Project OUTLIERS
(1) Adapt DVARS from exercices in findoutlie/metrics.py => DONE (2) Fill in findoutlie/detectors.py for IQR (interquartile range) (but not only) - It is the region between the 75th and 25th percentile (meaning 75-25=50% of the data) - TO DO : lier le script 'metrics' (calcul DVARS) avec 'detectors' (calcul IQR) (3) Fill in findoutlie/outfind.py - Dans detect_outliers : il faut que l'on calcule une métrique (example : métrique 1 = FD de l'image pour chaque pas de temps = une liste de valeur, injectée dans une des fonctions de detectors) ; important: on appelle à fois les fonctions de '/metrics' et à la fois de '/detectors' - Dans detect_outliers : (au-dessus) on va pouvoir le faire pour d'autres métriques (4) Test on one subject (5) Test on the 9 other subjects (6) Add other image quality metrics: (cf. detect_outliers) ARTIFACTS MEASURE: - Framewise displacement: controls for movement => was the outlier detected because of a sudden move? FORMULA: FDi =∣Δxi∣+∣Δyi∣+∣Δzi∣+∣Δαi∣+∣Δβi∣+∣Δγi∣ (from Power et al., 2012) EXPLANATION: expresses instantaneous head-motion; i is the timepoint; x, y, z are the translational realignment parameters (RPs); α, β, γ are the rotational RPs TEMPORAL INFORMATION: - tSNR: gives information on the temporal evolution of the signal to noise ratio => was the outlier detected because of a disturbance in signal? FORMULA: tSNR = (Mean timecourse signal) / (SD timecourse noise) (from Krüger et al., 2001) EXPLANATION: Mean timecourse signal, is the average BOLD signal across time and SD timecourse noise, the SD of the corresponding voxel timecourse; higher values are better. (7) Decision criteria: Based on the combination of these 3 parameters: - DVARS - FD - tSNR And the combination of these 2 detectors: - IQR - ?
=> OBJECTIF : 3 métriques (DVARS, FD, tSNR?) + 2 detectors (IQR, ?)