Fractal dimension analysis of resting state functional networks in schizophrenia from EEG signals

Abstract

Fractal dimension (FD) has been revealed as a very useful tool in analyzing the changes in brain dynamics present in many neurological disorders. The fractal dimension index (FDI) is a measure of the spatiotemporal complexity of brain activations extracted from EEG signals induced by transcranial magnetic stimulation. In this study, we assess whether the FDI methodology can be also useful for analyzing resting state EEG signals, by characterizing the brain dynamic changes in different functional networks affected by schizophrenia, a mental disorder associated with dysfunction in the information flow dynamics in the spontaneous brain networks. We analyzed 31 resting-state EEG records of 150 s belonging to 20 healthy subjects (HC group) and 11 schizophrenia patients (SCZ group). Brain activations at each time sample were established by a thresholding process applied on the 15,002 sources modeled from the EEG signal. FDI was then computed individually in each resting-state functional network, averaging all the FDI values obtained using a sliding window of 1 s in the epoch. Compared to the HC group, significant lower values of FDI were obtained in the SCZ group for the auditory network (p < 0.05), the dorsal attention network (p < 0.05), and the salience network (p < 0.05). We found strong negative correlations (p < 0.01) between psychopathological scores and FDI in all resting-state networks analyzed, except the visual network. A receiver operating characteristic curve analysis also revealed that the FDI of the salience network performed very well as a potential feature for classifiers of schizophrenia, obtaining an area under curve value of 0.83. These results suggest that FDI is a promising method for assessing the complexity of the brain dynamics in different regions of interest, and from long resting-state EEG signals. Regarding the specific changes associated with schizophrenia in the dynamics of the spontaneous brain networks, FDI distinguished between patients and healthy subjects, and correlated to clinical variables.

Keywords: EEG; fractal dimension; functional network; resting state; schizophrenia.

Figure 1

(A) Sources activity for 1 s. (B) Representation in 3D of the sources activity at two different time samples (0.06 and 0.73 s respectively). (C) 3D representation after binarization. (D) Point clouds defined by sources after binarization (cortical activations). (E) Cortical activations in each functional network (shown for the time sample of 0.73 s).

Figure 2

Cortical parcellation of the resting-state functional networks analyzed.

Figure 3

3DFD evolution for 1 s comparing a healthy subject and a patient. HFD, Higuchi fractal dimension of the curve described by the 3DFD values.

Figure 4

Boxplots with differences in fractal dimension index (FDI) between groups. The FDI values in each group are the average of the 150 FDI values computed for each subject in the corresponding network. Values of p for the Quade’s ANCOVA test controlling for age, gender, and education (^*p < 0.05). Values of p were corrected for multiple comparisons with the Bonferroni post hoc method. No significant differences were found between groups for the whole Brain nor the DMN and VIS networks.

Figure 5

FDI evolution in 150 s for HC and SCZ groups. Each FDI point is the average of the FDI values of all the subjects of the group for the corresponding second.

Figure 6

Significant Spearman correlations between cognitive and psychopathological scores and FDI in the SCZ group. Values of p were corrected for multiple comparisons with the Bonferroni post hoc method. Only significant correlations (p < 0.05) are shown. VLi, Immediate verbal learning; VLd, Delayed verbal learning; VF, Verbal fluency; WM, Working memory; PS: Processing speed; PANSS, Positive and negative syndrome scale; PANSS-P, PANSS—positive subscale; PANSS-N, PANSS—negative subscale; and PANSS-G, PANSS—general subscale.

Figure 7

ROC curve analysis assessing the performance of FDI as a classifier for schizophrenia. The AUC value for each FDI measure is shown between brackets.