Abstract
The work developed and presented in this thesis continues the research
line started with another PhD. thesis entitled ”Estudio de métodos para
procesamiento y agrupación de señales electrocardiográficas”. The algorithms
and methods used here have been developed to automatically process
the information contained within a special kind of long-time electrocardiographic
registers (called Holter ECG). The aim of the work is to
provide doctors and cardiologists with a complete set of tools that make
them easier the task of analysing and diagnosing the cardiac diseases. To
perform this, we will apply to the Holter ECG a clustering process in order
to automatically group the heart beats that compose the signal into
a very few clusters from where doctors, by means of the manual inspection
of a representative beat from each cluster, easily and quickly provide
a diagnosis. Finally, to achieve the clustering objective, we will study the
ECG morphological features using the large amount of databases available
through the internet, the development of new ECG signal applications,
the comparison among methods, and some algorithm optimization
tasks have been also performed. A polygonal approximation algorithm
used for ECG compression, the Principal Component Analysis (PCA)
scheme applied to the feature selection stage or heart beat modelling by
means of the Hidden Markov Models (HMM) for feature reduction too.
The application of all the methods described above has become an improvement
of the final Holter ECG clustering process.