Mapping semantic representations of depression using simple lexical choices
Poster No:
1015
Submission Type:
Abstract Submission
Authors:
Line Kruse1, Roberta Rocca2, Mikkel Wallentin3
Institutions:
1Aarhus University, Aarhus, Aarhus, 2Aarhus University, Aarhus , Aarhus , 3Aarhus University, Aarhus C, DK
First Author:
Co-Author(s):
Introduction:
Depression is characterized as a disorder of self, involving maladaptive distortions in the experiential and narrative self (Newell et al., 2018; Davey & Harrison, 2022). Evidence indicate that differences in mental states are well reflected in both language use and processing and predict many psychiatric disorders, as well as personality and demographic traits (Behdarvandirad & Karami, 2022; Christian et al., 2021; Lai et al., 2021). Spatial demonstratives (in English "this" and "that") are typically used to distinguish peri- and extrapersonal space and have been shown to engage brain regions associated with spatial information processing (Rocca et al., 2020). Recent work based on the Demonstrative Choice Task (DCT) showed that choice of demonstratives are not only indicative of distance in a physical space, but also of the experienced or emotional proximity to the self (Rocca & Wallentin, 2020). Capturing important dimensions concerning self-focused mental representations, the DCT may encode information relevant to inferring the presence of depression and may provide means to identify and study the structure of semantic representations underlying individual differences in depression and other disorders of self.
Methods:
This work included two behavioral studies and an fMRI study using a 300-item DCT. Behavioral analyses were based on two independent samples of 775 and 879 participants, respectively. A PHQ9 sum score of 10 was used as threshold for partitioning participants into control and clinical group (Kroenke et al., 2001). Classification models were trained to predict outcome label (control or clinical) based on principal component representations of DCT responses.
fMRI analyses included 69 participants of which 32 were diagnosed with clinical depression and 37 were healthy. Data were preprocessed using standard fMRIprep preprocessing pipelines (Esteban et al., 2019). First-level T-contrasts and second-level intercept models addressed neural differences between trials associated with a proximal compared to distal demonstrative choice. Voxel-wise encoding models using wordnet categories as feature space were computed to assess individual differences in neural representations of semantic features (Huth et al., 2012). Encoding models included L2-regularized linear regression models fit to each subject independently.
fMRI analyses included 69 participants of which 32 were diagnosed with clinical depression and 37 were healthy. Data were preprocessed using standard fMRIprep preprocessing pipelines (Esteban et al., 2019). First-level T-contrasts and second-level intercept models addressed neural differences between trials associated with a proximal compared to distal demonstrative choice. Voxel-wise encoding models using wordnet categories as feature space were computed to assess individual differences in neural representations of semantic features (Huth et al., 2012). Encoding models included L2-regularized linear regression models fit to each subject independently.
Results:
Behavioral results showed that DCT based classification models predicted depression group significantly better than chance with F1-scores between 62% and 66% across samples. Semantic analyses showed that proximal responses for words scoring high on the features sadness, fear, disgust and anger predicted higher PHQ9 symptom scores, while the opposite was true for the features valence, joy and trust (Figure 1). fMRI analyses showed significantly increased activity in the left precuneus, p<.05 (Figure 2), for trials associated with proximal compared to distal demonstrative forms. Voxel-wise encoding models indicated individual differences in neural representations across semantic features related to depression symptom scores.
·Figure 1
·Figure 2
Conclusions:
Results indicated that a simple lexical choice task reliably captured semantic characteristics of experiential states that are predictive of depression symptom severity and recover semantic effects of valence previously associated with depression. The paradigm engaged posterior parietal regions typically associated with spatial processing and a core node in the DMN, indicating that choice of proximal demonstratives on the DCT involves increased self-referential processing of DCT items. Assessing subject-specific semantic maps of neural representations across semantic features of DCT items support depression-related differences observed in the behavioral effects. Investigating and characterizing semantic representations underlying individual experiential states may contribute importantly to symptom profiling approaches and provide novel information on individual differences in depressive states.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 2
Language:
Language Comprehension and Semantics 1
Keywords:
Affective Disorders
Cognition
Emotions
Language
Psychiatric Disorders
1|2Indicates the priority used for review
Provide references using author date format
Christian, H., et al. (2021). Text based personality prediction from multiple social media data sources using pre-trained language model and model averaging. Journal of Big Data, 8(1), 68.
Davey, C. G., & Harrison, B. J. (2022). The self on its axis: A framework for understanding depression. Translational Psychiatry, 12(1), Article 1.
Esteban, O. et al. (2019). fMRIPrep: A robust preprocessing pipeline for functional MRI. Nature Methods, 16(1), Article 1.
Huth, A. G. et al. (2012). A Continuous Semantic Space Describes the Representation of Thousands of Object and Action Categories across the Human Brain. Neuron, 76(6), 1210–1224.
Kroenke, K. et al. (2001). The PHQ-9: Validity of a brief depression severity measure. Journal of General Internal Medicine, 16(9), 606–613.
Lai, Y. et al. (2021). Neural Demographic Prediction in Social Media with Deep Multi-view Multi-task Learning. In C. S. Jensen, E.-P. Lim, D.-N. Yang, W.-C. Lee, V. S. Tseng, V. Kalogeraki, J.-W. Huang, & C.-Y. Shen (Eds.), Database Systems for Advanced Applications (pp. 271–279). Springer International Publishing.
Newell, E. E. et al. (2018). You Sound So Down: Capturing Depressed Affect Through Depressed Language. Journal of Language and Social Psychology, 37(4), 451–474.
Rocca, R. et al. (2020). Language beyond the language system: Dorsal visuospatial pathways support processing of demonstratives and spatial language during naturalistic fast fMRI. NeuroImage, 216, 116128.
Rocca, R., & Wallentin, M. (2020). Demonstrative Reference and Semantic Space: A Large-Scale Demonstrative Choice Task Study. Frontiers in Psychology, 11.