Electrocardiogram analysis based on the Karhunen-Loève Transform

Karhunen-Loève Transform (KLT) is the statistically best block transform. In terms of decorrelation of the signal samples and repacking energy distributing, the signal dependent transform of KLT was used in noise eliminating, data compression and features extraction for ECG. Experimental results proved that KLT was proper for suppressing noise of low self-correlated property, e.g. white noise, but has little use for noise of high self-correlation such as power line interference. KLT got high compression ratio at the cost of low information fidelity, though the coding procedure might remove certain noise. KLT was more remarkable for its role in feature extraction, e.g. to distinguish between normal sinus beats from abnormal waveform morphologies. In conclusion, KLT is well suited in ECG processing of noise canceling and data compression, especially a good candidate for robust feature extraction.