Deep grey matter quantitative susceptibility mapping from small spatial coverages using deep learning.

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Quantitative Susceptibility Mapping (QSM) brain-iron studies often focus on deep grey matter (DGM).
• Full brain coverage QSM is standard, but reduced coverage could shorten scan times or improve resolution.
• Reduced coverage may lead to significant DGM susceptibility underestimation.

Purpose of the Study:

• To assess the accuracy of a deep learning-based QSM method (xQSM) on reduced brain coverages.
• To compare xQSM with conventional dipole inversion methods for DGM susceptibility mapping.
• To evaluate the impact of reduced spatial coverage on QSM accuracy.

Main Methods:

• Investigated a deep learning-based QSM method, xQSM.
• Compared xQSM with two conventional dipole inversion methods.
• Used simulated and in vivo data from 4 healthy subjects at 3T with varying spatial coverages (100%–400% of DGM size).

Main Results:

• xQSM demonstrated the lowest DGM contrast loss and smallest susceptibility variation across coverages.
• xQSM improved DGM susceptibility underestimation by over 20% in simulations with small coverages.
• In vivo scans achieved <5% DGM susceptibility error with 48mm slabs using xQSM, vs. 112mm for conventional methods.
• Background field removal worsened with reduced coverage, impacting dipole inversion.

Conclusions:

• The deep learning-based xQSM method significantly enhances DGM QSM accuracy from reduced spatial coverages.
• xQSM offers a substantial reduction in DGM QSM acquisition time compared to conventional algorithms.
• This advancement is crucial for efficient and accurate brain-iron quantification in DGM studies.