Ecg beat classifiers adapt to new patients using normal-anchored geometric transport
Transport What Shares the Conduction Pathway: Normal-Anchored SPD Geometric Transport for Source-Free Patient Adaptation of Holter and Patch ECG
Machine Learning
Summary
ECG machines that classify heartbeats often struggle when used on new patients or different devices because they rely on data they were originally trained on. The authors introduced a method that adapts these classifiers without needing labeled data from new patients, by using the normal heartbeat patterns in the first few minutes of each recording. This approach improved accuracy in recognizing abnormal heartbeats across many recordings and devices. Their method is quick to apply and performs better than several existing adaptation techniques.
What this means in practice
- •For medical device developers: Improve ECG-based arrhythmia detection on new patients without requiring labeled adaptation data by applying normal-anchored geometric transport techniques.
- •For healthcare data engineers: Increase accuracy of arrhythmia classification in large-scale patch and Holter ECG datasets from diverse patients using fast, label-free patient adaptation.
Authors
JinKook Kim
Abstract
Electrocardiogram (ECG) beat classifiers lose accuracy on new patients and devices, and source-free adaptation methods that assume class-balanced target batches fail on real recordings, whose label-free prefix is 87.9--98.6% normal beats. We propose Normal-Anchored Geometric Transport (NAGT, GA_NS), which estimates a closed-form symmetric positive-definite congruence map from label-free normal anchors in the first 5 min of each recording and applies it only to the normal and supraventricular prototypes, keeping the ventricular and fusion prototypes fixed. The score offsets and decision rules were set on MIT-BIH only; the rate-invariant RR head (relA) was adopted after the development databases (INCART, SVDB) had been opened; the frozen model was then evaluated once on Icentia11k chest-patch recordings under a pre-registered protocol. In 493 patients (973 recordings, 3,615,388 evaluated beats), NAGT reached an arrhythmia macro-$F_1$ $J = \frac{1}{2}(\text{S-}F_1 + \text{V-}F_1) = 48.7\%$ [95% CI 42.1, 54.7] versus 34.5% without adaptation ($ΔJ = +14.1$ [+8.6, +19.3] pp; 59.8% versus 43.9% in non-AF) and outperformed all five Holm-adjusted comparators, including AdaBN, Tent and T3A. In the 75-patient test subgroup of the historical split (73 analyzed), the gain was not significant (+6.1 [-8.5, +23.4] pp). Adaptation takes a median of 5.8 ms per record without back-propagation.