A Computational Account of the Mechanisms Underlying Face Perception Biases in Depression

Abstract

Here, we take a computational approach to understand the mechanisms underlying face perception biases in depression. Thirty participants diagnosed with Major Depressive Disorder and thirty healthy control participants took part in three studies involving recognition of identity and emotion in faces. We used signal detection theory to determine whether any perceptual biases exist in depression aside from decisional biases. We found lower sensitivity to happiness in general, and lower sensitivity to both happiness and sadness with ambiguous stimuli. Our use of highly-controlled face stimuli ensures that such asymmetry is truly perceptual in nature, rather than the result of studying expressions with inherently different discriminability. We found no systematic effect of depression on the perceptual interactions between face expression and identity, suggesting that depression is not associated with difficulty attending to one of these dimensions while filtering out the other. We show through simulation that the overall pattern of results, as well as other biases found in the literature, can be explained by a neurocomputational model in which neural populations encoding positive expressions are selectively suppressed.

Changes within neural population codes can be inferred from psychophysical threshold studies

“…Second, there are multiple reports that an aversive CS+ captures attention in search tasks [48,49,50,51]. A reasonable assumption is that bottom-up attentional capture depends on the overall level of neural activity that the CS+ produces in comparison with concurrently-presented stimuli that compete for attention [43]. If we think of that overall level of neural activity as the result of both the number of neurons selective for the CS+ as well as their firing rates, we see that both suppression and outward shift are mechanisms likely to reduce the overall level of neural activity produced by the CS+, whereas an inward shift is likely to produce the opposite effect.…”

“…The tuning functions used by the brain to encode such dimensions might be different than what is represented by the standard population model used here. For example, face features are thought to be encoded through monotonic tuning functions (e.g., sigmoidal; see [5,45,43]). Using computational modeling and visual adaptation, it has been found that the effects of categorization on perception of face identities along the category-relevant dimension [59,60,61] can be best explained using a specific gain mechanism [45].…”

“…In addition, additional psychophysical studies may be able to differentiate between the two mechanisms, as they should differ in the overall level of activity produced by presenting the target. Assuming that the overall level of activity elicited by a stimulus determines its relative salience, producing bottom-up attention, then a target under specific gain should capture attention more easily in visual search tasks than a target under specific tuning [43]. More generally, at the heart of our approach is the idea that no single study allows one to infer the correct mechanism of encoding change underlying some behavior of interest.…”

“…Fourth, we assumed a bellshaped tuning function, which as far as we know is the only type of function used in research applying inverted encoding modeling [e.g., 9,38,31,15,13,11,12,10]. For other stimulus dimensions, such as those characterized by monotonically increasing or decreasing tuning functions (e.g., face dimensions; [44,5,45,43]), new simulations would be required. Finally, we assumed a homogeneous encoding population for the baseline condition.…”

“…For orientation stimuli, we chose to model channels as symmetrical Gaussians with Poisson noise based on previous literature, but there are other ways to characterize tuning functions that may be more appropriate for other types of stimuli. For example, face dimensions are often modeled using asymmetric sigmoidal tuning functions [44,5,45,43]. Also, another common assumption for neural channel noise is that it follows a Gaussian distribution [e.g., 6, 7], a common assumption implicit in inverted encoding modeling analyses of neuroimaging data [see 8,14].…”