Decoding individual identity from brain activity elicited in imagining common experiences

The research question centers on whether individual differences in the mental simulation of common experiences can be detected and quantified through neuroimaging. The study's context is the burgeoning field of cognitive neuroscience, which seeks to understand how personal memories and experiences are encoded and processed in the brain. Previous fMRI studies have identified brain networks associated with episodic memory and mental imagery, but these have largely focused on group-level averages, failing to capture individual differences. This study aims to address this gap by using personalized models to decode individual neural signatures from fMRI data during the mental imagery of common experiences. The importance of this research lies in its potential to advance our understanding of how individual experiences shape brain activity, with implications for both basic cognitive science and clinical applications, such as diagnosing and treating memory disorders. The hypothesis is that fMRI activity patterns elicited while participants imagine common scenarios will reflect interpersonal differences in information recalled and that these differences can be predicted using personalized models of their imagery.

Prior fMRI research has identified a network of brain regions consistently activated during the recollection and imagination of experiences. These regions show considerable overlap, suggesting similar neural mechanisms are involved. Studies have also shown relationships between activation patterns and individual differences in remembering and imagining events, with regions like the medial parietal cortex, prefrontal cortex, and temporal lobes showing variations. However, it has remained unclear if these differences reflect functionally relevant information or mere noise. Recent work has demonstrated the ability to predict individual differences in cognitive domains using fMRI, including resting-state activity and task-related brain responses. While these studies show individual differences can be discerned from fMRI data, it is unclear whether these methods generalize to the specific task of imagining personal scenarios. Furthermore, studies of between-group differences in autobiographical memory-related fMRI activation exist (e.g., in Alzheimer's disease), but these don't fully address the subtle differences between individuals within the same group. Existing neural decoding models often rely on inter-subject averages or generic semantic models, potentially obscuring individual differences.

Twenty-six healthy participants (17 female, 9 male, mean age 73) vividly imagined and verbally described 20 common scenarios (e.g., dancing, shopping, wedding). They then rated their imagined scenarios on 20 experiential attributes (sensory, motor, cognitive, temporal, emotional). Verbal descriptions were mapped to a distributional semantic model (GloVe) to create a 'verbal model'. An 'attribute model' was created from the attribute ratings. Participants underwent fMRI while re-imagining the scenarios. fMRI data were preprocessed using standard techniques, and the brain was parcellated into 90 anatomical regions of interest (ROIs). Representational Similarity Analysis (RSA) was used to compare model and fMRI data. The primary analysis examined whether fMRI representations were better predicted by personal models than group-average models. A secondary analysis tested whether individual identity could be decoded from fMRI activity by matching participants' fMRI activation to their personal models. This decoding analysis was repeated across various ROIs.

Representational Similarity Analysis (RSA) revealed that participants' neural representations were better predicted by their own personalized models (integrating verbal and attribute data) than by group-average models. This effect was observed across multiple brain regions, including the medial parietal cortex (MPC), left temporoparietal junction (LTPJ), and left dorsolateral prefrontal cortex (DLPFC). The results strongly supported the hypothesis that fMRI activation patterns during the imagination of common scenarios reflect interpersonal differences. A further analysis demonstrated that individual identity could be decoded from fMRI activity with high accuracy (around 75%) across several ROIs. This finding indicates that the brain activity during the imagery task contains information unique to each individual, allowing for their identification based solely on neural data. Supplementary analyses confirmed the robustness of these results across various parameter adjustments (e.g., number of voxels, model types).

The findings demonstrate that fMRI can measure individual differences in brain activity during the imagination of complex events. The success in predicting fMRI data using personalized models represents a significant advancement over previous approaches relying on group-level models. The involvement of regions like the MPC (associated with episodic memory and scene perception), LTPJ (linked to self-perspective and event segmentation), and DLPFC (involved in cognitive control) highlights the complex neural processes underlying individual differences in mental imagery. The high accuracy in decoding individual identity underscores the unique neural signatures encoded in brain activity during mental imagery. The results support the integration of episodic and semantic memory within a unified predictive modeling framework, offering potential advancements in understanding memory and its associated disorders. The potential of these methods in clinical settings is significant, offering possibilities for characterizing and diagnosing memory-related disorders.

This study successfully decoded individual identity from fMRI brain activity during the imagination of common scenarios, demonstrating that personalized models significantly improve the prediction of neural activity compared to group-level approaches. The findings highlight the unique neural signatures reflecting individual differences in mental imagery and suggest promising applications for clinical diagnostics and personalized treatments. Future research should explore the role of perspective-taking in these models, investigate younger populations, and further examine the relationships between these neural signatures, thought profiles, and psychosocial traits.

The study focused on older adults (mean age 73), limiting the generalizability of findings to younger populations. The sample size, while substantial, could be further increased for enhanced statistical power. The use of a specific set of scenarios and attributes might limit the generalizability to other types of events or experiences. The reliance on self-reported verbal descriptions and ratings may introduce subjective biases into the models.