Introduction

Gliomas comprise a family of malignant brain tumors whose response to treatment is often challenging to assess amid a background of nonspecific changes on conventional anatomical imaging. With the recent success in using hyperpolarized carbon-13 (HP-¹³C) MRI techniques to probe glioma metabolism^1-4, there has been a drive toward improving capabilities for monitoring disease serially. In this work, we implemented an automated atlas-based prescription⁵ of HP-¹³C echo-planar imaging (EPI) to ensure consistent volumetric coverage across exams and match ¹H spectroscopy orientation, alongside other methodological advances.

Methods

Study Overview. Analysis of 29 patients previously diagnosed with glioma and 7 healthy volunteers included: atlas-based versus manual/axial HP-¹³C EPI prescription (3 patients; 14 exams); 2s versus 5s EPI acquisition delays triggered post HP tracer injection (31 subjects; 82 exams); and comparison of pyruvate frequency referencing methods⁵ (36 subjects; 106 exams).
Atlas-based HP-¹³C EPI. Dynamic HP-¹³C imaging was performed on a 3T MR750 scanner with a dual-tuned 24/8-channel ¹³C/¹H phased-array receiver. Data were acquired using a frequency-specific 2-D multislice EPI sequence⁷ (TR/TE=62.5ms/21.7ms, 8 slices, 20 timepoints, 3s temporal resolution) that featured 1.5cm isotropic resolution and constant 𝛼_pyruvate/𝛼_lactate/𝛼_bicarbonate = 20^o/30^o/30^o flip angle scheme. Automated prescription entailed 1) registering T₁ pre-contrast images to the MNI152⁵ brain atlas, 2) transforming pre-defined ¹H point-resolved spectroscopy (PRESS) coverage templates back to patient space; 3) exporting the 3-plane oblique orientation as an XML file⁵; and 4) extracting the center/orientation for EPI (Fig. 1A). A SPINlab system provided dynamic nuclear polarization of [1-¹³C]pyruvate; 0.43mL/kg was intravenously injected at 5mL/s, followed by 20mL saline².
HP-¹³C Post-processing. EPI data were prewhitened⁸, channel-combined using complex weights from the fully sampled pyruvate signal⁹, phased and denoised¹⁰. Combined kinetic modeling of pyruvate-to-lactate (k_PL) and pyruvate-to-bicarbonate (k_PB) conversion was performed on dynamic data using an inputless model¹¹.
Atlas-based Analysis. Consistency of volumetric coverage for serial atlas-based EPI was measured using Volume_PRESS Dice coefficients: $$Dice = \frac{2\mid{Volume1_{PRESS} \cap }Volume2_{PRESS}\mid}{{\mid{Volume1_{PRESS}}}\mid + \mid{Volume2_{PRESS}}\mid}$$ Kinetic data from manual and atlas-based prescriptions acquired within the same exam were qualitatively compared for consistency within voxels containing >30% normal-appearing white matter (NAWM). Since atlas-based EPI was hypothesized to improve hemispheric data symmetry, asymmetry maps¹² were generated for normalized k_PL data with zero mean and unit standard deviation. Given the set of points P_L (p∈P_L) and P_R (q∈P_R) that are perfectly mirrored about a left-right line of symmetry l, local asymmetry¹² s was defined as $$s(p;l) = {\min_{q\epsilon P_{R}}} K(f(p), f(q))$$, where K is the Euclidean distance between vectors $$ f(p)= \binom{f_{patch}(p)}{wf_{coord}(p)}, f(q)= \binom{f_{patch}(q)}{wf_{coord}(q)}$$ , which contain the elements of 3x3 patches with central points p/q and coordinates that are empirically weighted by w (and mirrored about l). Global asymmetry S was calculated as $$S =\frac{1}{N}\sum_{p\epsilon P_L\cup P_R} s(p;l)$$ , where N is the number of data samples, and was compared for 11 total manual versus atlas-based serial EPI scans in patient P-01 (Wilcoxon ranked-sum).
Frequency reference. Pyruvate f₀ referenced using a coil-embedded 1mL 8M-urea phantom² versus water with empirical correction factor (0.251491899) was evaluated via FID-CSI with non-selective excitation (Wilcoxon).
Acquisition delay. EPI data acquired with 5s versus 2s delays triggered following injection were compared according to how much of the initial rise in pyruvate was captured (Wilcoxon). Spline-interpolated dynamic traces were temporally shifted and resampled back to 3s resolution to simulate the effects of delaying acquisition >2s on capturing the rise in pyruvate and kinetic analysis.

Results

Atlas-based EPI. A diagrammatic description of atlas-based HP-¹³C EPI is provided in Figure 1, along with example serial data from patient P-01, whose 4 scans spanning 159 days demonstrate consistent volumetric coverage based on dice scores ranging 0.965-0.978. Values of k_PL-NAWM and k_PB-NAWM are shown in Figure 2 to be consistent across atlas-based EPI exams (P-01) and comparable to manually-prescribed EPI exams acquired for second injections in select patients (P-01, P-02). Atlas-based EPI (n=4) also showed greater hemispheric symmetry in k_PL for patient P-01 (Fig. 3A) versus manual prescription (n=7, p=0.006; Fig. 3B), which helped distinguish tumor from non-pathologic k_PL elevation.
Frequency reference. Absolute pyruvate f₀ offsets for urea/water references (8.5±11.6/3.7±5.4Hz; median±SD; n=68/38; p=0.00003) are shown with EPI excitation spectral response in Fig. 4A,B.
Acquisition delay. Figure 4C,D defines percent rise in pyruvate and shows a histogram for 5s (67±28%; median±SD) versus 2s (90±22%) acquisition delays, with the 2s delay enabling the capture of additional inflow dynamics (p=0.04). In Figure 5, the simulated effects of acquisition delay in a patient replicate the 67% median %rise in pyruvate captured among exams with ~5s delay. The relative effects on k_PL and k_PB modeling are presented in Fig. 5C-F for different thresholds of NAWM, and show that error estimates increase most prominently with <40% capture of pyruvate rise.

Discussion

This work demonstrated methodological improvements in serial HP-¹³C metabolic imaging of patients with glioma. Atlas-based EPI prescription achieved >96% consistency in volumetric coverage across serial scans, enabling voxel-wise comparisons between exams, while matching the 3-plane oblique orientation of ¹H spectroscopy. Besides offering similar NAWM kinetics, the atlas-based prescription showed improved hemispheric symmetry in k_PL that helped highlight tumor. The pyruvate center frequency was best estimated using a water reference and the 2s acquisition delay ensured adequate capture of pyruvate inflow to reduce potential kinetic modeling error.

Acknowledgements

This work was supported by NIH Grants P01 CA118816, P41 EB0341598, P50 CA097257, and T32 CA151022 (AA).