Described herein is a method and system for adaptively processing ambulatory electrocardiogram (ECG) signals to reduce noise in the output signals. Principal component analysis (PCA) or independent component analysis (ICA) is used to transform multiple input ECG signals into principal or independent components in component space in accordance with reference signals. The transformation is dynamically adjusted in accordance with noise characteristics within the ECG signals. A subset of principal or independent components is selected in accordance with the noise characteristics determined from the ECG signals, from additional non-ECG related signals and from environment- related signals. The subset is processed in accordance with the noise characteristics. In one embodiment, an inverse transform is applied to transform the principal or independent components back to ECG signals with reduced noise from which ECG properties can more accurately be determined. In another embodiment, the selected subset of principal or independent components may have a beat detection or arrhythmia detection algorithm applied directly thereto to determine instantaneous heart rate values or atrial activity.
