Introduction

Dynamic oxygen-17 (¹⁷O) MRI allows non-invasive measurement of cerebral metabolic rate of oxygen (CMRO₂) by tracking the accumulation of metabolically generated ¹⁷O-water (H₂¹⁷O) after inhalation of ¹⁷O-oxygen (17O₂).¹ In vivo experiments have shown a linear increase in ¹⁷O signal during a brief period (2 to 3 min) of ¹⁷O₂-inhalation, with the slope of the line proportional to CMRO₂.^2,3 However, dynamic ¹⁷O-MRI suffered from limited spatial and temporal resolution due to low MR sensitivity and fast T₂ decay of the ¹⁷O signal. Model-based reconstruction has been applied to estimating physiologic parameters in dynamic contrast-enhanced MRI studies for improved measurement accuracy. In this study, we aimed to evaluate the efficacy of using model-based approach for CMRO₂ mapping in both simulation and in vivo studies on post-stroke mice.

Methods

Data Acquisition: Dynamic ¹⁷O data was acquired by a 3D ultrashort echo-time (UTE) sequence with a flip angle of 90° and TE/TR of 0.075/10 ms. Sampling of k-space data used center-out, golden-means-based 3D radial trajectory to allow near uniform coverage of k-space with arbitrary number of radial spokes.⁴ Each radial spoke acquired 8 data points in 1-ms readout time covering an FOV of 24x24x24 mm³, leading to a nominal isotropic resolution of 1.6 mm.
Model-Based CMRO₂ mapping: Maps of CMRO₂ were estimated by solving the following optimization problem,_⁵
$$\min_{CMRO_2,C_0}\frac{1}{2}\parallel F[f(t;CMRO_2,C_0,\alpha,\beta)-k(t)]\parallel^2+\lambda\cdot TV(CMRO_2)$$
where $$$F$$$ represents the non-uniform Fourier transform operator, $$$k(t)$$$ represents the acquired k-space data. $$$R(\cdot)$$$ is the regularization function calculating the total variation in the CMRO₂ map,⁵ $$$\lambda$$$ is the corresponding regularization factor. The signal model $$$f$$$ is related to CMRO₂ and the initial H₂¹⁷O concentration ( $$$C_0$$$) as
$$f(t;CMRO_2(r),C_0(r),\alpha,\beta(r))=\beta(r)\cdot (C_0(r)+2\alpha\cdot CMRO_2(r)\cdot t)$$
$$$\alpha$$$ is the ¹⁷O₂-enrichment level and $$$\beta$$$ is the coil sensitivity map estimated from data acquired at baseline.
Simulation Study: A 3-compartment digital phantom was used in the simulation. The ¹⁷O signal of an 11-min data acquisition protocol was simulated with 8 min of baseline acquisition and 3 min of data acquisition during ¹⁷O₂-inhalation. CMRO₂ in each compartment was assumed to be 1.3, 2.0, and 2.5 μmol/g/min, respectively. Literature value of baseline H₂¹⁷O concentration (20.35 μmol/g) was used.⁶ White noise at two different SNR levels ($$$SNR_{high}=2\cdot SNR_{low}$$$) were added to the simulated k-space data. Data were simulated at a higher spatial resolution (0.8 mm) with 16 data points acquired on each radial spoke. CMRO₂ maps were generated using model-based reconstruction. The choice of regularization parameter was evaluated.
In Vivo Study: Adult male C57BL/6 mice (N=1) underwent MRI studies at 2 hours after 60 min of middle cerebral artery occlusion (MCAO) surgery. All MRI studies were performed on a 9.4 T Bruker scanner. ¹⁷O-MRI used a 2-cm surface coil. 250-ml of gas composed of 30% ¹⁷O-enriched oxygen (70% enrichment) and 70% N₂ mixed with 1% isoflurane was delivered to the animal through a nose cone for 2.5 to 3 min. Dynamic ¹⁷O data were continuously acquired at baseline (8 min) and during inhalation. CMRO₂ was estimated using model-based reconstruction. The mice were euthanized after MRI studies. The brains were harvested and sectioned into 1-mm slice for 2,3,5-Triphenyltetrazolium chloride (TTC) staining.

Results

Simulation Results: Figure 1 shows the cost function and CMRO₂ maps at selected iteration steps. With a regularization factor of 5e6, the regularization term only contributed to ~1% of the cost function yet overfitting to the noise can be effectively prevented. Figure 2 compares the CMRO₂ maps estimated at different noise levels using different regularization factors. In general, lower SNR required stronger regularization. Our results suggested that regularization consisting of 0.1 to 5% of the cost function can maintain fidelity to data while avoid overfitting.
In Vivo results: Figure 3 shows the CMRO₂ maps of a post-stroke mouse brain. A regularization factor of 2e5 was used consisting of 0.6% of the cost function. The average CMRO₂ was ~2.1 μmol/g/min in the contralateral hemisphere, while reduced to ~1.7 μmol/g/min on the ipsilateral side. The infarct shown in the TTC staining reside in the same area with CMRO2 at ~1 μmol/g/min.

Discussion and Conclusion

The model-based reconstruction allowed CMRO₂ mapping in the mouse brain with 1.6-mm nominal resolution at 9.4 T. Low CMRO₂ was found in the TTC-stained infarct core. Other regularization method can be tested in the future.

Acknowledgements

This work was supported by a grant from the National Institute of Health (R01 EB23704).