Another example is the heart rate detection which is estimated after detecting the QRS complex from the beat sequence.Figure 2.ECG typical elementary waveforms and intervals.A variety of methods such as adaptive filtering [11], singular value decomposition (SVD) [12], independent component analysis (ICA) [13], neural networks [14], wavelet transform [15] have been introduced in the previous literature for ECG signal denoising. However, almost all these methods are not suitable for wearable sensor ECG online filtering in body sensor networks. Adaptive filtering and SVD method are simple and fast in operation, but they inherently suffer from the effects of noise residuals. Wavelet transform and ICA are effective in eliminating the common ECG noises, yet both of them require high computational complexity, which is rather challenging to achieve in real-time ECG sensor filtering.

Neural Networks require a large amount of time for ECG noise training and are not robust enough for individualized ECG filtering. In comparison, the ICBS filter has the advantages of fast denoising on ECG sensor nodes as well as the satisfactory filtering outcome.In ECG signal segmentation and classification, Hamid and Dana [16] have proposed the local trigonometric basis technique for ECG signal segmentation. However, the algorithm’s corresponding criterion is not universally suitable for the variable ECG morphological structure of interest. Macfarlane and Lawrie [17] sought to segment the characteristic QRS complex first, and then tried to segment the P subwave and T subwave besides this complex segment.

This method is effective in implementation, yet not robust enough for ECG signals with muscle noises and electrode motion artifacts. Some other methods like heuristic rules [18], wavelet transform [15], neural networks [5], Kalman filtering [19] have been proposed. However, the HMM-based algorithm outperforms Drug_discovery those methods in ECG feature extraction and classification in Body Sensor Networks. The reasons are given below: (1) HMM preserves the ECG structure��a typical ECG waveform usually consists of P subwave, QRS complex, T subwave and isoelectrics between these waves in a typical cardiac cycle [4]; (2) due to the fact that HMM is a probabilistic model rather than a heuristic model, it is more adaptive to dynamics [10]; (3) HMM is low in computational complexity in comparison with some commonly used methods like neural networks. Despite the merits of HMM in cardiovascular signal classification, one-layer HMM is merely capable of segmenting and classifying the ECG data [10].