Searchlight single subject significance

[Örjan de Manzano]

Dear All,

We have in the past done searchlight classification to obtain images for second level analyses. We however this time see some individual differences in the spatial distribution of representation and would therefore like to have a closer look on the single subject level. Is there a convenient way to implement permutation testing into the searchlight classification (supposedly for each searchlight) to understand which accuracy values are significant within each person? In principle, we would like add a significance threshold to the searchlight accuracy map from each person.

Best,

Örjan

[Nick Oosterhof]

Greetings,

On 28 Sep 2023, at 11:54, Örjan de Manzano <orjan.d...@gmail.com> wrote:



Dear All,

We have in the past done searchlight classification to obtain images for second level analyses. We however this time see some individual differences in the spatial distribution of representation and would therefore like to have a closer look on the single subject level. Is there a convenient way to implement permutation testing into the searchlight classification (supposedly for each searchlight) to understand which accuracy values are significant within each person? In principle, we would like add a significance threshold to the searchlight accuracy map from each person.

cosmo_montecarlo_cluster_stat provides such functionality using feature_stat = none. To use it for data from one participant, one has to compute a set of null maps for that participant. Cosmo_randomize_targets can be used to shuffle labels.

For TFCE multiple comparisons correction, it seems appropriate to convert the map values to z-scores. 

Does that help?