What problem does it solve? Researchers analyzing fMRI data often need to run multivariate pattern analysis, ROI-level statistical tests, or voxel-wise decoding maps, but wiring together scikit-learn, statsmodels, and Nilearn pipelines with correct cross-validation and multiple-comparison correction is error-prone. This Skill provides a unified CLI that runs all three decoding modes with consistent inputs, outputs, and reproducible run manifests. ## Core Features & Use Cases - ROI MVPA: Trains cross-validated classifiers or regressors on ROI/parcel feature CSVs, with scaling and feature selection kept inside cross-validation folds. - ROI-wise GLM: Fits mass-univariate general linear models across ROIs using an explicit design matrix, reporting effect, standard error, P value, and Benjamini-Hochberg FDR-corrected Q value per ROI. - Voxel-wise SearchLight: Runs Nilearn SearchLight decoding over aligned NIfTI images within a group mask, producing a voxel-wise score map (searchlight_scores.nii.gz). - Use Case: Given a CSV of ROI features and subject diagnoses, run the mvpa mode to get cross-validated SVM classification metrics, then run roi-glm to identify which ROIs show significant group effects after FDR correction. ## Quick Start Ask the assistant to run ROI MVPA decoding on your roi_features.csv with subject_id as the subject column and diagnosis as the target using 5-fold cross-validation.
