Task-dependent fractal patterns of information processing in working memory

Jeremi K. Ochab, Marcin Wątorek, Anna Ceglarek, Magdalena Fąfrowicz, Koryna Lewandowska, Tadeusz Marek, Barbara Sikora-Wachowicz, Paweł Oświȩcimka

TL;DR

This paper demonstrates that the "fractalness" of brain activity—how signal patterns repeat across different time scales—changes dynamically based on the specific cognitive task at hand. By applying advanced fractal and spectral analyses to fMRI data, the researchers show that these temporal patterns are more sensitive than traditional correlation methods at distinguishing between different phases of working memory, such as memorizing versus recalling information.

What Problem Does This Paper Address?

Neuroscientists often study working memory by looking at where the brain is active or how different regions "talk" to each other via simple correlations. However, brain signals (specifically BOLD signals from fMRI) are "non-stationary" and complex; they exhibit long-range temporal correlations, meaning what the brain did a few seconds ago influences what it does now in a mathematically self-similar (fractal) way.

The challenge is that standard tools like Pearson correlations often miss the subtle "texture" of these signals. This paper investigates whether the fractal characteristics of brain activity change depending on the type of information being processed (verbal vs. non-verbal) and the stage of the memory process (encoding vs. retrieval). Essentially, they ask: Does the "rhythm" of information processing change when you're trying to remember a word versus a shape?

Methods

The researchers employed several sophisticated mathematical techniques to analyze fMRI time-series data:

  • Detrended Fluctuation Analysis (DFA): Used to calculate the Hurst exponent ($H$), which measures the "memory" or persistence of a signal. An $H > 0.5$ indicates a persistent signal (trends tend to continue), while $H \approx 0.5$ indicates random noise.
  • Power Spectral Density (PSD): A complementary frequency-domain analysis used to find the spectral exponent ($\beta$), helping confirm the fractal scaling properties.
  • Eigenanalysis of Detrended Cross-Correlations (DCC): Unlike standard correlations, this method accounts for non-stationarity in the data. By looking at the "eigenvalues" of these correlation matrices, the team could identify how entire networks of brain regions synchronize during specific tasks.
  • Wavelet Transform: Used to create "scalograms" (visual heatmaps of signal intensity across different scales) to qualitatively observe the hierarchical structure of brain activity.

Datasets

The study involved 54 healthy participants (32 females, mean age ~24). The data consisted of fMRI BOLD signals recorded during:

  • Four Visual Tasks: Two verbal (semantic words and phonological pseudowords) and two non-verbal (global and local processing of abstract figures).
  • Two Memory Phases: Encoding (memorizing the stimuli) and Retrieval (identifying if a "probe" stimulus was seen before).
  • Resting State: A 10-minute baseline with eyes open.
  • Diurnal Variation: Participants performed tasks in both morning and evening sessions to account for time-of-day effects.

Key Insights and Ideas

  • Tasks Break the "Resting" Rhythm: The brain is most "fractal" and persistent during rest ($H \approx 1.1$). When engaged in a task, the Hurst exponent drops ($H \approx 0.7$), meaning the signal becomes less predictable and more "efficient" for processing new information.
  • Encoding is More Taxing than Retrieval: The drop in $H$ was significantly more pronounced during the encoding phase than the retrieval phase, particularly in the visual cortex. This suggests that the act of "writing" information into working memory requires a more complex, less persistent neural state than "reading" it back out.
  • Verbal vs. Non-Verbal Signatures: Verbal tasks showed lower $H$ values in sensory-motor and visual regions compared to non-verbal tasks. This implies that processing language-based information alters the temporal structure of neural activity differently than processing abstract shapes.
  • Fractals Beat Pearson: The authors found that fractal scaling was a more sensitive metric for distinguishing between task types and memory phases than standard Pearson correlations, which often failed to catch the subtle differences between encoding and retrieval.

Key Results

  • Hurst Exponent ($H$): Averaged across the brain, $H$ was ~0.049 lower during tasks than at rest. During the encoding phase, it dropped by an additional ~0.020 compared to retrieval.
  • Regional Specificity: The Auditory network showed the highest persistence (highest $H$), while the Visual II network showed the lowest $H$ during tasks, marking it as a primary hub for task-dependent changes.
  • Network Clustering: Eigenanalysis revealed that the Default Mode Network (DMN) and Executive Function networks showed opposite activation patterns compared to the Sensory-Motor and Visual networks, a pattern that remained consistent across memory phases but changed in intensity.

Limitations and Future Work

  • Temporal Resolution: fMRI has a low sampling rate (TR = 1.8s in this study). Fractal analysis is generally more robust with higher-frequency data, such as that from EEG or MEG.
  • False Memories: While the study used the DRM paradigm (famous for inducing false memories), the current analysis did not distinguish between "true" and "false" retrieval events. The authors suggest this as a future direction.
  • Task Complexity: The study focused on relatively simple visual tasks. Future work could explore if more complex, multi-modal tasks (like social interaction or navigation) produce even more distinct fractal "fingerprints."

Reference: Ochab, J. K., et al. (2022). Task-dependent fractal patterns of information processing in working memory. arXiv:2209.13297v1 [q-bio.NC].

Summary updated

Abstract

We applied detrended fluctuation analysis, power spectral density, and eigenanalysis of detrended cross-correlations to investigate fMRI data representing a diurnal variation of working memory in four visual tasks: two verbal and two nonverbal. We show that the degree of fractal scaling is regionally dependent on the engagement in cognitive tasks. A particularly apparent difference was found between memorisation in verbal and nonverbal tasks. Furthermore, the detrended cross-correlations between brain areas were predominantly indicative of differences between resting state and other tasks, between memorisation and retrieval, and between verbal and nonverbal tasks. The fractal and spectral analyses presented in our study are consistent with previous research related to visuospatial and verbal information processing, working memory (encoding and retrieval), and executive functions, but they were found to be more sensitive than Pearson correlations and showed the potential to obtain other subtler results. We conclude that regionally dependent cognitive task engagement can be distinguished based on the fractal characteristics of BOLD signals and their detrended cross-correlation structure.
BibTeX
@inproceedings{ochab2022task,
  author = {Jeremi K. Ochab and Marcin Wątorek and Anna Ceglarek and Magdalena Fąfrowicz and Koryna Lewandowska and Tadeusz Marek and Barbara Sikora-Wachowicz and Paweł Oświȩcimka},
  title = {{Task-dependent fractal patterns of information processing in working memory}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-21375-1},
}

Loading related content...

Figure