Bridge router improves biosignal model accuracy by selecting pathways
BridgeRouter: Gain-Aware Instance Routing over Cross-Modal Bridge Portfolios
Artificial Intelligence
Summary
Many models trained on brain signals (EEG) don’t work as well when applied to other body signals like muscle or heart data without retraining. The authors created BridgeRouter, a system that chooses the best transfer pathway for each individual signal instance, improving detection accuracy without needing to retrain the model. BridgeRouter learns which path is likely to perform better per signal and uses that during application, leading to more accurate results on tasks like detecting freezing gait and classifying sleep from different signals.
What this means in practice
- •For medical device engineers: Improve wearable health devices by dynamically selecting the best EEG-based model transfer route to interpret other biosignals like EMG or ECG without retraining.
- •For signal processing software developers: Build biosignal interpretation tools that automatically route individual biosignal instances to the most accurate pretrained model bridge, enhancing prediction accuracy without extra labels.
Authors
Ruoxi Piao, Alicia Lu, Yunzhen Zhang, Mustafa Mısır, Hasan Onur Keleş
Abstract
Lightweight representation bridges make pretrained, task-adapted EEG foundation models executable on other biosignal modalities without retraining their backbones. However, selecting a single bridge pathway globally assigns every instance to the same transfer route and ignores complementary errors across routes. We therefore formulate multi-bridge transfer as an instance-level algorithm-selection problem and introduce BridgeRouter, a gain-aware router over a portfolio of frozen cross-modal pathways. A validation-selected Single Best Solver (SBS) serves as a protected default route. During training, task labels are used to compute the reduction in negative log-likelihood provided by each alternative route relative to the SBS, and a lightweight MLP predicts these route-specific gains from a target-modality representation extracted by a frozen encoder. At inference, BridgeRouter observes only the target modality, requires neither EEG, task labels, nor predictions from all candidate routes, and executes exactly one bridge pathway: the route with the largest predicted gain when that gain exceeds a validation-calibrated threshold, and the SBS otherwise. We evaluate seven EEG foundation-model bridge pathways for EMG-based freezing-of-gait detection on FOG and ECG-based sleep classification on ISRUC. Across five seeds, BridgeRouter achieves $81.38\pm1.11\%$ balanced accuracy on FOG, improving over the BIOT SBS by $11.49$ percentage points, and $62.92\pm2.17\%$ on ISRUC, improving over the CBraMod SBS by $3.83$ points. Per-instance bridge oracles reach $99.96\%$ and $78.61\%$, respectively, revealing substantial remaining portfolio headroom. These results show that anchor-relative gain prediction can convert complementary cross-modal bridge behavior into deployable instance-level routing improvements while retaining single-route inference.