Supporting data for "DUNE: a versatile neuroimaging encoder captures brain complexity across three major diseases: cancer, dementia and schizophrenia"

This dataset contains the supporting code, models, and processed data for DUNE (Deep feature extraction by UNet-based Neuroimaging-oriented autoEncoder), a neuroimaging workflow that transforms multisequence brain MRI scans into standardized compact patient-level embeddings. <br>DUNE uses a UNet-based autoencoder trained on 3,814 brain MRI scans from healthy volunteers and glioma patients to generate low-dimensional embeddings that capture clinically relevant patterns across multiple neurological conditions including cancer, dementia, and schizophrenia. <br>The dataset includes model weights, preprocessing scripts, feature extraction code, and evaluation results demonstrating the model's ability to predict clinical variables, molecular features, and survival outcomes across multiple independent validation cohorts.