Abstract
Theory of Mind (ToM)—the ability to infer the mental states of others—is a cornerstone of human social interaction. Deficits in ToM are central to various mental disorders, yet traditional diagnostic methods often rely on subjective assessments that treat ToM as an opaque "black box." This paper introduces the Visual Theory of Mind (v-ToM) framework, a novel conceptual model that decomposes social inference into a four-stage computational process. By integrating eye-tracking technology with an audio-visual paradigm, we demonstrate that specific "gaze signatures" can objectively distinguish individuals with Depressive Disorder (DD) and Autism Spectrum Disorder (ASD) from healthy controls. Our findings, based on a sample of over 700 participants, provide a path toward AI-aided digital medicine and precision identification in mental health.
Introduction
The capacity to navigate social environments depends on our ability to attribute beliefs (cognitive ToM) and emotions (affective ToM) to others. When these processes are impaired, individuals face significant challenges in social integration. Historically, clinical assessments of ToM have utilized verbal tasks and questionnaires. However, these methods are limited by their subjectivity and their inability to capture the fine-grained, moment-to-moment processing involved in social cognition. To address these gaps, there is an urgent need for objective, quantifiable biomarkers that can pinpoint specific breakdowns in the ToM process.
The Visual Theory of Mind (v-ToM) Framework
We propose the v-ToM framework to model social inference as a measurable, four-stage computational sequence:
- S1: Visual Processing: The initial perceptual detection of basic features such as color, orientation, and borders.
- S2: Mental Processing: The conversion of visual information into internal "propositions" (e.g., identifying that a character is looking at a specific object).
- S3: Assessment: The comparison of these propositions against "prototypes" or schemas stored in long-term memory (e.g., recognizing that a helpful agent's presence implies a specific outcome).
- S4: Transfer & Implementation: The use of these mental representations to guide subsequent behavior and eye movements.
By breaking down ToM into these stages, we can identify exactly where the social-cognitive process deviates in different clinical populations.
Methods and Datasets
The framework was validated through two large-scale studies involving over 700 participants. We employed an audio-visual paradigm where eye movements were recorded (using Tobii and EyeLink trackers) while participants viewed social and emotional scenes.
- Study 1 (Affective ToM): Compared 212 individuals with Depressive Disorder (DD) against 492 healthy controls using a "competing-priming effect" paradigm (contrasting emotional faces against backgrounds).
- Study 2 (Cognitive ToM): Compared 332 children with Autism Spectrum Disorder (ASD) against 612 typically developing (TD) children using social inference tasks involving a character interacting with objects and social agents.
Key Results
Attentional Bias in Depressive Disorder
In the affective ToM tasks, individuals with DD exhibited a distinct "disengagement" deficit. Specifically, DD patients showed significantly larger transit times when shifting their gaze from negative backgrounds to positive faces ($p < .01$). This suggests that negative information "traps" their attention, representing a measurable physical manifestation of the cognitive biases associated with depression.
Social Inference Deficits in Autism
In the cognitive ToM tasks, 4- and 5-year-old TD children significantly increased their fixations on a target object when a helpful social agent was present ($p < .01$), indicating they had successfully inferred the agent's intent. In contrast, the ASD group showed no significant change in fixation patterns ($p = .35$), suggesting a failure to integrate social "prototypes" into their mental processing (Stage S3).
Discussion and Future Directions
The v-ToM framework transforms an abstract psychological concept into a computable model. By identifying specific gaze signatures, we move away from subjective assessments toward objective digital biomarkers. These indicators not only allow for the precision identification of DD and ASD but also offer potential for transdiagnostic interventions.
Future work will focus on applying these models to dynamic, real-world social interactions and developing AI-driven personalized therapy programs that "train" the brain to process social information more effectively. The ultimate goal is a fully integrated, AI-aided precision identification system for mental health.
Conclusion
The integration of eye-tracking data with the v-ToM framework provides a robust foundation for the next generation of mental health diagnostics. By quantifying the stages of social inference, we can better understand the underlying mechanisms of mental disorders and pave the way for more effective, data-driven clinical interventions.
Summary updated