Meeting Banner
Abstract #3037

Robust Perfusion Maps in Arterial Spin Labeling by Means of M-Estimators

Camille Maumet1, Pierre Maurel1, Jean-Christophe Ferr1, 2, Christian Barillot1

1Inria, IRISA, RENNES, Brittany, France; 2Department of Neuroradiology, CHU Rennes, Rennes, Brittany, France

In Arterial Spin Labeling (ASL), the perfusion signal is usually extracted by averaging the volumes acquired over several repetitions. Unfortunately, the presence of artefacts is a well-known source of outliers and can drastically alter the perfusion map obtained by averaging. In this paper, we propose to compute ASL perfusion maps using Huber's M-estimator, a robust statistical function that is not overly impacted by outliers. This method is compared to an empirical approach, previously introduced in the literature, based on z-score thresholding. Overall, Huber's M-estimator is more robust than z-thresholding. Both robust approaches outperform the sample mean in the presence of outliers.

Abstract PDF Presentation Video

Keywords

abnormal acquisition added advise annals another appears arterial assessed audience authors automatic averaging avoid behavior behind better blood brain category cause cerebral clearly clinical compute computed computer computing contrast control correlation corrupted corruption datasets date deal decimated detrimental deviation diagnosed displayed distribution doctors dynamic empirical employ ensures estimator estimators example experiment feasible final flow frequent function general ground grows idea ignore illustrated impacted improved include index instead intensities introduced known label labeling linear location long mainly maps masking mathematical motion noise often outliers outperform overall overly pairs pairwise paper particular patient patients peaks performances perfusion populations post presence processing produce pronounced proposed quality randomly real rejection related repeated repetitions robust sample score sensitive series several simulated slice solution specifically spin squared starting statistical statistics still strong studied subject subset subtracting sudden suffering sufficient suggested superiority susceptibility target term thanks theoretical though threshold thresholding truth tumors unbiased uniform validated various volume volumes