Lecturer: Dr. Moritz Wurm
Student: Maximilian Herrmann
Date: 08th May 2025
Interpretation of the cross-classification data alaysis
The task was to decode for the video session in the 4 ROIs (MTG, SPL, PMC, IFG) using the fMRI_dataset2. The dataset provided data from different brain regions and their response to person-directed or object-directed targets in a study.
Then I plotted the results and computed t-tests.
Below are the plotted results:
In the cross-classification plot, you can see 4 ROIs on the x-axis:
- MTG (middle temporal gyrus)
- SPL (superior temporal lobe)
- PMC (premotor cortex)
- IFG (inferior frontal gyrus)
On the y-axis you can see the accuracies for the two different tasks:
- person-directed vs non-person-directed across object-directedness
- object-directed vs non-object-directed across person-directedness
Results
- person-directed: significant decoding above chance only in MTG, SPL and PMC at the border(very slightly above chance)
- object-directed: all brain regions seem to decode above chance, suggesting stronger decoding in this context
Interpretation
- MTG:
- MTG decodes both person-directedness and object-directedness when holding the other factor. Could show that decodes for abstract concepts as “social” vs. “non-social” and for “object” vs. “non-object”. independent of low-level visual input.
- IFG:
- does not decode “social” concepts as the MTG, but generalizes slightly above chance for object-directedness,
- SPL & PMC:
- only show robust above-chance decoding for object-directedness and no significant above-chance for person-directed
- seems not to decode higher “social” concepts, but rather spatial movement itself
Conclusion
Our cross-context decoding demonstrates that these regions go beyond remembering single videos, but also form a concept of an action category, so they abstract information.
When training a classifier, it shows that the underlying brain structures reflect high-level properties such as “social” and “non-social” and “object-directed” and “non-object” rather than only low-level features (single details).
In other words, MTG, IFG, SPL and PMC abstract from single videos to represent meanings put into categories.