Data-driven synthetic MRI FLAIR artifact correction via deep neural network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Synthetic MRI FLAIR imaging can produce artifacts, limiting diagnostic capabilities.
• Artifacts stem from partial volume effects and flow, which are challenging to correct analytically.
• A deep learning (DL) method was developed to address these synthetic FLAIR artifacts without analytical modeling.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) based method for correcting artifacts in synthetic FLAIR MRI.
• To improve the diagnostic quality of synthetic FLAIR images by reducing image artifacts.

Main Methods:

• A retrospective study involving 80 subjects was conducted.
• A deep learning model was trained and validated on synthetic FLAIR data acquired using a multiple-dynamic multiple-echo (MDME) sequence at 3T.
• Quantitative metrics (NRMSE, SSIM) and qualitative assessment by neuroradiologists were used to evaluate image quality and artifact reduction.

Main Results:

• Deep learning correction significantly improved NRMSE from 4.2% to 2.9% (P<0.0001) and SSIM from 0.85 to 0.93 (P<0.0001).
• NRMSE reductions were observed across white matter, gray matter, and CSF regions.
• Qualitative analysis showed improved overall image quality and reduced artifacts in critical brain areas.

Conclusions:

• The developed deep learning approach effectively corrects artifacts in synthetic FLAIR MRI.
• This DL-based method offers a promising solution for enhancing the diagnostic utility of synthetic FLAIR imaging.
• The findings suggest a potential advancement in MRI techniques for neurological assessments.