1. Introduction
A dynamic system is characterized by its evolving state over time, which can be fully described by a set of state variables at any given moment. These models allow for the analysis and prediction of system behavior under different inputs or conditions. Studying and understanding dynamic systems provides valuable insights into their behavior, allowing for predictions and even the control or optimization of their performance in various applications.
Dynamical systems can exhibit different dynamic properties, such as stability, periodicity, or chaos. Chaotic systems exhibit sensitivity to initial conditions, known as the butterfly effect, where small changes in initial conditions can lead to significant differences in the system’s behavior over time [
1]. This is a fundamental characteristic shared by various physical systems, including neural networks, with the factors contributing to this sensitivity yet to be fully understood.
The dynamics of the brain show complex nonlinear characteristics, and their underlying mechanisms are still not well understood. While concrete evidence of chaos in cerebral dynamics, from a mathematical standpoint, has primarily been observed at the level of axons, individual cells, and paired cells, findings suggest that brain signals may exhibit chaotic patterns across all levels of their hierarchy [
2,
3,
4].
Neural information processing is a challenging topic that requires an understanding of the intricate mechanisms underlying neuronal activity. One key aspect of this process is the nonlinear nature of neuronal dynamics, leading to complex phenomena such as chaos, synchronization, and bifurcations. These phenomena have significant implications for neural coding and computation. Analyzing the dynamics of nonlinear systems is not a straightforward task, especially when there are no available differential equations to model the system under study, as is the case with the brain.
One method of capturing the brain’s activity over time is through temporal recordings of its electrical activity via intercellular local field potentials (LFPs). These recordings reflect the extracellular electrical activity of nearby neurons and are obtained using deep electrodes inserted into the brain, resulting in intracranial electroencephalography (iEEG) signals, which are temporal voltage series.
The application of information theory to analyze iEEG signals enhances the understanding of brain activity by providing quantitative and objective measures. It enables the extraction of temporal characteristics and quantification of the information contained in the brain signals. Furthermore, information theory allows investigation of the complexity and organization of brain signals, revealing underlying patterns and organizational structures.
Utilizing ordinal patterns to compute measures based on permutation entropy enables the extraction of causal characteristics from signals and the assessment of nonlinear dynamics within these systems. The approach to computing the probability distribution function (PDF) from ordinal patterns, as proposed by Bandt and Pompe (BP), is widely embraced for electroencephalography (EEG) analysis, including iEEG, and has demonstrated superior results compared to conventional analyses [
5].
Extensive studies have been conducted on non-intracranial EEG signals, such as the analysis of time series in epilepsy EEG [
6,
7], the distinction between brain death and coma [
8], and the study of states of consciousness with the use of anesthetics [
9], sleep stages [
5,
10,
11,
12], and signal discrimination [
13,
14,
15]. In iEEG, this method has been applied to epilepsy for seizure prediction [
16], detecting the epileptogenic focus [
17], identifying preictal markers [
18], extracting different characteristics for identifying the epileptogenic focus [
17], and differentiating epileptic signals in an unsupervised manner [
19,
20,
21]. These findings underscore the potential of permutation analysis in delving into diverse facets of EEG signal processing and analysis, empowered researchers to investigate the causal characteristics of signals, and evaluate the nonlinear dynamics within neural systems.
A growing body of research indicates sex-related differences in neural activity recorded in human EEG data [
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38,
39,
40,
41,
42,
43,
44]. Quantifiers derived from information theory, including markers based on the entropy of EEG background activity [
45], have shown potential for identifying these sex differences. Several approaches have been proposed to exploit this potential, such as estimating differences in brain status through entropy measurements [
46] and using permutation entropy to extract features for gender identification in emotional-based EEG datasets [
47] and non-stationary EEG signals [
48].
Applying these analyses to iEEG facilitates a deeper understanding of how men and women encode information during cognitive processes. A recent development involves creating a database to compile iEEG data from normal brain regions [
49,
50,
51]. Given the impracticality of screening asymptomatic patients using invasive and costly methods, this database, comprising epileptic patients with identified epileptogenic foci and distinct healthy (non-affected) regions, currently provides the closest approximation for studying signals from normal areas.
iEEG signals, which record the brain’s electrical activity, are obtained through electrodes implanted in the cranial area or brain tissue. These signals offer superior spatial resolution and less tissue attenuation compared to EEGs, facilitating the precise localization of specific brain regions. They reflect the activity of ensembles or populations of neurons responsible for encoding sensory, motor, or cognitive information. Adopting the perspective of neural population [
52] allows for investigating how the brain integrates and combines the activity of multiple neurons or brain regions to generate a coherent and efficient representation of internal or external stimuli.
This study aims to investigate how biological sex differences influence the way the brain encodes information through iEEG signals and to explore their impact on neuronal encoding dynamics. To achieve this, the study employs information theory quantifiers based on ordinal patterns that preserve causal coding features, using the Bandt and Pompe method. Complete signals are analyzed using the complexity-entropy causality plane and power spectral density. In each region, the mean, median, and Mann–Whitney U test were applied to entropy and complexity to compare between sexes. The analysis is based on iEEG signals sourced from healthy sectors of the human brain sourced from the Open iEEG Atlas [
49,
50,
51].
The objective is to enhance the understanding of sex-based differences in brain activity through a comprehensive exploration of local field potentials. This involves the segregation of signals by sex, region, and hemisphere, followed by their analysis and the overlaying of results for graphical comparisons and statistical tests. Ultimately, the purpose of this study is to broaden the foundation for future research in neuroscience, thereby advancing comprehension of the observed diversity in the human brain and the influence of biological sex on cerebral dynamics.
3. Results
This section presents the results of Shannon entropy and statistical complexity across different brain regions in the left and right hemispheres of both male and female individuals, using embedding dimensions of and , with a time window of 15 s and a delay of . Additionally, power spectra and the complexity-entropy plane are shown for those regions where significant differences between sexes were identified using the Mann–Whitney U test.
The first step involved examining the age distribution of the patients to account for any potential age-related differences in the results.
Table 1 displays the number of patients (n), mean age, and standard deviation for each region. Within each region, the age ranges for males and females were similar and fell within the standard deviation, which supported the assumption that no significant disturbances in the calculated quantifiers arose from age differences.
Differences in Shannon entropy and statistical complexity across different brain regions in the left and right hemispheres of male and female individuals are presented in
Table 2. The analysis was conducted using the embedding dimension
. Only regions with a minimum of five female and five male patients were included in the analysis. For each region, the mean and median values of Shannon entropy and statistical complexity were calculated separately for males and females. The statistical significance of the differences between male and female values was assessed using the Mann–Whitney
U test, with the results presented as
p-values and
h-index. Significant differences in both Shannon entropy and statistical complexity between males and females were observed in certain brain regions, as highlighted in the table. Conversely, the remaining regions showed no significant differences between sexes for either measure.
The following figures display the PSD and complexity-entropy causality plane for various brain regions in the left and right hemispheres. Each figure shows results from both male and female patients, normalized to the total power in each channel. The format is consistent across all figures: (A) and (B) present the median PSD for female and male patients respectively (where the shaded area represents the interquartile range, IQR), (C) shows the median PSD for both sexes, and (D) illustrates the complexity-entropy causality plane for each sex using . Each scatter plot includes a boxplot for each sex, showing the median, IQR, outliers, and the notch, which indicates the 95% confidence interval of the median.
The figures correspond to the brain regions highlighted in
Table 2, which showed significant differences between males and females based on the statistical test with
. These regions include the superior parietal lobule (
Figure 2), supramarginal gyrus (
Figure 3), precuneus (
Figure 4), posterior cingulate (
Figure 5), supplementary motor cortex (
Figure 6), the triangular part of the inferior frontal gyrus (
Figure 7), and the middle temporal gyrus in the left hemisphere (
Figure 8), as well as the superior temporal gyrus in the right hemisphere (
Figure 9).
In
Table 2, using
, it can be seen that differences between the entropy and statistical complexity results for both biological sexes were observed in half of the analyzed regions, as quantified by a Mann–Whitney test. The normalized spectral analysis of the signals also revealed differences between the median results for both sexes. The figures provide a qualitative analysis, showing not only the medians but also the dispersion of the data through interquartile ranges (IQRs).
In the left hemisphere, seven regions with significant differences were identified for
. In the superior parietal lobule, the median complexity and entropy values showed disjoint notches in the boxplots, and the statistical test indicated that the entropy values come from different distributions (
Figure 2D).
Figure 2C shows that the PSD had a peak for both men and women. For women, this peak was located at the boundary of the
bands, while for men, it was shifted towards the
band.
The supramarginal gyrus region exhibited differences in the PSD spectra. A peak was observed in the spectrum for women at the end of the
band, while the male spectrum lacked peaks in both the
and
bands (
Figure 3C). The complexity-entropy plane also showed differences, with notched medians indicating distinct distributions for men and women (
Figure 3D), as confirmed by statistical testing.
Similar patterns were observed in the precuneus region. The female PSD displayed a peak between the
bands, whereas no peaks were seen in the male PSD within these regions (
Figure 4C). Although the notches in the medians had some overlap (
Figure 4D), the statistical test revealed differences in entropy values between men and women.
In the posterior cingulate region (
Figure 5), both spectra showed a small peak in the boundary between the
and
bands, with the male peak shifted to slightly lower frequencies (
Figure 5C). Regarding complexity and entropy (
Figure 5D), while some overlap was observed between the notches in entropy, the values for complexity were well separated. Statistical tests indicated that the distributions for both quantifiers were different.
The supplementary motor cortex presented a peak at the beginning of the
band in both spectra. However, as frequencies increased, differences emerged: the female spectrum showed a peak in the
band, which was absent in the male spectrum (
Figure 6C). Although the medians for complexity and entropy were well differentiated with non-overlapping notches, statistical tests revealed significant differences only for entropy (
Figure 6D).
In the triangular part of the inferior frontal gyrus region, both spectra showed a peak in the
band, but the female peak was shifted toward lower frequencies. Additionally, the male PSD featured a second peak centered in the
band which was not present in the female spectrum (
Figure 7). Despite overlapping notches in the medians for complexity and entropy, statistical tests indicated differences between sexes for entropy.
The middle temporal gyrus region showed differences between the
and
bands in the PSD. A peak in the female spectrum occurred at the boundary between these bands, whereas the male spectrum displayed two peaks in the
band (
Figure 8C). While entropy medians were well differentiated, complexity medians presented overlapping notches (
Figure 8D). Statistical tests revealed significant differences in complexities, but not in entropies.
In the right hemisphere, among the four regions with acceptable statistics, only the superior temporal gyrus region displayed differences for
. Both spectra showed a peak, with the female peak located between the
and
bands, and the male peak shifted to the left in the
band with higher amplitude (
Figure 9C). Non-overlapping notches for both entropy and complexity were observed (
Figure 9D), but statistical tests revealed significant differences only for complexity.
To improve the statistics of BP and ensure that the length of the time series (M) was much greater than
, an embedding dimension of
was used with the same time window. Results for mean, median, and statistical tests of Shannon entropy and statistical complexity using
are presented in
Table 3.
With , the regions of the supramarginal gyrus, posterior cingulate, supplementary motor cortex, and middle temporal gyrus in the left hemisphere showed differences between males and females. Shannon entropy differences were detected in the first three regions, while all four regions displayed differences in statistical complexity. In the right hemisphere, differences were observed in two regions: the superior temporal gyrus, which showed differences in both quantifiers, and the middle temporal gyrus, which showed differences in entropy.
In summary, significant differences were observed in both and analyses in the following regions: the supramarginal gyrus, the posterior cingulate, the supplementary motor cortex, and the middle temporal gyrus in the left hemisphere, and the superior temporal gyrus in the right hemisphere.
4. Discussions
In summary, the findings of this study indicate that biological sex-based differences in brain function can be detected through iEEG signal analysis. These variations are evident in multiple brain regions, affecting both spectral characteristics and measures of complexity and entropy.
Further research is needed to explore the underlying causes of these differences. The invasive nature of iEEG signal collection prevents its application to healthy subjects, resulting in a restricted amount of LFP data from normal brain regions. However, despite constraints on patient numbers per region, iEEG analysis offers unparalleled spatial resolution compared to EEG, enabling the capture of region-specific LFPs and allowing for comparisons across sexes. Additionally, signals collected from electrodes implanted directly in the brain minimize attenuation caused by the skull and other tissues, a common issue with EEG, leading to less interference. These advantages come with challenges, as the “healthy” regions analyzed originate from the brains of epilepsy patients. This not only limits the dataset size but may also introduce bias, as the signals are taken from regions deemed “normal”, rather than from truly “healthy” individuals.
To address potential factors influencing the results, a strategy was employed to adjust for age within the analyzed sample, despite its limited size. The dataset was subsampled to include five patients for each region, aiming to minimize age differences between males and females while reducing standard deviations. The results of this analysis are detailed in this Section for the interested reader. The tables for
show that the number of regions exhibiting differences increased under this restriction (see
Table 4 and
Table 5) in both hemispheres. Additionally,
Table 6 provides the results of a significance analysis using the Benjamini–Hochberg FDR correction for
and
, applied to both the entire patient cohort and the subsets created to minimize differences and deviations. Reducing age differences in the dataset helps to minimize the impact of age-related variables or confounding factors that could introduce noise or skew the results. By reducing age variability between groups (i.e., males and females), the analysis can more directly focus on the effects of the variables of interest, reducing the influence of age as a potential source of noise or bias. This approach also increases sensitivity to variations attributable to biological sex, reinforcing the current findings.
Neuroscience research must increasingly acknowledge the vast diversity present in human brains. A key step in this direction is adopting a binary, proportional, and inclusive segmentation based on common biological classifications of male and female. To better understand brain dynamics, it is essential to reduce biases in sample selection and analyze the data both collectively and separately. Future research should also investigate whether these differences persist when considering perceived gender identity and hormonal backgrounds, as these factors can significantly influence brain structure and function. Incorporating these variables could provide a more comprehensive understanding of brain dynamics and help mitigate biases in neuroscience research.
This study found that both Shannon entropy and MPR statistical complexity can serve as sensitive biomarkers for biological sex in iEEG from awake patients. Recently, we proposed a framework constructed with Renyi entropy and its generalized statistical complexity, where we detected sex-based differences related to scale-free phenomena in non-REM sleep stages N1 and N2, as well as REM sleep [
67].
These findings underscore the necessity of addressing biases in experimental design. Including both males and females in the study population is crucial for ensuring that results are representative and not skewed by gender imbalances. Differences in average brain dynamics between sexes highlight the importance of inclusivity and equality in research. Despite advancements in gender equality, preclinical neuroscience often inadvertently excludes women, leading to biased data and conclusions. Thus, considering biological sex and gender differences is essential for obtaining a comprehensive understanding of brain function. Incorporating these variables into future studies could enhance our understanding of brain dynamics and contribute to reducing biases in neuroscience research.