SHARQnet - Sophisticated harmonic artifact reduction in quantitative susceptibility mapping using a deep...

Area of Science:

• Medical Imaging
• Neuroscience
• Computational Physics

Background:

• Quantitative susceptibility mapping (QSM) detects pathological changes in neurological and hepatic diseases.
• QSM processing involves complex, iterative steps like background field removal and field-to-source inversion.
• Current QSM methods are computationally intensive and require parameter tuning, hindering clinical use.

Purpose of the Study:

• To develop an efficient and reliable method for background field removal in QSM.
• To extend a deep learning approach for QSM inversion to address background field artifacts.
• To facilitate the clinical application of QSM by improving processing efficiency and accuracy.

Main Methods:

• A deep convolutional neural network (SHARQnet) was developed for background field removal in QSM.
• SHARQnet was trained on simulated background fields and validated on 3T and 7T brain MRI datasets.
• The method leverages a feed-forward multiplication approach, avoiding iterative optimization and parameter selection.

Main Results:

• SHARQnet demonstrated superior performance compared to existing background field removal techniques.
• The method showed excellent generalization across diverse 3T and 7T brain imaging data.
• No parameter adjustments were needed for SHARQnet, highlighting its robustness.

Conclusions:

• Deep learning can effectively learn and correct artifacts caused by background fields in QSM.
• SHARQnet offers an efficient, reliable, and parameter-free solution for QSM background field removal.
• This advancement significantly contributes to the clinical translation of QSM for disease diagnosis.