Automatic Electrocardiogram Interpretation

The goal of the work behind this thesis was to design and partially implement an automatic system for the analysis of electrocardiograms. The thesis describes an original model of the human ECG analysis process which demonstrates that it breaks naturally into three stages: reduction, labelling and diagnosis. It is shown that to make a correct diagnosis, each of these stages must be both expressive and complete. A review of the literature reveals that what had previously been perceived as a fundamental limitation of the machine diagnosis problem is, in reality, a lack of expressiveness at the reduction stage. Original algorithms to overcome this problem are described for the reduction and labelling stages, with examples of their implementation being given. Future work on the diagnosis stage is also suggested and discussed