Project 1

Efficient DSP for Cardiology Applications


  • Noel Walsh

Current Work

The ECG signal is a biomedical signal measured on the bodies surface above the heart to describe its electrical activity. The waveform is characterised by a P wave, QRS complex and a T wave indicating the depolarisation and re-polarisation of various heart muscle during each heart beat.  This signal allows cardiologists to non-invasively, examine a patients heart to determine if any abnormalities are present. A “typical” heart beat is shown in Figure 1.



Figure 1. Illustrative diagram of heart beat, showing major morphological elements.

My project is concerned initially with the compression of ECG signals, and will also examine the automatic processing of ECG in order to extract information about e.g. possible heart problems. Of particular interest is efficient processing in an ambulatory environment. A high-level block diagram of the various elements is shown in Figure 2.

Figure 2. High-Level Block diagram of ECG Processing System

The main focus of the research is concerned with the three final stages, feature extraction, classification and compression. Although these areas have been extensively researched, comparatively little work has been carried out from the point of view of low complexity and power consumption with portable devices.

The three main stages in the system require signal processing, and the methods which we hope to implement are all based on the wavelet transform. For feature extraction and classification the wavelet transform produces a time-frequency decomposition of the signal which separates individual signal components more effectively than traditional methods. Its advantages for compression methods allow for good data de-correlation to highlight the redundancy in the waveform.

At present the research has mainly focused on compression of ECG with various state of the art methods implemented. The two main methods studied were the set partitioning in hierarchical tress (SPIHT) and the embedded zero-tree wavelet (EZW) method. The SPIHT and EZW algorithms are also based around the wavelet transform and the two major strengths of these are:

1)      The bit-stream is embedded, therefore the coefficients are sent in order of importance to the decoder, allowing the encoding process to be stopped at any stage.

2)      It exploits the similarity of the coefficients of similar orientation between the subbands of the wavelet transform to provide greater compression.

These advantages allow variation of the compression rate of the signal depending on the final signal quality required. The graph in Figure 3 shows the performance of SPIHT in relation to compression ratio and percentage root-mean-square difference (PRD); as the compression ratio increases the PRD (signal distortion) also increases.



Figure 3. Reconstruction Fidelity (PRD) vs. Compression Ratio for SPIHT

Implementing the SPHIT and EZW on portable devices has some drawbacks, in particular, they require complex memory structures and sorting/list procedures, increasing the complexity for the systems. To overcome these complications, I am currently testing a method which I have developed to reduce the memory structures required, while still providing an adaptable algorithm that could compete with these methods.

To date the proposed algorithm has been tested on simple data sets to verify its logic and behaviour, the results so far have been favourable and future work will be carried out to compare its performance against SPIHT and EZW on ECG data. Further tests will be carried out to also determine the effect of normal and abnormal heart conditions on the algorithm.

Once the compression stage has been finalised and fully tested the research will follow on to develop methods for both feature extraction and classification.