Qualitative and quantitative assessment of accelerated liver DWI using deep learning reconstruction in oncologic patients

Motivation: Deep learning (DL) reconstructions can improve image quality and/or reduce acquisition time in diffusion-weighted imaging (DWI). Goal(s): This study aims to assess, quantitatively and qualitatively, DL-accelerated DWI in 50 patients with colorectal cancer undergoing liver examinations. Approach: Three DWI series are compared: a moderately-accelerated DL-DWI, a corresponding standard reconstruction and a highly-accelerated DL-DWI. Results: The moderately-accelerated DL reconstruction method provides better image quality than a standard reconstruction. Its ADC estimates in liver, spleen and liver metastases are slightly higher than ADC estimates from the standard reconstruction. Impact: This study evaluated DL-accelerated DWI in 50 patients undergoing liver examinations by comparing three DWI series. The moderately-accelerated acquisition with DL reconstruction provided better image quality versus the standard reconstruction; its ADC was slightly higher than the standard-based ADC.