Brain connectivity model improves prediction across diverse fMRI tasks
CoFlu-Brain: An fMRI Connectivity Foundation Model with Cofluctuation-Based Pretraining
Neural and Evolutionary Computing
Summary
Brain scans often analyze connections between regions by looking at average data, but this misses changes over time. The authors created CoFlu-Brain, a model that uses detailed time-based brain activity patterns to better capture how brain regions work together. They trained it on lots of brain scan data and found it works well for predicting tasks like classification and age estimation from brain scans. This approach performed better than other models on many tests.
What this means in practice
- •For medical imaging teams: Provide improved brain connectivity features for clinical diagnostic models using fMRI data with better representation of time-varying brain interactions.
- •For health data engineers: Integrate pretrained brain connectivity encoders to enhance demographic and clinical prediction workflows across multiple fMRI datasets.
Authors
Ziteng Sui, Junfeng Xia, Wenhao Ye, Mo Wang, Quanying Liu
Abstract
Functional magnetic resonance imaging (fMRI) commonly characterizes large-scale brain organization through functional connectivity (FC). However, an observed FC matrix is not a fixed brain graph but a statistical estimate determined by the finite BOLD frames used to compute it. Conventional FC-based approaches typically represent each scan with a single Pearson-correlation matrix, leaving frame-resolved variations in regional interactions unexploited. To address this limitation, we introduce CoFlu-Brain, which organizes framewise cofluctuation observations into multiple retained and dropped views, each summarized as an FC matrix. The model learns connectivity structure reproducible across retained views and predicts the representations of the corresponding dropped views, while adopting SIGReg from LeJEPA to prevent representation collapse. We pretrained CoFlu-Brain on 13,444 subjects from eight dataset groups and evaluated its representations via frozen-encoder linear probing on seven classification and two age-prediction tasks across seven datasets. CoFlu-Brain showed strong transfer across demographic, developmental, and clinical prediction tasks, outperforming the evaluated released models on most tasks. Random frame partitioning improved aggregate transfer, while the complete objective achieved the strongest overall performance. Code is available.