Introduction

Systemic energy deficiency, traditionally measured by ³¹PMRS (phosphorous magnetic resonance spectroscopy) as the delayed phosphocreatine (PCr) recovery in exercised muscle, has been implicated in various disorders, including primary mitochondrial disorders(1), muscle injury(2), cardiovascular disease(3) and diabetes mellitus(4). However, ³¹PMRS suffers from poor spatial resolution. Recently developed 3D implementation of CrCEST, with temporal resolution of 30s(5), provides higher SNR and increased coverage. Exploiting these two advantages, we present a non-local spatially regularized approach to reconstruct parametric map for recovery time constants.

Methods

3D CrCEST MRI was acquired at a 7.0T MRI scanner (Siemens) using a 28-Channel phased-array knee coil. First baseline CrCEST data was performed for 2 minutes, followed by 2 minutes of mild plantar flexion exercise (with 30 push/second, air pressure 8 pound per square inch) and then 8 minutes of post exercise CrCEST imaging. The sequence consisted of the pulse train (5x100ms Hanning windowed, duty cycle 99%, B1rms =2.9 μT), followed by single shot GRE read out with TR =3.5ms, TE =1.47ms, BW =710Hz/pixel, T1 recovery delay =5s. Other imaging parameters were: FOV = 160x160x40mm³; FOV phase 100%; spiral centric encoding order with elliptical scanning, GRAPPA factor = 2, averaging = 1; # slices = 8; BW = 710 Hz/pixel; shots = 1 in transverse orientation. Raw CEST images were acquired at 6 saturation offset frequencies ±1.5, ±1.8, ±2.1ppm (relative to water resonance). WASSR images (from ±0 to ±0.9 ppm with a step-size of ±0.15 ppm), with a saturation pulse at B1rms of 0.29μT with 200ms duration, was used to correct for B0 in homogeneities(6). B1 Calibration: Baseline CEST weighted images were acquired for following B1rms= 1.31, 1.6, 2.04, 2.47, 2.91, 3.2, 3.64, 4.08, 4.37 μT and corresponding CrCEST maps were calculated. Finally, the calibration was calculated using linear model (i.e. polynomial of order 1) on the voxelwise basis. Two healthy human volunteer (males, 28, 40 years old) participated in the approved study protocol. This volunteer #1 was known to have slow creatine recovery (recovery time constant >100s), most likely due to his ~2 hour/day bike commute. Based on post exercise PCr spectra, acquired in a separate session, the acidosis was confirmed.
Recovery Parameters Fitting Algorithm: Mono-exponential course was assumed for creatine recovery curve
$$y_{k}(t) = A_0 + A_1 exp(-\frac{t}{\tau_{Cr}})$$ where $\tau_{Cr}$ = Cr recovery time constant, A₀ = baseline level and A₁ = interpolated values of post exercise CrCEST level at t approaches 0s. Starting with “noisy” time series data at any voxel “i" and time frame “k”: y_exp(i) = y_k(i), the weighted average over spatial domain can be written as:
$$y_{k}^{Smooth}(i) = \sum{W(i,j)y_{k}(j)}, where \, k \in {1,2,3, ...t} ...[1]$$
Calculation of weights: Out of the 16 time points acquired, only first eight of the decay curve were used for calculating the weight factors. The selected segment of time recovery curve was normalized and then Exponential Divergence score, Dij, was calculated between voxels i,j using formula given in(7). The weight (w(i,j)<1) , between i^th and j^th voxels in a neighborhood of radius 7, was calculated by combining the similarity and distance metric:
$$ W(i,j) = \frac{1}{\sqrt{(i_x-j_x)^2 + (i_y-j_y)^2 + (i_z-j_z)^2)}} \times \frac{(e^{-\frac{Dij}{h}})}{\sum_j{(e^{-\frac{Dij}{h}})}}$$ where h is the tuning parameter and was set to be ~0.001 empirically. Finally, we inverted the smoothed data (Eqn. [1]) using nonlinear least square solution utilizing Levenberg-Marquardt-Fletcher algorithm(8).

Results

Both volunteers utilized lateral gastrocnemius (LG) and medial gastrocnemius (MG) muscles during the exercise. A major part of LG muscle show slow recovery. Fig. 1 compares the performance of the regularized approach against non-regularized approach. By suppressing the noisy instability, the regularized approach resulted in better depictions of underlying parametric maps in LG, MG. As other muscle group did not get utilized in the exercise, asymmetry change post exercise were not significant. The parametric map for volunteer 2 showed significantly slower recovery kinetic in terms rate constants and a higher post exercise elevation in % Cest asymmetry (Fig. 2). Volunteer 2 showed much faster recovery, which is what we typically expect from a healthy volunteer with sedentary to moderately physically active life style (Fig. 2).

Discussion

The proposed algorithm resulted in a noise robust maps of recovery time constants and other related parameters. Parts of the parametric maps, for volunteer 2 with faster Cr recovery, were noisy/patchy. On the basis of empirical evidence, it appears that numerically robust parametric map reconstructions are only possible if τ_Cr >3x temporal resolution. Fortunately, this condition would easily be met for patients with compromised creatine kinetics(1), (Cr recovery rates >1.5 minutes). However, this condition may not met be satisfied for healthy volunteers with sedentary to moderately physically active life style. We are currently working on developing experimental methods to reduce the temporal resolution to ~15s and that would eliminate this restriction. In future, we will perform simulations to investigate the τCr cut-off for reliable parametric reconstruction. We also plan to perform scan-rescan on a few patients with delayed CK kinetics.

Conclusions

To the best of our knowledge, this is the first time demonstration of the ability to produce inter- and intra- muscular variabilities of creatine recovery specific parameters in exercised muscle and may provide a better imaging biomarker in disease conditions with delayed CK recovery.

Acknowledgements

This project was supported by National Institute of Biomedical Imaging and Bioengineering of the National Institute of Health through grant number P41-EB015893 (NIH/NIBIB) and R56-AG062665 (NIH).