individual compartments. Then T₂^*-values were iteratively adapted until the change in T₂^*-value was <1%.

In-vivo measurements: Data of three healthy volunteers (age 25±2) were acquired with a 3D-DAPR-sequence (echos: TE=0.56ms, 1.56ms, 2.5ms, 3.5ms, 6.0ms, 8.0ms; FOV: 260x260x260mm³) (Fig.2). A proton 3D-GRE-sequence (TR/TE=8.1ms/4.88ms, Θ=10°, (1mm)³) was used as registration basis. For PVC masks high resolution anatomical data were acquired using a 24-channel ¹H-head-coil. PV correction was performed for all data-sets considering three compartments (CSF, grey and white matter). Data were fitted mono-exponentially and T₂^*-values were adjusted iteratively. ¹⁷O-signal of the first data-set (TE=0.56ms) was then PV corrected considering two CSF compartments (lateral ventricles (CSF_i)/sulci (CSF_o)).

The influence of PV effects on T₂^*-determination was directly verified in simulations: Without additional correction, the relaxation time of the agar-compartment was overestimated by 40% (T₂^*=2.8ms). PVC in the first step reduced the discrepancy to 5%-7.5% and the following values remained within ±2.5% of the actual value (T₂^*=2.05ms). T₂^*-values of 0%-agar-compartments remained stable and showed little discrepancy. Similar results were obtained for experimental data of 5%-agar-compartments where the discrepancy to the final result (T₂^*=1.9ms) was even higher (60%, T₂^*=3.2ms). T₂^*-values for 0%-agar-compartments showed a slight difference (~13%, T₂^*=5.7/6.5ms) before and after correction. Exemplary fits are shown in Fig.3. In-vivo data showed a change in determined relaxation behavior for all three compartments (Tab.1). With corrected relaxation properties PVC water-content was quantified with different T₂^*-assumptions for PSF-simulation (Tab.2).PV effects strongly influence the determination of relaxation parameters. Phantom studies revealed a discrepancy for T₂^*-values of up to 60% if signal was not corrected properly. Phantom simulations allowed verification of the method: In simulations correct T₂^*-values were recovered and comparable results for ¹⁷O-MRI experiments were obtained. The iterative PVC led to an increase in CSF-T₂^* and to a decrease of grey-matter-T₂^* (Tab.1) as expected. T₂^* of white matter remained stable over all iterations as PV effects of neighboring long-relaxation compartments (CSF) have only a small influence. With adapted T₂^*-values for PVC improved water content quantification was possible where discrepancy between CSF_i and CSF_o was minimal and expected values where closest to literature values⁸ (Tab.2). However, grey and white matter values are still underestimated by 4-12%, most likely due to not fully corrected transverse relaxation.

The presented iterative approach leads to improved T₂^*-determination in ¹⁷O-MRI. More precise T₂^* relaxation times enable a more accurate ¹⁷O-MRI signal quantification. Combining exact T₂^*-values with PVC algorithms is of particular interest for signal correction of CMRO₂ experiments. This approach is also well transferable to other nuclei which face similar problems.