Hello, our lab has recently started acquiring multi-band FMRI data, which will typically be in the range of 2.5mm isotropic and a TR of 1s. melodic can handle small datasets like this (ie, 4-5minutes of scan time) but when i concatenate an entire hour long session, it fails to converge (its been running for 4 days now), while a 5 minute scan converges in ~10 minutes.

anyone know a way around this? i like to concatenate my datasets because i feel it gives the algorithm more info to work with, but this is just my general feeling. 

also, on a related note, i find that when i run melodic separately on several 5 minute datasets, it will estimate something like 120 components, but when i concatenate several datasets (to make say, 25 minutes of continuous FMRI data) it estimates roughly 5x the number of components? this seems strange to me, because obviously the number of components hasn't changed in the brain (besides maybe a few more motion artifacts)...