Robustness of MRI radiomics features in abdomen: impact of deep learning reconstruction and accelerated acquisition

Motivation: The deep learning reconstruction and accelerated acquisition using HASTE technique allow scans with comparable image quality and diagnostic performance. However, its impact on MRI-radiomics is unclear. Goal(s): This study aims to investigate the impact of deep learning reconstruction and accelerated acquisition on the radiomics robustness in abdominal scans using a HASTE sequence. Approach: Seventeen volunteers were prospectively scanned using deep learning reconstruction and accelerated acquisition protocols, and their influence on radiomics were assessed. Results: Deep learning reconstruction and accelerated acquisition led to a poor reproducibility of radiomics, but more than a half of them varied within an acceptable range. Impact: Deep learning reconstruction and accelerated acquisition significantly impacts on radiomic features, necessitating caution to the generalizability when performing radiomic analysis using images from different reconstruction algorithms and acquisition protocols.