Feasibility study of super-resolution deep learning-based reconstruction using k-space data in brain...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Diffusion-weighted imaging (DWI) is crucial for brain MRI, providing insights into tissue microstructure.
• Image quality and accurate quantitation of apparent diffusion coefficient (ADC) are vital for reliable diagnostic interpretation.
• Super-resolution deep learning-based reconstruction (SR-DLR) is an emerging technique that processes k-space data to potentially improve image resolution and quality.

Purpose of the Study:

• To evaluate the impact of SR-DLR on image quality in brain DWI.
• To assess the effect of SR-DLR on ADC quantitation in brain DWI.
• To determine if SR-DLR improves signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) in brain MRI.

Main Methods:

• Retrospective analysis of 34 patients undergoing 3T brain MRI with DWI.
• Comparison of DWI images reconstructed with and without SR-DLR (684x684 vs. 228x228 matrix).
• Quantitative assessment of SNR, CNR, and full width at half maximum (FWHM), alongside qualitative radiologist assessments.

Main Results:

• Images reconstructed with SR-DLR demonstrated significantly higher SNRs and CNRs compared to those without (p < 0.001).
• No statistically significant differences were observed in ADC values for white matter (p=0.945) and grey matter (p=0.235) between the two reconstruction methods.
• The full width at half maximum (FWHM) was significantly lower in images reconstructed without SR-DLR (p < 0.001).

Conclusions:

• SR-DLR effectively enhances image quality in brain DWI MRI by improving SNR and CNR.
• SR-DLR does not significantly affect the accuracy of ADC quantitation, preserving its diagnostic value.
• This deep learning-based reconstruction technique shows promise for improving diagnostic confidence in brain MRI.