E. Kousi, M. Rata, D.-M. Koh, C. Messiou, N. Tunariu, G. Hopkinson, E. Evans, O. Darwish, E. Weiland, J. Winfield
Objectives: To evaluate the impact of an accelerated whole-body diffusion-weighted MRI (WB-DWI) protocol using deep learning (DL) reconstruction on image quality and apparent diffusion coefficient (ADC) quantification in patients with metastatic prostate cancer. Methods: This single-centre prospective study involved two patient cohorts undergoing standard-of-care WB-MRI at 1.5T. The DL WB-DWI acquisition used fewer signal averages and higher parallel imaging acceleration, reconstructed with a research application DL package based on a variational network. In Cohort 1 (n = 10), qualitative assessment of image quality was performed across four anatomical stations using a 4-point Likert scale by two radiologists, blinded to the DWI protocol. In Cohort 2 (n = 20), ADC values were evaluated in hypercellular focal bone metastases (as determined by a radiologist) and compared between the standard and accelerated DL WB-DWI protocols. Image quality scores were analysed using Wilcoxon signed-rank tests (Bonferroni-corrected), and ADC agreement was evaluated via Bland-Altman analysis. Results: Accelerated DL WB-DWI reduced WB-DWI acquisition time by 37% without compromising image quality. Radiologists consistently rated both protocols as good-to-excellent across all image quality metrics. Median ADC estimates in lesions showed no significant difference between protocols; interquartile range of ADC estimates also showed no significant difference. Conclusions: Accelerated DL WB-DWI enables 37% scan time reduction without degrading image quality or affecting ADC estimates of lesions in patients with metastatic prostate cancer. Advances in knowledge: DL-based reconstruction enables faster WB-DWI acquisition without compromising image quality or ADC quantification, supporting clinical implementation in metastatic prostate cancer staging and therapy assessment.