MULTIGROUP LOGISTIC CLASSIFICATION OF ELECTRO CARDIO GRAMS

Abstract

The electrocardiogram is animportant tool in the diagnosis of the diseases of the heart. With the debut, of computers, the .decision making in the process of disease classification has become simpler and straight forward. A -computer. software program is synthesized to identify the parameters of. an ECG Data Base and classify the data base amongst available disease classes. ",- There are two main approaches in the classification methods. The heuristic or deterministic method, employs a decision tree which.•ends up in a particular class, based on the heuristic knowledge. The second one, statistical method considers a broader spectrum of probabilities in which .;_the particular ECG data can be-classified Further, in statistical approach logistic classification employs logarithmic distributionof probabilities, and thus a wider range of classification groups can be accommodated in the system._ The -probability•of misclassification can •be minimized by assigning the parameters to a set having highest posterior probability. The present work capitalizes- on the 'advantages of logistic: classification . ,with- some heuristic features justifiably;--. -i.ncorporated..The results of the blending of.the two approaches has_. better. accuracy, -a higher .versati-lity and a less degree of misclassification. The computer program so developed here, uses a three dimensional spatial velocity for the detection'of the QRS complex. The locations of the onset and offset of the waves in individual leads of a 12 lead-simultaneous ECG data base are then determined. Thus mutual dependence of the wave features and the simultaneity in the onsets and offsets is used for abetter judgment.