A spatially adaptive regularization based three-dimensional reconstruction network for quantitative susceptibility...

Area of Science:

• Medical Imaging
• Biophysics
• Computational Neuroscience

Background:

• Quantitative susceptibility mapping (QSM) is an advanced MRI technique for non-invasive tissue characterization.
• Reconstructing QSM involves solving an ill-posed inverse problem, which is challenging for deep learning models due to differing physical units (Hz to ppm).

Purpose of the Study:

• To develop a novel deep learning framework, SAQSM, for accurate QSM reconstruction.
• To address the challenges of cross-modality quantitative mapping and improve feature detection in QSM.

Main Methods:

• Proposed SAQSM, a 3D reconstruction network with spatially adaptive regularization modules.
• Incorporated dynamic perceptual initialization in the network encoding for enhanced feature detection.
• Utilized field and magnitude data for adaptive adjustment of feature maps.

Main Results:

• SAQSM demonstrated more accurate QSM reconstruction with reduced susceptibility artifacts in healthy volunteers, hemorrhage patients, and phantom data.
• The network showed good stability and generalization capabilities, even in areas with severe lesions.
• Experimental results confirmed the effectiveness of the spatially adaptive modules in correcting information loss.

Conclusions:

• The SAQSM framework offers a significant advancement in QSM reconstruction accuracy and artifact reduction.
• This approach provides a valuable paradigm for quantitative mapping and multimodal reconstruction in medical imaging.
• SAQSM shows promise for improved characterization of tissue composition and microstructure.