Impact of Distortion on Local Radiomic Analysis of Quadriceps Based on Quantitative Magnetic Resonance Imaging Data

Abstract

This study investigates the impact of the MRI distortion that appears between 3D T1 Dixon Water-only images and both spoiled gradient echo and multi-echos T2 weighted spin-echo images when sequentially acquired. Recent studies focusing on radiomic features locally computed on muscle heads or bone marrow segmentations require precise corrections of the bias field and the distortion. Our results suggest that classically used rigid registration are not optimal for such fine study and that deformable registration should be preferred to limit significant error in radiomic feature extraction. However, from our experiments on our data, no significant change in radiomic statistic is observed whatever segmentation correction approach was applied. This indicates that radiomic features are not sensitive to segmentation refinement when considering large 3D regions.