Using Deep Learning to Simultaneously Reduce Noise and Motion Artifacts in Brain MR Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Motion artifacts (MA) and noise degrade the quality of brain Magnetic Resonance Imaging (MRI) scans.
• These artifacts can obscure important anatomical details, potentially leading to misdiagnosis.
• Deep learning offers a promising avenue for automated artifact and noise reduction in medical imaging.

Purpose of the Study:

• To develop and validate a deep learning-based method for reducing motion artifacts and noise in brain MRI.
• To enhance the clinical utility of MRI by improving image quality across different sequences (T1W, T2W, FLAIR).

Main Methods:

• A deep learning model was trained using simulated brain MRI images (T1W, T2W, FLAIR) with varying levels of noise and MA.
• Separate models were developed for each MRI sequence.
• Model performance was evaluated using quantitative metrics (SSIM, PSNR) and qualitative visual assessment by radio technologists.

Main Results:

• The deep learning models achieved high performance in removing noise and MA, with SSIMout >0.95 and PSNRout averaging 72 dB.
• Significant improvement in image quality was demonstrated compared to input images (IMPRs and IMPRp).
• Visual evaluation confirmed the effectiveness of the method in reducing artifacts and noise.

Conclusions:

• The proposed deep learning approach effectively removes motion artifacts and noise from brain MRI, irrespective of imaging direction or artifact orientation.
• Utilizing simulated data for training enabled the generation of robust models.
• This technique holds significant potential for improving the diagnostic accuracy and clinical applicability of brain MRI.