Brain fingerprints along the language hierarchy

Juan Zhang, Liping Zhuang, Jiahao Jiang, Menghan Yang, Shijie Li, Xiangrong Tang, Yingbo Ma, Lanfang Liu, Guosheng DingFrontiers in Human Neuroscience 2022 · DOI

TL;DR

This study investigates how individual brain "fingerprints," based on functional connectivity, change with increasing language processing complexity. They found that more complex language tasks lead to higher individual identifiability, suggesting that higher-order cognitive functions contribute more to our unique brain patterns than basic sensory processing. This has implications for understanding brain individuality and developing personalized biomarkers.

What Problem Does This Paper Address?

For a long time, neuroscience research has focused on identifying universal brain mechanisms shared across individuals. However, there's a growing interest in understanding what makes each brain unique – its "fingerprint." While previous work has shown that functional connectivity patterns can reliably identify individuals, the underlying sources of this individualization remain unclear. Specifically, this paper asks: does the complexity of cognitive tasks influence how uniquely identifiable a person's brain is? And are higher-level cognitive functions, like understanding meaning, more critical for individual identification than lower-level sensory processing, like hearing sounds?

Methods

The researchers used functional Magnetic Resonance Imaging (fMRI) to scan 30 healthy, bilingual participants. Participants underwent resting-state fMRI and fMRI while listening to stories presented in three conditions, representing a hierarchy of language processing:

  1. Backward: Stories played in reverse, primarily engaging low-level auditory processing.
  2. Sentence-scrambled: Sentences within the story were shuffled, requiring middle-level linguistic and semantic processing.
  3. Intact: Stories played normally, engaging high-level conceptual processing and inference.

Each task condition was scanned twice to allow for within-session comparisons. Functional connectivity (FC) matrices were computed by parcellating the whole brain into 368 regions and calculating Pearson correlations between their time series. Individual identification was then performed by comparing FC matrices between different scan sessions. They used a "pairing scheme" where one session served as a "database" and another as a "target," calculating an "identifiability index" (I_diff) based on the similarity of a subject's own data versus others. They also analyzed FC within specific functional networks (auditory, language, default mode network (DMN), and frontoparietal) to see if certain networks were more critical for identification.

Datasets

The study used fMRI data collected from 30 healthy college students. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Language Hierarchy Matters: The complexity of the cognitive task significantly impacts how uniquely identifiable a person's brain is. As language processing moves from basic auditory perception to complex comprehension, individual brain fingerprints become stronger.
  • High-Order Functions Drive Uniqueness: Higher-level cognitive functions, particularly those involved in story comprehension and inference, contribute more to individual brain uniqueness than basic sensory processing.
  • Network Specialization: Not all brain networks contribute equally to individual identification. While the auditory network shows low identifiability, networks involved in higher-order cognition, like the frontoparietal network, are highly individualized.
  • Inter-Subject Variability is Key: Increased individual identifiability along the language hierarchy is associated with greater variability in functional connectivity patterns between subjects. This suggests that unique ways of processing complex information are what make our brains stand out.
  • Intrinsic Processes are Foundational: Even at the baseline (resting state), individuals are identifiable, suggesting that intrinsic brain processes play a significant role in forming these unique fingerprints, which are then modulated by task demands.

Key Results

  • Whole-Brain Identification:
    • Resting state identification accuracy (baseline) was around 62.9%.
  • Backward condition (low-level auditory): Success Rate (SR) of 85.2%.
  • Sentence-scrambled condition (middle-level): SR of 83.3%.
  • Intact condition (high-level conceptual): SR of 90.74%.
  • Identification accuracy was significantly higher for all task conditions compared to the resting state.
  • Network-Based Identification:
    • The auditory network consistently showed the lowest identification accuracies (14.8% - 48.1%) across all conditions.
  • The frontoparietal network performed best, achieving SRs from 61.1% to 92.6%, and showed increasing identifiability with task complexity.
  • Within- vs. Between-Subject Similarity:
    • Within-subject similarity (stability) was highest for the sentence-scrambled condition and lowest for the intact condition.
  • Between-subject similarity was lower for higher-level tasks (intact) compared to lower-level tasks (backward, scrambled), indicating greater inter-subject variability in complex processing.

Limitations and Future Work

The authors acknowledge several limitations:

  • Small Sample Size: The study involved only 30 participants, and all were female. Future work with larger and more diverse samples is needed to confirm these findings and assess generalizability.
  • Short Data Length: The duration of the stimuli and scan sessions might influence FC estimation. While they validated results with longer data, further exploration is warranted.
  • Statistical Significance: While trends showed increased identifiability with task complexity, many of these changes did not reach statistical significance. More research is needed to confirm the meaningfulness of these subtle shifts.

Future work could explore:

  • Males and Diverse Populations: Extending the study to male participants and other demographic groups.
  • Longitudinal Studies: Investigating the stability of these "brain fingerprints" over longer periods.
  • Causal Mechanisms: Manipulating brain states to further understand how executive processes contribute to individual differences.
  • Clinical Applications: Exploring how these findings can inform the development of personalized biomarkers for neurological or psychiatric conditions.

Summary updated

Abstract

Recent studies have shown that the brain functional connectome constitutes a unique fingerprint that allows the identification of individuals from a group. However, what information encoded in the brain that makes us unique remains elusive. Here, we addressed this issue by examining how individual identifiability changed along the language hierarchy. Subjects underwent fMRI scanning during rest and when listening to short stories played backward, scrambled at the sentence level, and played forward. Identification for individuals was performed between two scan sessions for each task as well as between the rest and task sessions. We found that individual identifiability tends to increase along the language hierarchy: the more complex the task is, the better subjects can be distinguished from each other based on their whole-brain functional connectivity profiles. A similar principle is found at the functional network level: compared to the low-order network (the auditory network), the high-order network is more individualized (the frontoparietal network). Moreover, in both cases, the increase in individual identifiability is accompanied by the increase in inter-subject variability of functional connectivities. These findings advance the understanding of the source of brain individualization and have potential implications for developing robust connectivity-based biomarkers.
BibTeX
@inproceedings{zhang2022brain,
  author = {Juan Zhang and Liping Zhuang and Jiahao Jiang and Menghan Yang and Shijie Li and Xiangrong Tang and Yingbo Ma and Lanfang Liu and Guosheng Ding},
  title = {{Brain fingerprints along the language hierarchy}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2022},
  doi = {10.3389/fnhum.2022.982905},
}
Figure