Quantitative Susceptibility Mapping for Intracranial Arteries

QSM images were reconstructed from the magnitude and phase images acquired by multi-echo GRE sequence using STI Suite (https://people.eecs.berkeley.edu/~chunlei.liu/software.html) implemented in Matlab R2016a (The Mathworks)^{51 (link)}. The phase images from the multi-echoes were unwrapped using the Laplacian-based method, then the normalized phase was calculated based on a method by Li et al.^{52 (link)}. The magnitude images were masked to include the intracranial arteries while excluding other noisy regions (clivus and condyles of the occipital bone). The 9 masks were generated by adjusting the fractional intensity threshold using the FSL BET software⁵³ , and an appropriate mask containing the best ROI including target vessels was selected through visual inspection. Using the normalized phase and the magnitude mask images, the tissue phase images were created after background removal using the magnitude mask and variable-kernel sophisticated harmonic artifact reduction for phase data method. Finally, QSM was calculated using the streak artifact reduction for QSM method^{8 (link),9 (link),54 (link),55 (link)}. Overall processing of QSM reconstruction took about 5 min for each patient on a personal computer equipped with a 1.8 GHz processer (Intel Core i7; Intel) and 16 GB memory. QSM was normalized relative to the CSF in the posterior horns of the lateral ventricles, avoiding the surrounding brain tissue and choroid plexus.