Deep Learning-Based Super-Resolution Reconstruction on Undersampled Brain Diffusion-Weighted MRI for Infarction...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Diffusion-weighted imaging (DWI) is vital for detecting infarction stroke.
• Limited spatial resolution of conventional DWI hinders accurate visualization of stroke lesions.
• Super-resolution reconstruction techniques are being explored to overcome these limitations.

Purpose of the Study:

• To evaluate the image quality of deep learning (DL)-based super-resolution reconstruction for brain DWI in stroke detection.
• To assess the diagnostic confidence associated with DL-based super-resolution DWI compared to conventional methods.
• To compare the performance of DL-based super-resolution reconstruction (DWI_DL) against compressed sensing reconstruction (DWI_CS).

Main Methods:

• Retrospective analysis of 114 brain DWI examinations.
• Reconstruction of DWI images using two schemes: DL-based super-resolution (DWI_DL) and conventional compressed sensing (DWI_CS).
• Qualitative assessment of image quality, lesion conspicuity, and diagnostic confidence; quantitative assessment of SNR, CNR, ADC, and edge rise distance.

Main Results:

• DWI_DL demonstrated significantly improved overall image quality, lesion conspicuity, and diagnostic confidence for infarction stroke lesions (<1.5 cm) compared to DWI_CS (P < .001).
• Quantitative analysis showed a reduced edge rise distance with DWI_DL (P < .001).
• No significant differences were observed in SNR, CNR, and ADC values between the two methods (P > .05).

Conclusions:

• DL-based super-resolution reconstruction offers superior image quality for brain DWI in infarction stroke detection.
• This technique enhances diagnostic confidence compared to conventional compressed sensing reconstruction.
• DL-based super-resolution is a feasible and effective method for improving stroke imaging.