AUTHOR=Si Wenbin , Guo Yihao , Zhang Qianqian , Zhang Jinwei , Wang Yi , Feng Yanqiu 

TITLE=Quantitative susceptibility mapping using multi-channel convolutional neural networks with dipole-adaptive multi-frequency inputs

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2023.1165446

DOI=10.3389/fnins.2023.1165446

ISSN=1662-453X

ABSTRACT=Quantitative susceptibility mapping (QSM) quantifies the distribution of magnetic susceptibility and shows great potential in assessing tissue content such as iron, myelin, and calcium in numerous brain diseases. The accuracy of QSM reconstruction was challenged by ill-posed field-to-susceptibility inversion problem, which is related to the impaired information near the zero frequency response of dipole kernel. Recently, deep learning methods demonstrated great capability in improving the accuracy and efficiency of QSM reconstruction. However, the construction of neural networks in most deep learning-based QSM methods did not take the intrinsic nature of dipole kernel into account. In this work, we proposed a Dipole kernel-Adaptive Multi-channel CNN (DIAM-CNN) method for the dipole inversion problem in QSM. DIAM-CNN first separated the original tissue field into high-fidelity and low-fidelity components by thresholding the dipole kernel in frequency domain, and it then input the two components as additional channels into a multichannel 3D Unet. QSM maps from Calculation of Susceptibility through Multiple Orientation Sampling (COSMOS) was used as training labels and evaluation reference. DIAM-CNN were compared with two conventional model-based methods (Morphology Enabled Dipole Inversion or MEDI; improved sparse linear equation and least squares or iLSQR) and one deep learning method (QSMnet). High-frequency error norm (HFEN), peak signal-to-noise-ratio (PSNR), normalized root mean squared error (NRMSE), and structure similarity index (SSIM) were reported for quantitative comparisons. Experiments on healthy volunteers demonstrated that the DIAM-CNN results had superior image quality to those of MEDI, iLSQR or QSMnet results. Experiment on data with simulated hemorrhagic lesion demonstrated that DIAM-CNN produced less shadow artifacts around the bleeding lesion than the compared methods. This study demonstrates that the incorporation of dipole-related knowledge into the network construction has potential to improve deep learning-based QSM reconstruction.