--- redirect_from: - "/features/markdown/preprocessing" title: |- Preprocessing Recommendation pagenum: 3 prev_page: url: /features/markdown/Speakers.html next_page: url: /features/notebooks/Download_Localizer_Data.html suffix: .md search: spatial perform preprocessing different realignment average csf activity motion e g using recommended currently agreed upon conventions naturalistic neuroimaging data here speakers share current recommendations luke chang our work standard normalization also basic denosing removing global multivariate spikes linear quadratic trends covariates centered squares derivative squared derivatives very cautious performing high pass filtering effects interested occur slower frequencies including mask helps lot reducing types physiological related artifacts typically apply smoothing depending question dont always step feature selection rarely searchlights instead parcellations allows us quickly prototype analysis ideas smaller numbers parcels n increase want greater specificity usually parcellation neurovault org collections comment: "***PROGRAMMATICALLY GENERATED, DO NOT EDIT. SEE ORIGINAL FILES IN /content***" ---
Preprocessing Recommendation

Recommended Preprocessing

There are currently no agreed upon conventions for preprocessing naturalistic neuroimaging data. Here different speakers share their current recommendations.

Luke Chang

In our work, we perform standard realignment and spatial normalization. We also perform basic denosing by removing global and multivariate spikes, average CSF activity, linear/quadratic trends, and 24 motion covariates (e.g., 6 centered realignment, their squares, derivative, and squared derivatives). We are very cautious about performing high-pass filtering as many of the effects we are interested in occur in slower frequencies. We find that including average activity from a CSF mask helps a lot in reducing different types of physiological and motion related artifacts. We typically apply spatial smoothing, but depending on the question we don't always perform this step. For spatial feature selection, we rarely use searchlights and instead use parcellations. This allows us to quickly prototype analysis ideas using smaller numbers of parcels (e.g., n=50) and then increase the number if we want greater spatial specificity. We usually use a parcellation that we developed based on meta-analytic coactivation using the neurosynth database.