FMRI multitask deep phenotyping: subject-specific deviation analysed across contrasts and tasks
Poster No:
655
Submission Type:
Abstract Submission
Authors:
Philipp Sämann1, BeCOME study team1, Michael Czisch1
Institutions:
1Max Planck Institute of Psychiatry, Munich, Germany
First Author:
Co-Author(s):
Introduction:
Task fMRI is considered powerful to capture individual and task specific brain states and by this phenotype neuropsychiatric conditions. In the BeCOME study that is in part aligned to the RDoC classification principles we are currently deep-phenotyping subjects in a spectrum from super-healthy to moderate to severe anxiety and depression disorders [REF]. One major component of the neurobiological measurements is a task-fMRI battery distributed over two fMRI session on separate days (see methods). The major challenge lies in breaking down the fMRI task data to meaningful, condense information substrate that captures inter-individual variability on one hand, but allows for interpretation of the feature space. Here we pursue an BOLD activation based approach and similarity analyses between subjects to rank them, depending on 'atypicality' within the sample, both using separate key contrasts of the task model, the entire task, or all tasks. Our main questions are: (1) How independent is the information from different tasks and task contrasts when it comes to subject ranking along these axes? (2) How important is BOLD amplitude infomation gained from deactivations (also referred to as task-negative networks)?
Methods:
In the BeCOME study we include healthy subjects and subjects with current or past disorders of the anxiety/depression spectrum.
A data freeze was made July 2023 which left us with 238 subjects of which a homogeneous set of the following four tasks was available. TASKS. (i) N-Back task (with contrasts NB2<>NB0, NB2<>NB1, NB2<>NB1 and NB<>fixation), (ii) Reward anticipation task (money gain, verbal feedback, control; with contrasts money<>control, verbal<>control and money<>verbal), (iii) Hariri emotional faces matching task (with contrasts: all faces<>geometry, negative<>geometry, negative<>neutral) and (iv) Time Estimation Task with three feedback types (with contrasts correct<>false, correct<>uncertain and false<>uncertain). fMRI ANALYSIS: State-of-the-art preprocessing with slice timing correction, motion correction, spatial normalisation (DARTEL), physiological noise correction using CompCor, classical general linear models to estimate regressor influences and to generate individual contrast maps, SECOND LEVEL ANALYSIS AND ROI DEFINITION: ROIs were positioned in centers of group activation and deactivation maps; a 7x7x7 voxel box was weighted with the group T statistics. A total of 24 contrasts for the four tasks was defined, with 39 (N-back) , 31 (Reward), 33 (Hariri) and 22 (TET) ROIs, and contrast values extracted from 238 subjects, resulting in 24 specific poly-regional BOLD response vectors per subject. CALCULATION OF INDIVIDUAL RANK: Per contrast (and later per task and across all tasks) we calculated the simple Eucledian distance which can be understood as a measure of (statistical) atypicality or abnormality.
A data freeze was made July 2023 which left us with 238 subjects of which a homogeneous set of the following four tasks was available. TASKS. (i) N-Back task (with contrasts NB2<>NB0, NB2<>NB1, NB2<>NB1 and NB<>fixation), (ii) Reward anticipation task (money gain, verbal feedback, control; with contrasts money<>control, verbal<>control and money<>verbal), (iii) Hariri emotional faces matching task (with contrasts: all faces<>geometry, negative<>geometry, negative<>neutral) and (iv) Time Estimation Task with three feedback types (with contrasts correct<>false, correct<>uncertain and false<>uncertain). fMRI ANALYSIS: State-of-the-art preprocessing with slice timing correction, motion correction, spatial normalisation (DARTEL), physiological noise correction using CompCor, classical general linear models to estimate regressor influences and to generate individual contrast maps, SECOND LEVEL ANALYSIS AND ROI DEFINITION: ROIs were positioned in centers of group activation and deactivation maps; a 7x7x7 voxel box was weighted with the group T statistics. A total of 24 contrasts for the four tasks was defined, with 39 (N-back) , 31 (Reward), 33 (Hariri) and 22 (TET) ROIs, and contrast values extracted from 238 subjects, resulting in 24 specific poly-regional BOLD response vectors per subject. CALCULATION OF INDIVIDUAL RANK: Per contrast (and later per task and across all tasks) we calculated the simple Eucledian distance which can be understood as a measure of (statistical) atypicality or abnormality.
Results:
Key contrasts were inspected to detect preprocessing or modelling errors. Figure 1 exemplifies the positioning of a total of 13 ROIs (9 in the positive, 4 in the negative contrast of the NB2<>NB0 condition. Comparing the cross-correlation of the 24 ED vectors among eachother (Figure 2A-C) we found that (1) there is dissimilarity between the tasks compared against within-task, (2) a checkerboard-like structure indicating the independence of task-positive and task-negative subject ranking, (3) highest values for faces<>geometry against negative<>geometry (expected as most faces had a negative valence). In addition, the N-Back and Reward task were rather similar whereas the Hariri task was moderately related to N-Back and Reward, but unrelated to TET.
·(A) Reward anticipation task (B) N-Back-task with ROIs marked
·Cross-correlation of Eucliedian distance vectors across N=238 subjects
Conclusions:
Calculating the Eucledian distance of a subject in relation to a group is a useful tool to investigate, if features generated from task-based fMRI are independent or redundant. Task-negative responses (typical default mode regions) seem to carry an independent information, possibly with task-specificity.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI)
Methods Development
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems 2
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Affective Disorders
Computational Neuroscience
FUNCTIONAL MRI
Modeling
Psychiatric
Psychiatric Disorders
Statistical Methods
1|2Indicates the priority used for review