Deep Learning Reconstruction of Diffusion-weighted MRI Improves Image Quality for Prostatic Imaging.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Deep learning reconstruction (DLR) is a novel technique with the potential to improve MRI image quality.
• The specific impact of DLR on prostate diffusion-weighted imaging (DWI), particularly at high b values, remains underexplored.

Purpose of the Study:

• To evaluate the effectiveness of DLR in enhancing the image quality of prostate diffusion-weighted MRI.
• To assess image quality improvements at high b values (1000-5000 s/mm²).

Main Methods:

• Retrospective analysis of prostate DWI scans (b=0, 1000, 3000, 5000 s/mm²) from 60 prostate cancer patients.
• Image reconstruction with and without DLR.
• Quantitative assessment using signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR).
• Qualitative assessment using a 5-point visual scoring system.
• Comparison of apparent diffusion coefficients (ADCs) between DLR and non-DLR images.

Main Results:

• DLR significantly increased SNR and CNR compared to conventional reconstruction (P < .001).
• Qualitative image quality scores were significantly higher with DLR across all high b values (P ≤ .002).
• Mean SNR for DWI1000 with DLR was 38.7 ± 0.6 vs 17.8 ± 0.6 without DLR.
• Mean CNR for DWI1000 with DLR was 18.4 ± 5.6 vs 7.4 ± 5.6 without DLR.
• ADCs derived with and without DLR showed no substantial differences (P > .99).

Conclusions:

• Deep learning reconstruction demonstrably improves image quality in prostate cancer diffusion-weighted MRI.
• DLR enhances SNR, CNR, and qualitative scores at high b values.
• The technique does not alter ADC quantitation, preserving diagnostic information.