Table 1.

Qualitative Differences between the Proposed CNMF Method, PCA/ICA, and Plain NMF

Method	PCA/ICAa	Plain NMFb	Constrained NMFc
Features
Handles spatial overlap well	No	Yes	Yes
Eliminates “cross-talk”	No	Medium	Yes
Robustness in low SNR	Medium	Medium	Yes
Produces localized estimates	Yes	No	Yes
Unsupervised (after initial parameter setting)	Yes	No	Yes
Deconvolves neural activity	No	No	Yes
Models background activity	No	Yes	Yes
Enables merging	No	Manual	Automated

^aMukamel et al. (2009).

^bMaruyama et al. (2014).

^cThis paper.