ECG arrhythmia live demo

Pick a real heartbeat below — each one comes from the MIT-BIH database, from patients the model never saw during training. The 1D CNN classifies it in your browser; nothing is uploaded.

Educational demo — not a diagnostic tool. This classifies pre-recorded research heartbeats to show how a model reads an ECG. It cannot read a live ECG, and it is not a substitute for a clinician or an approved cardiac monitor. If you have symptoms or a heart-rhythm concern, seek medical care.

Pick a heartbeat

Beats are labelled by their true class (as annotated by cardiologists): N normal, S supraventricular, V ventricular, F fusion. Watch what the model gets right — and where it slips.

What am I looking at?

Each trace is a single heartbeat: a ~0.7-second window of the ECG signal centered on the beat's sharp peak (the R-peak). The model was trained only on the shape of these windows. That's why it reads ventricular beats — which look visibly different — but struggles with supraventricular beats, whose shape is nearly normal and whose real giveaway is timing the model can't see. Fixing that is exactly what the next iteration tackles.