Icasso: software for investigating the reliability of ICA estimates by clustering and visualization

Abstract

Abstract. A major problem in application of independent component analysis (ICA) is that the reliability of the estimated independent components is not known. Firstly, the finite sample size induces statistical errors in the estimation. Secondly, as real data never exactly follows the ICA model, the contrast function used in the estimation may have many local minima which are all equally good, or the practical algorithm may not always perform properly, for example getting stuck in local minima with strongly suboptimal values of the contrast function. We present an explorative visualization method for investigating the relations between estimates from FastICA. The algorithmic and statistical reliability is investigated by running the algorithm many times with different initial values or with differently bootstrapped data sets, respectively. Resulting estimates are compared by visualizing their clustering according to a suitable similarity measure. Reliable estimates correspond to tight clusters, and unreliable ones to points which do not belong to any such cluster. We have developed a software package called Icasso to implement these operations. We also present results of this method when applying Icasso on biomedical data.

Demonstration of Brain Tumor-Induced Neurovascular Uncoupling in Resting-State fMRI at Ultrahigh Field

“…7T rsfMRI data were corrected for slice time and spatial realignment, followed by nuisance regression using anatomical component-based noise correction (aCompCor) per Behzadi and associates (2007), scrubbing of outlier volumes detected by the ART repair toolbox (Mazaika et al, 2009), band-pass filtering at 0.1-0.01 Hz, and smoothing with a 6-mm FWHM Gaussian kernel. Independent component analysis (ICA) with 35 target components was subsequently performed using the GIFT toolbox (Correa et al, 2005) utilizing the Infomax algorithm with ICASSO to determine the reliability of ICA results (Himberg and Hyvarinen, 2003). Three experienced raters ( J.J.P., S.A., and H.I.S.)…”

Performance of blind source separation algorithms for fMRI analysis using a group ICA method

“…Consistency analysis using ICASSO-ICASSO [31] presents a visualization technique that enables analysis of the differences among estimates obtained from multiple runs due to the iterative nature of FastICA. In [32], ICASSO was used to analyze ICA of fMRI data using FastICA.…”

Abnormal connectivity between attentional, language and auditory networks in schizophrenia

“…Stability of the components, i.e. investigating whether a component has the tendency to split or merge with another component (Rosazza et al, 2011), was validated by running the ICASSO toolbox implemented in GIFT using twenty iterations with both random iterations and bootstrapping (Himberg and Hyvärinen, 2003;Himberg et al, 2004).…”

“…First of all, it is difficult to determine the correct number of components, and to prevent the splitting or merging of components (Fox and Raichle, 2007;Smith et al, 2009). We tried to reduce this risk by estimating the optimal number of components (Li et al, 2006) and running the ICASSO toolbox (Himberg and Hyvärinen, 2003;Himberg et al, 2004), which showed good stability. Moreover, we also ran ICA with more or less components (20, 25, 35 and 40), and this did not change our findings (data not shown), which is in agreement with Rosazza et al (2011).…”