gradecc

Brain Large-scale manifolds (Gradients) in a Motor Reinforcement Learning task.

Task

Dataset

We had 46 subjects in functional MRI during a Motor Reinforcement Learning. Stored as different files in RL_dataset_Mar2022/.

Seven subjects are remove due to behavioural issues. It's marked in ./data/participants.tsv. Also, subject SH1 excluded for not having subcortical data. Total of 8.

Atlas

Cortical atlas is stored in ./data/Schaefer2018_1000Parcels_7Networks_order.

Time-series extraction

Cortical by Dan Gale. Subcortical/Cerebellum by Corson Areshenkoff.

Epoch

Time-periods during the task as follows. Each epoch is set to be 216 time-trials of each ~ 2 seconds. Other time-trials dismissed.

• rest Subject is not doing the task. 297 trs. First 3 trs dismissed.
• baseline Subject is doing the task but no reward is given. 219 trs. First 3 trs dismissed.
• learning Subject starts getting rewards. 619 trs. Divided into early and late sections to differentiate learned period.
  • early When subject starts knowing how the task has changed. First 3 trs dismissed => 3:219 trs.
  • late When some subjects got it right. The last 216 trs.

Analysis

• Correlation matrix by Nilearn

• Gradient analysis by Brainspace

  • measure Any value for a brain region. For example, value for gradient 2 on for 7Networks_LH_Vis_3.
  • eccentricity Euclidian distance to the center of PCA space. Sum of top 3 or 4 gradient components squared.

• Behavioural analysis. Based on task scores.

Statistical analysis

• Pairwise t-tests

• Repeated-measures ANOVA by pingouin

  After including subcortical regions in gradient analysis, number of significant regions decreased from 57 to 50. No significant regions found in subcortex.

• False discovery rate (FDR) correction by Benjamini-Hochberg method

Post-hoc analysis

Seed connectivity of Regions of interest. Comparing shifts in functional connectivity pattern.

Plots

• Connectivity matrix
• Gradients
• Statistics
• Seed connectivity
• Behavioural