Evaluation of deep learning reconstruction on diffusion-weighted imaging quality and apparent diffusion coefficient...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging Technology
• Image Reconstruction Algorithms

Background:

• Diffusion-weighted imaging (DWI) is crucial for assessing tissue microstructure.
• Accurate apparent diffusion coefficient (ADC) quantification is vital for diagnostic interpretation.
• Deep learning reconstruction (DLR) offers potential improvements in MRI image quality.

Purpose of the Study:

• To evaluate the impact of deep learning reconstruction (DLR) on diffusion-weighted image (DWI) quality.
• To assess DLR's influence on the accuracy, precision, and repeatability of apparent diffusion coefficient (ADC) measurements.
• To investigate DLR performance across different b-values and slice thicknesses in MRI.

Main Methods:

• An ice-water phantom with known diffusion properties was imaged using a 3T MRI scanner.
• Diffusion-weighted images (DWIs) were acquired at various b-values (0-4000 s/mm²) and slice thicknesses (1.5, 3.0 mm).
• Images were reconstructed with and without DLR; ADC maps were generated and analyzed for SNR, accuracy, precision, and within-subject coefficient of variation (wCV).

Main Results:

• DLR significantly improved signal-to-noise ratio (SNR) for b-values up to 2000 s/mm², with diminished effect at 4000 s/mm².
• No significant difference in ADC values was observed between DLR and conventional reconstruction.
• DLR enhanced the precision and repeatability (wCV) of ADC measurements, particularly at higher b-values and thinner slices.

Conclusions:

• Deep learning reconstruction (DLR) enhances SNR and measurement precision in MRI-based ADC quantification.
• DLR does not improve the accuracy of ADC values, which remained lower than true values in specific configurations.
• DLR shows promise for improving the reliability of ADC measurements in diffusion MRI studies.