## Document Type

Thesis - Open Access

## Award Date

1971

## Degree Name

Master of Science (MS)

## Department / School

Electrical Engineering

## Abstract

Considerable use is presently being made of computers to classify electrocardiograms (ECGs) into various heart defect categories. An example of such a program is the "Electrocardiographic Analysis Program" developed jointly by the Mayo Clinic and IBM. In one of Dr. Caceres published reports, he tells of using a computer program which involves 5,000 instructions. Dr. Caceres does not say whether this is a commercial program or one of bis own. His description does place this program in the same general category as IBM's. These programs have the common criteria of examining the ECG for the same information that a cardiologist uses to analyze an ECG. This necessitates using many instructions and large memory storage facilities; thus, a large computer is necessary. It would be most desirable to have a computer method of analyzing ECGs that would utilize a small special purpose computer and eliminate the use of the large expensive computer. A second choice would be a small special purpose computer capable of sorting the cardiograms into the categories of normal and possibly abnormal. The possibly abnormal ECGs would then be analyzed on the large computer for diagnostic purposes. Bailey, in his thesis, examined the possibility of changing the ECG from a time-varying plot to a statistical plot not involving time. This probability density function was analyzed to determine the ECG category. He was able to show that correlation exists between the probability density function and the ECG classification of normal versus possibly abnormal. Young and Huggins tried approximating the ECG as a twelve dimensional vector utilizing the orthonormal exponentials as the basis vectors. The coefficients of these basis vectors were used to categorize the ECGs. Seventy-five percent correct diagnosis was obtained on 65 cardiograms; 53 of these cardiograms comprised the training set. It is of interest to repeat what these authors call a fundamental assumption justifying this type of approach. "Since the physicians are able to distinguish different pathological categories from the similarities and the dissimilarities of ECG waveforms, and since the waveform corresponds to direction in signal space, it is logical to conclude that the signal space may be separated into several subspaces, each subspace corresponding to a pathological category. A transformation may be found to relate the subspaces with the pathological categories. The ECG signal vector which falls into a certain subspace ray then be considered as belonging to that category." This thesis utilizes the above fundamental assumption and attempts to assign ECGs to one of two classes (normal or possibly abnormal) on the basis of examples given for each class; thus, a pattern recognition problem is the result. To generate the basis vectors that span the signal space, the Hotelling Method is used. This method utilizes a linear transformation to obtain orthogonal basis vectors (the uncorrelated random variables of statistics). The vectors are generated so that the first vector yields the largest possible variance; the second vector yields the next largest possible variance; etc. Variance is a direct measure of the amount of information contained in the vector. | Mattson and Dammen were able to segregate voice patterns by using only the first vector obtained by the Hotelling Method. The theory used by these gentlemen to acquire the vector was slightly different from Hotelling's approach; their theory produced only the vector yielding the largest variance. A better classification might have resulted if more than one vector had been considered.

## Library of Congress Subject Headings

Electrocardiography

Computers

## Format

application/pdf

## Number of Pages

78

## Publisher

South Dakota State University

## Recommended Citation

Schwartz, Edward G., "Screening Electrocardiograms Using Hotelling's Method of Principle Components" (1971). *Electronic Theses and Dissertations*. 5298.

https://openprairie.sdstate.edu/etd/5298