Fast Diffusion Kurtosis Mapping of Human Brain at 7 Tesla With Hybrid Principal Component Analyses

Diffusion kurtosis has become an important magnetic resonance imaging (MRI) modality for non-invasively mapping the microstructural variations in living tissues. Theoretically, the spatial resolution of diffusion kurtosis imaging (DKI) can be significantly improved by acquiring data at ultra-high magnetic fields (UHF, ≤ 7 Tesla) because of the increased signal-to-noise ratios. However, issues such as increased susceptibility artefacts and rapid signal attenuation inherent in UHF-MRI have impeded the adoption of DKI in research and clinics. In this paper, we developed a new image reconstruction algorithm for fast DKI at UHF. By integrating the one-dimensional and two-dimensional principal component analysis and compressed sensing technologies, the new algorithm can reconstruct kurtosis maps from highly undersampled data. The technique was validated using randomly undersampled brain images with a control database of fully sampled DKI acquisitions from healthy human participants. We compared the technique with zero-filling Fourier transform and similar compressed sensing algorithms by evaluating the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) between images and assessing the reproducibility of results using the Bland-Altman method. We found that our methods can achieve at least a five-fold reduction in data acquisition time at UHF with high image quality. Moreover, the PSNR and SSIM of five diffusion metrics generated by our methods were superior to the other algorithms when the undersampling rate is high and the echo time is short. The proposed method can be valuable for fast functional and dynamic-contrast imaging techniques at 7 Tesla or higher.