Model-based multichannel diagnosis of cardiac conduction abnormalities

A model-based method for real-time ECG diagnosis has been formulated in which information from multiple leads is processed simultaneously with a set of multivariable, linear, time-invariant filters in parallel. A diagnosis is reached by examining the filter outputs (residuals) using statistical criteria. Individual criteria diagnostics are fused into a single, final diagnostic. The methodology is applied to the discrimination of complete RBBB and LBBB from normals, using the QRS complex from 30 independent cases of a standard database. The cases are assigned randomly into a design set and a testing set of 15 cases each. Correct diagnosis assessment is 100% for the design set, and 93% for the testing set. The method is extendible to other types of cardiac abnormalities and to multiple, distinct physiologic sensing devices