Selective MMIE training of hidden Markov models for cardiac arrhythmia classifi...

• Author(s): Lima, C. S. ; Cardoso, Manuel J.
• Date: 2007
• Persistent ID: http://hdl.handle.net/1822/17511
• Origin: RepositóriUM - Universidade do Minho
• Subject(s): Hidden Markov models; MMI training; Cardiac arrhythmia; Pattern recognition
• Document Type: Conference Object
• Language: English

Description

Centre Algoritmi This paper is concerned to the cardiac arrhythmia classification by using Hidden Markov Models. The types of beat being selected are normal (N), premature ventricular contraction (V) which is often precursor of ventricular arrhythmia, and two of the most common class of supra-ventricular arrhythmia (S), named atrial fibrillation (AF) and atrial flutter (AFL). The approach followed in this paper is based on the supposition that atrial fibrillation, atrial flutter and normal beats are morphologically similar except that the former does not exhibit the P wave, while the later exhibits several P waves following the QRS. Regarding to the HMM modelling this can mean that these three classes can be modelled by HMM’s of similar topology and sharing some similar parameters excepting the part of the HMM structure that models the P wave. This paper shows, under that underlying assumption, how this information can be compacted in only one HMM, increasing the classification accuracy by using MMI (Maximum Mutual Information) training, and saving computational resources at run-time decoding. This paper also shows that the similarities among normal, atrial fibrillation and atrial flutter beats, which main difference is the lack or repetitions of the P wave, can be taken into consideration to improve the classifier performance by using MMI training, in a single model/triple class framework, which is similar of having three different models sharing several parameters. The algorithm performance was tested by using the MIT-BIH database. Better performance was obtained comparatively to the case where one different HMM models each class when using MLE (Maximum Likelihood Estimation) training alone.