Integrated deep learning reconstruction combining phase correction, denoising, and super-resolution for prostate diffusion-weighted imaging
Related Experiment Video
Updated: Nov 9, 2026
Three-Dimensional Phase Resolved Functional Lung Magnetic Resonance Imaging
Published on: June 21, 2024
Noriko Nishioka1, Maki Onodera1, Daisuke Kato2
1Department of Diagnostic and Interventional Radiology, Hokkaido University Hospital, N14 W5, Kita-ku, Sapporo 060-8648, Japan; Department of Diagnostic Imaging, Faculty of Medicine and Graduate School of Medicine, Hokkaido University, N15 W7, Kita-ku, Sapporo 060-8638, Japan.
Objective:
To evaluate integrated deep learning reconstruction (Integrated DLR), which combines phase correction, denoising, and super-resolution, for prostate diffusion-weighted imaging (DWI) by comparing it with compressed sensing (CS) and deep learning-based denoising reconstruction (Denoising DLR).
Methods:
This single-center, retrospective study included 47 patients scanned on a single-vendor 3-T system. Three DWI datasets were reconstructed from identical raw data using CS, Denoising DLR, and Integrated DLR. Two radiologists rated overall image quality, noise, artifact, and prostate delineation. Apparent signal-to-noise ratio (SNR) was assessed in all patients. In 21 patients with pathologically confirmed prostate cancer, we assessed lesion conspicuity, lesion localization/anatomical relationship, Prostate Imaging Reporting and Data System (PI-RADS) version 2.1 DWI scores, apparent lesion contrast-to-noise ratio (CNR), and apparent diffusion coefficient (ADC). Comparisons were performed using Friedman tests followed by Bonferroni-corrected Wilcoxon signed-rank tests.
Results:
Scores for overall image quality, noise, and prostate delineation were highest with Integrated DLR, followed by Denoising DLR and CS, for both readers (all p < 0.001); apparent SNR showed the same pattern (p < 0.001). Artifact scores did not differ among methods. Lesion conspicuity differed among methods for Reader 2 (p = 0.044) but not for Reader 1 (p = 0.199), whereas Integrated DLR yielded the highest lesion localization/anatomical relationship scores for both readers (both p < 0.001). Integrated DLR also yielded the highest apparent CNR and lesion ADC (both p < 0.001), while PI-RADS DWI scores were unchanged.
Conclusion:
Integrated DLR improved prostate DWI image quality and lesion visibility compared with CS and Denoising DLR, potentially facilitating visual assessment of prostate DWI.
