What problem does it solve? Running fMRI deep learning research requires juggling multiple model implementations, preprocessing pipelines, and training configurations. This Skill unifies the full lifecycle of the NeuroSTORM platform — from raw .nii files to trained checkpoints — across 8 models and 3 input modalities under one entry point. ## Core Features & Use Cases - Multi-model training: Pretrain (MAE or contrastive) and fine-tune NeuroSTORM, SwiFT, BrainGNN, BNT, LG-GNN, Com-BrainTF, IBGNN, and BrainNetCNN via a single main.py interface. - Full preprocessing pipeline: Convert MNI152-space fMRI volumes into PT or H5 tensors, extract ROI time series with atlases like CC200, and compute functional connectivity matrices. - Flexible inference: Run single-subject or batch dataset inference for age regression, gender classification, phenotype prediction, and disease diagnosis. - Use Case: A researcher downloads HCP1200 data, preprocesses it into 4D voxel tensors, MAE-pretrains NeuroSTORM, then fine-tunes it for sex classification and evaluates on a held-out test split. ## Quick Start Ask the assistant to preprocess your HCP fMRI dataset and fine-tune the NeuroSTORM model for gender classification using the pretrained MAE checkpoint.
