Framework Classifies Heart Rhythms on Wearable Devices
A split federated learning method runs arrhythmia classification on small client-side networks while keeping patient ECG data on each device.
Updated on Nov. 4, 2026 in Heart Disease
Researchers proposed FedIoMT-RPM, a split federated learning framework for ECG arrhythmia classification. Each device trains a 28.4 KiB client-side network, and the framework reported 93.97 ± 1.72% accuracy under a patient-independent evaluation protocol.
Why it matters
Wearable electrocardiogram monitoring requires classifiers that run on resource-constrained devices and learn across hospitals without pooling patient data. The framework targets both constraints within the Internet of Medical Things.
The framework reported a Matthews correlation coefficient of 0.731 ± 0.039 and 64.8% sensitivity for supraventricular beats, based on 1,381 of 1,781 test beats from a single record. Fusion-beat detection remained near zero, and generalization beyond the evaluated datasets is not yet established.
The details
Each device trains only a client-side network, and bounded adaptive weighted aggregation adjusts each client's contribution according to its smoothed validation performance. Reconstruction attacks recovered the ECG from transmitted activations, while membership inference stayed near chance.
Timeline
• The framework was reported on November 4, 2026.
Health Landscape
The framework builds on SplitFed, the split federated learning baseline used to train models across hospitals without centralizing patient records. By shrinking the client-side network and lowering communication against that baseline, it moves the approach closer to deployment on wearable monitors.
Wearable ECG monitors that run classifiers locally could flag arrhythmias without sending raw heart-rhythm data off the device. Any clinical use would depend on validation beyond the current benchmark results, including improved detection of fusion beats.
The takeaway
Privacy-preserving training can reduce device burden for heart monitoring, but detection gaps for certain beat types remain before the approach is ready for patient care.
Further reading
Explore more on Heart Disease for context on arrhythmia monitoring and cardiac care.
Source note: This article includes information reported by Nature.
