For histological estimation of brain tumors, it is essential to identify the most aggressive tumor areas for sampling¹. Whereas high resolution anatomical MR images are available, assessment of functional indicators like cerebral blood volume (CBV) is usually restricted to low spatial resolutions due to a trade-off between temporal resolution and quantification fidelity. This work aims at performing cerebral perfusion acquisitions with high spatial resolution without sacrificing temporal resolution. This is achieved using PEAK-EPI^2,3 and by employing a new reconstruction technique: PS-SPIRiT.

Dynamic-susceptibility-weighted imaging of cerebral perfusion, incorporating peripherarily injected gadoteridol tracer, is performed based on k-t-undersampled EPI acquisitions with inplace Nyquist-sampled ACS data and PEAK-GRAPPA⁴ reconstruction. Further, a new reconstruction method that integrates temporal basis functions (derived based on the partial separability^5,6 of (k,t)-space data) with parallel imaging (employing SPIRiT^7,8) is evaluated.

Two repeated half-dose first-pass bolus perfusion acquisitions were performed (acquisition parameters in Fig.$$$~$$$1) in six patients (only two shown) with different diagnostic backgrounds on a$$$~$$$3T clinical scanner (Tim TRIO, Siemens, Erlangen, Germany): 1.$$$~$$$a standard clinical EPI protocol and 2.$$$~$$$PEAK-EPI at reduction factor $$$R=5$$$ with our in-house developed (PEAK-)EPI sequence^2,3. With PEAK-EPI, a nominal spatial resolution of 1.4$$$~$$$mm was achieved, as opposed to a nominal resolution of 1.8$$$~$$$mm in the standard measurement.

The further proposed PS-SPIRiT reconstruction was achieved based on solving the following inverse problem for the unknown image weights $$$U\in\mathbb{C}^{N_{x}N_{y}\times{L}}$$$ and given the sampled data $$$d\in\mathbb{C}^{N_x(N_y/R)N_tN_c}$$$:

$$\arg\min_{U\in\mathbb{C}^{N_{x}N_{y}\times{L}}}~||~d-(\mathbf{\Omega}\circ\mathbf{F})(U\hat{V})~||~^2_2+||~\mathbf{\hat{G}}(U\hat{V})-U\hat{V}~||^2_2+R(U),$$

where $$$\hat{V}\in\mathbb{C}^{L\times{N_t}}$$$ and $$$\mathbf{\hat{G}}$$$ refer to the a priori determined temporal basis functions and SPIRiT-reconstruction operator, both estimated from the Nyquist-sampled ACS data at full temporal resolution. The operators $$$\mathbf{F}$$$ and $$$\mathbf{\Omega}$$$ denote the coil-wise Fourier transform and undersampling mask, respectively. The subspace dimension was restricted to $$$L=8$$$ and $$$R(U)$$$ was chosen to promote spatial-spectral sparsity,$$$~$$$e.g.⁹.

The standard EPI was additionally reconstructed offline with POCS¹⁰ to account for Partial-Fourier-sampling. CBV maps were derived off-line and re-scaled to $$$\left[0,1\right]$$$ for depiction with a common threshold in case of PEAK-, PS-SPIRiT- and POCS-EPI. The first two were further compared in terms of tSNR, average temporal evolution and correlation of relative CBV values.

Figure$$$~$$$2 and$$$~$$$3 demonstrate the improved spatial resolution for PEAK-EPI acquisitions with PEAK-GRAPPA and PS-SPIRiT reconstruction. Fine capillary structures - whose composition yield a marker of tumor growth - and prominent veins draining into the left ventricle are much better distinguishable in the proposed acquisition as compared to the standard protocol (Fig.$$$~$$$2). Due to the better spatial resolution in the PEAK-EPI acquisition, the small lesion in the cingulate cortex as shown in Fig.$$$~$$$3 is easier to distinguish from the vessels of the anterior interhemispheric fissure than in standard EPI. The high resolution CBV maps further exhibit finer gradations of CBV within the tumorous area. As further demonstrated in Fig.$$$~$$$3, the PEAK-EPI measurement is less affected by susceptibility artifacts due to the reduced readout length (white oval). Comparison between PEAK- and PS-SPIRiT-EPI indicate a slightly better tSNR for the latter (Fig.$$$~$$$4) and correlated CBV values with higher -$$$~$$$yet relative$$$~$$$- values for PS-SPIRiT-EPI (Fig.$$$~$$$5). However, only one case is regarded in the comparison.

The achieved echo train length reduction and faster k-space traversal mitigates distortion and susceptibility artifacts, while facilitating an increase of spatial resolution without sacrificing temporal resolution. In the Cartesian acquisition scenarios used in our method, this was achieved in both, PEAK-EPI with PEAK-GRAPPA reconstruction as well as with the new PS-SPIRiT reconstruction. High resolution information is substantial in angiogenesis assessment. When comparing relative CBV values emulating equivalent windowing, the additional information provided by the high resolution CBV maps is clearly visible.

The flexible PS approach allows for various undersampling patterns, which were not assessed here. Next steps therefore comprise further analysis of the optimal parameters for PS-SPIRiT and to explore more complex undersampling scenarios, potentially enabling higher reduction factors. Currently, a static SPIRiT-kernel was employed which will be extended to dynamic SPIRiT-kernels in further experiments.

The presented scenarios can be readily applied in multi-echo acquisitions, e.g.¹¹, where either the higher reduction factor can be used to increase the echo train length or SNR can be improved with the same reduction due to PEAK-GRAPPA¹². Our methods can be readily combined with multiband-EPI^13,14,15 to further increase temporal resolution of the measurement.

The high resolution CBV maps obtained with PEAK-EPI acquisition and the two presented reconstructions facilitate detailed delineation of tumor borders and provide a more precise location of the most aggressive areas in tumorous tissue. Therefore, PEAK-EPI provides high resolution functional information ready to assist more precise grading of tumors in stereotactic surgery biopsies.