BeatGraph: Self-Supervised Heartbeat Graphs for Infant ECG Representations from the Home Environment
BeatGraph learns infant ECG from heartbeat graphs and releases 3,408 hours of home recordings.
BeatGraph represents each 30-second infant ECG window as a graph of heartbeats instead of fixed patches that can split a beat. A shared beat encoder, temporal Transformer, and residual graph attention are pretrained by predicting masked-beat embeddings, then fine-tuned for sleep-wake detection, infant state, activity source, and affect. Macro-F1 improves by 0.076 to 0.158 over the strongest baseline on each task. The authors release 3,408 hours of single-channel ECG from 143 infants; the model scores 0.892 AUROC on ZZU-pECG and matches a leading self-supervised model under linear evaluation on adult PTB-XL.
- Each 30-second window is modeled as a graph of heartbeats.
- Macro-F1 rises 0.076 to 0.158 over the strongest baseline.
- Reaches 0.892 AUROC on the pediatric ZZU-pECG benchmark.
- Public corpus has 3,408 hours from 143 infants aged 3–11 months.
Full article278 words · extracted from arxiv.org · click to collapse
Electrocardiogram (ECG) foundation models typically tokenize the signal into fixed-length patches that ignore cardiac structure, so a patch may split a heartbeat and the number of beats in each patch shifts with heart rate. This matters most for infants, whose heart rates are higher and whose ECG differs from the adult, clinic-recorded 12-lead data these models are built on. A model for infant ECG should therefore reason about heartbeats directly rather than recover them from arbitrary patches. We propose BeatGraph, which makes the heartbeat its unit of representation, modeling each 30-second window as a graph of beats. A shared beat encoder embeds each heartbeat from its waveform and inter-beat intervals, a Transformer with positional encoding orders the beats in time, and residual graph attention layers relate every beat to every other before attention pooling yields a window embedding. We pretrain BeatGraph on our new corpus of unlabeled infant recordings by predicting masked-beat embeddings, then fine-tune it for each task. One backbone supports sleep-wake detection, infant-state classification, activity-source identification (infant- or caregiver-initiated movement), and affect recognition, improving macro-F1 over the strongest baseline on each task by 0.076 to 0.158. It also transfers across age groups, reaching 0.892 AUROC on the ZZU-pECG pediatric benchmark (ages 0 to 14), within 0.001 of the best published self-supervised ECG model, and matching that model under linear evaluation on the adult PTB-XL benchmark despite infant-only pretraining. Finally, to our knowledge, we release the first public infant ECG corpus collected in homes, classrooms, and laboratory settings with state and affect labels. It contains 3,408 hours of single-channel ECG from 143 infants aged 3 to 11 months, with unlabeled pretraining data, benchmark tasks, and subject-level splits.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.31546