Super-resolution deep learning reconstruction for brain fluid-attenuated inversion recovery: image quality and white...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Neuroimaging

Background:

• Two-dimensional (2D) fluid-attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI) is crucial for visualizing white matter hyperintensities (WMH).
• Image quality in 2D FLAIR can be limited by noise and resolution, impacting automated WMH volumetry.
• Deep learning reconstruction (DLR) techniques offer potential for image quality improvement.

Purpose of the Study:

• To assess if super-resolution deep learning reconstruction (SR-DLR) enhances 2D brain FLAIR image quality.
• To compare SR-DLR against Gaussian-filtered reconstruction (GA) and denoising DLR (dDLR).
• To evaluate SR-DLR's ability to preserve automated white matter hyperintensity (WMH) volumetry.

Main Methods:

• Thirty-six healthy volunteers underwent 3T axial 2D FLAIR MRI.
• Images were reconstructed using GA, dDLR, and SR-DLR (twofold in-plane upscaling).
• Quantitative metrics (noise, SNR, CNR, sharpness) and qualitative image quality scores were assessed. WMH volumes were measured using a U-Net model and compared across reconstructions, including a high-resolution reference.

Main Results:

• SR-DLR significantly outperformed GA and dDLR, showing lower noise and higher SNR, CNR, and sharpness (p < 0.001).
• Qualitative assessments revealed superior noise reduction, sharpness, and overall image quality with SR-DLR (p < 0.001).
• Mean WMH volumes were comparable across all reconstructions, with SR-DLR volumes closely matching the high-resolution reference standard.

Conclusions:

• SR-DLR substantially improves 3T 2D brain FLAIR image quality by enhancing key metrics.
• SR-DLR effectively preserves WMH volumetry, yielding results consistent with high-resolution acquisitions.
• SR-DLR represents a promising advancement for neuroimaging analysis, particularly for WMH assessment.