PMC

Grey matter and vCSF templates used for spatial correlation with the thresholded Z-map of the independent component.

Flowchart of the optimal features/components selection, explaining how to identify the final optimal components from the raw data.

The extracted joint component with the largest J-divergence from feature combination of [SM,SM,SBP,AODT]. The activations of the 3 features are transferred to Z score and thresholded at ∣Z∣>2.5.

Top 14 ICs with a J-divergence larger than 1.5 among all extracted ICs of 15 combinations. Their feature combination names are displayed on the bar. Component’s number of the top 3 ICs is also shown in bracket.

J-divergence of the optimal components selected by 2 criteria: the smallest p value (blue bar) and the largest J-divergence (green bar), for 15 feature combinations. The modulus of the above two J-divergence values are plotted as red bar, which is treated as the group discriminating power of each feature combination.

The most remarkable characteristics for 8 components decomposed from an fMRI dataset with a single Sternberg_probe task. Note that the 3rd IC is sparsely-distributed noise, the 5th IC is a vCSF artifact, the 7th IC manifests as movement-related artifact; and others appear to be ICs of interest. The descriptive measures used to classify the components are listed below in a table. All the 3 artifacts show zero (green background) in either of the criteria listed in the first two rows.

(a), (b) are the spatial maps of the top 3 optimal components, which are converted to Z-scores and thresholded at ∣Z∣>2.5; (c) shows the overlapping regions of the 4 features with their original spatial map values, these activated regions are important for group discrimination and may serve as biomarkers of schizophrenia patients; (d) displays the difference between the back-reconstructed sources (HC-SZ) on the combined highlighted regions of the top three optimal ICs in (c), the regions where HC>SZ in ∣Z∣ score are shown in orange, otherwise are shown in blue.