Processing of Cardiograms for Pattern Recognition.
Abstract
Conventional 12-lead electrocardiographic signals must be subjected to noise removal and data reduction schemes before pattern recognition techniques may be applied for automated diagnosis. A batch-processing and an adaptive recursive filtering scheme are developed and studied experimentally. A physical model for the generation of cardiographic signals is exploited to study the relationship of the standard electrocardiogram (ECG), to the three-channel vectorcardiogram (VCG) using the Frank-Orthogonal-lead system. A linear time-invariant relationship between the ECG and the VCT is postulated on the basis of the physical model. Estimation of the VCG from the ECG, based on the various models, is studied experimentally over an ensemble of nine patients. The results indicate that on an individual basis the models give quite good results, but that a single satisfactory fixed model valid for the whole ensemble of patients is not specifiable. Factor Analysis and the Karhunen-loeve expansion are then considered as alternatives to preliminary data reduction of the ECG by estimation of the VCG. A signal processor that can be easily implemented on a minicomputer and that reduces the full set of 12-time signals (the ECG) to a set of a few constant parameters (the pattern vector) is determined.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1974
- Accession Number
- ADA028852
Entities
People
- Adnan Akant
Organizations
- Charles Stark Draper Laboratory