Robust Multivariate Assessment of Empathic Function without Empathy Tasks
Poster No:
802
Submission Type:
Abstract Submission
Authors:
Leonardo Christov-Moore1, Nicco Reggente1
Institutions:
1Institute for Advanced Consciousness Studies, Santa Monica, CA
First Author:
Co-Author:
Introduction:
Deficits in empathic function have deleterious effects on individual, relational and community function, encouraging isolation, increasing the risk of unemployment and homelessness, and impacting long-term health outcomes. Assessing empathic function in vulnerable neurodivergent or nonverbal populations using self-reports and in-scanner tasks is frequently unfeasible. Encouraging evidence suggests that characteristic interactions between brain networks underlying empathy are observable at rest. Leveraging this to assess empathic function without self-reports or in-scanner tasks could be invaluable for clinical practice. We tested whether machine learning-aided analysis (LASSO) of resting fMRI data could predict subdimensions of empathy (empathic concern, personal distress, and perspective-taking) in 74 healthy participants.
Methods:
Participants: 74 equivalently recruited participants aged 18-26 (38 female) scanned at the Ahmanson-Lovelace Brain Mapping Center at UCLA on a Siemens Trio 3T between 1/12/2015 and 6/22/2016. Eligibility criteria included: right handed, no prior or concurrent diagnosis of any neurological, psychiatric, or developmental disorders, and no history of drug or alcohol abuse.
Procedure :Participants filled out the Interpersonal Reactivity Index (IRI), to assess three subdimensions of empathy: Empathic Concern: sympathetic reactions to the distress of others; Perspective Taking: the tendency to take other's perspective; Personal Distress: aversive reactions to the distress of others.
Resting state data was acquired via a series of MRI scans conducted in a Siemens Trio 3T scanner housed in the Ahmanson-Lovelace Brain Mapping Center at UCLA. Participants passively observed a white fixation cross on a black screen. Images were acquired over 36 axial slices covering the whole cerebral volume using an echo planar T2*-weighted gradient echo sequence. A T1-weighted volume was also acquired.
Following motion correction, high-pass filtering (.01Hz) and smoothing (6mm FHWM). Preprocessed data was subjected to probabilistic ICA in MELODIC. Noise components were identified and removed. Functional data was coregistered to standard space (MNI 152 template) via nonlinear registration (FNIRT). We created resonance network and control network with 22 ROIs within the Seitzman atlas. Using mean time-courses from each ROI, correlation matrices were created for each participant. We leveraged a LASSO regression model built on N-10 participants' feature sets for each IRI subscale. The model's intercept and outcome beta values were used as coefficients for each left-out subject's feature set-obtaining a predicted subscale measure for that individual. After N folds, we correlated the predicted values with the actual values, yielding Pearson's R– a measure of our model's ability to capture variance across participants. We repeated this cross-validation 50 times, then averaged R values to converge on a true test statistic estimate. To correct for multiple comparisons, matrices of p-values were created using a Benjamini-Hochberg approach in R and corrected p-values were considered significant at the 5% positive tail. To enhance replicability, we also applied Bonferroni correction and generally treated this as 'full' significance.
Procedure :Participants filled out the Interpersonal Reactivity Index (IRI), to assess three subdimensions of empathy: Empathic Concern: sympathetic reactions to the distress of others; Perspective Taking: the tendency to take other's perspective; Personal Distress: aversive reactions to the distress of others.
Resting state data was acquired via a series of MRI scans conducted in a Siemens Trio 3T scanner housed in the Ahmanson-Lovelace Brain Mapping Center at UCLA. Participants passively observed a white fixation cross on a black screen. Images were acquired over 36 axial slices covering the whole cerebral volume using an echo planar T2*-weighted gradient echo sequence. A T1-weighted volume was also acquired.
Following motion correction, high-pass filtering (.01Hz) and smoothing (6mm FHWM). Preprocessed data was subjected to probabilistic ICA in MELODIC. Noise components were identified and removed. Functional data was coregistered to standard space (MNI 152 template) via nonlinear registration (FNIRT). We created resonance network and control network with 22 ROIs within the Seitzman atlas. Using mean time-courses from each ROI, correlation matrices were created for each participant. We leveraged a LASSO regression model built on N-10 participants' feature sets for each IRI subscale. The model's intercept and outcome beta values were used as coefficients for each left-out subject's feature set-obtaining a predicted subscale measure for that individual. After N folds, we correlated the predicted values with the actual values, yielding Pearson's R– a measure of our model's ability to capture variance across participants. We repeated this cross-validation 50 times, then averaged R values to converge on a true test statistic estimate. To correct for multiple comparisons, matrices of p-values were created using a Benjamini-Hochberg approach in R and corrected p-values were considered significant at the 5% positive tail. To enhance replicability, we also applied Bonferroni correction and generally treated this as 'full' significance.
·A priori Resonance and Control functional networks
Results:
rsFC within a priori networks predicted empathic concern and perspective-taking. Empathic concern was predicted by a far wider array of systems than personal distress and perspective-taking: out of 12 individual and 14 combined networks tested, empathic concern was significantly predicted by 7 individual and 8 combined networks. In contrast, Personal Distress was predicted by 1 individual and 2 combined, and Perspective-Taking by 2 individual and 4 combined, respectively.
·Networks significantly predictive of Empathic Concern
Conclusions:
Trait empathy can be robustly predicted from resting brain activity, with possible applications for diagnosis in vulnerable populations.
Emotion, Motivation and Social Neuroscience:
Social Cognition 1
Modeling and Analysis Methods:
Multivariate Approaches 2
Keywords:
Cognition
Data analysis
Machine Learning
MRI
Social Interactions
Other - empathy
1|2Indicates the priority used for review