Metadata-Driven Federated Learning of Connectional Brain Templates in Non-IID Multi-Domain Scenarios

TL;DR

This work proposes a metadata-driven federated CBT learning framework, called MetaFedCBT, which overcomes the non-IID issue of existing methods by generating informative brain connectivities for privacy-preserving holistic CBT learning with guidance using metadata.

Abstract

A connectional brain template (CBT) is a holistic representation of a population of multi-view brain connectivity graphs, encoding shared patterns and normalizing typical variations across individuals. The federation of CBT learning allows for an inclusive estimation of the representative center of multi-domain brain connectivity datasets in a fully data-preserving manner. However, existing methods overlook the non-independent and identically distributed (non-IDD) issue stemming from multidomain brain connectivity heterogeneity, in which data domains are drawn from different hospitals and imaging modalities. To overcome this limitation, we unprecedentedly propose a metadata-driven federated learning framework, called MetaFedCBT, for cross-domain CBT learning. Given the data drawn from a specific domain (i.e., hospital), our model aims to learn metadata in a fully supervised manner by introducing a local client-based regressor network. The generated meta-data is forced to meet the statistical attributes (e.g., mean) of other domains, while preserving their privacy. Our supervised meta-data generation approach boosts the unsupervised learning of a more centered, representative, and holistic CBT of a particular brain state across diverse domains. As the federated learning progresses over multiple rounds, the learned metadata and associated generated connectivities are continuously updated to better approximate the target domain information. MetaFedCBT overcomes the non-IID issue of existing methods by generating informative brain connectivities for privacy-preserving holistic CBT learning with guidance using metadata. Extensive experiments on multi-view morphological brain networks of normal and patient subjects demonstrate that our MetaFedCBT is a superior federated CBT learning model and significantly advances the state-of-the-art performance.

Figures (7)

• Figure 1: Illustration of heterogeneous multi-view brain connectivities drawn from different hospitals (i.e., domains) for federated learning of holistic CBT across multiple domains, each marked by a corresponding color. To address the non-IID issue associated with the source domains (circled by the solid lines), we propose a data-generation federated learning scheme, where the meta-domains marked by the dashed circled lines are generated to shift the locally learned CBTs towards a more centered template across all federation domains.
• Figure 2: An overview of MetaFedCBT. (a) indicates how metadata is incorporated into FL using the metadata-driven connectivity generators $\texttt{MCG}_k$ and metadata regressors $\texttt{MR}_k$ for improved performance. For each hospital, we predict its metadata $\textbf{m}$ with a metadata regressor (c) from corresponding network residual weights between the local hospital and the server during the FL. The statistical property is then employed to determine a symmetric Gaussian noise matrix $\mathbf{N} \sim \mathcal{N}({\mu},{\sigma})$ to increase disturbance on the original multi-view dataset $\mathcal{T}$. This facilitates the generation of a meta-domain $\mathcal{M}$ through a metadata-driven connectivity generator (b) for each hospital. Thanks to the high-quality meta-domain generated in a supervised manner, our MetaFedCBT effectively improves the FL of CBTs.
• Figure 3: Illustration of our unsupervised (a) and supervised (b) connectivity generation methods. Through adjusting the distribution parameters of a noisy matrix, our metadata-driven connectivity generator (b) achieves a guided brain connectivity generation.
• Figure 4: Quantitative comparison of the centeredness of CBTs given by MetaFedCBT, the comparison models, and the ablated versions. The averaged Frobenius distance of all folds is adopted for evaluation. We also report the $p$-values using a two-tailed paired t-test between MetaFedCBT and each comparison method. If the $p$-value is less than 0.05/0.01/0.001/0.0001, it is flagged with one/two/three/four stars (*/**/***/****), respectively.
• Figure 5: Visual comparison of CBTs given by different models. Experiment was performed on the fold one of ASD-LH dataset.