Abstract
Validity of Fourier coefficients as parameters for the electrocardiographic pattern recognition by means of the threshold logic units and the maximum likelihood method was verified. As the traditional methods of Fourier transform required much time, a new method of fast Fourier transform of folding type, which differed from FFT of Cooley and Tukey, was developed. Sampling points used in the present investigations totaled 100. Using scalar ECG of Frank's orthogonal system (X, Y and Z leads) 15 normal subjects, myocardial infarction cases and ostium secundum type of atrial septal defect cases respectively were analyzed by this new method, and the coefficients such as An, Bn, √An2+Bn2 and Arctan (An/Bn) were obtained.√An2+Bn2 was considered as the most useful parameter among them, because it is invariant for the parallel shift of amplitude and time axis.
Normals, myocardial infarction and atrial septal defect cases were classified, in 100 per cent success, by threshold logic units and maximum likelihood method in the twelve dimensional information space, respectively using the initial four Fourier coefficients of the X, Y and Z leads as parameters.
Fourier coefficient parameters were estimated by t-tests, and then the effort was made to reduce the numbers of parameters of information space.
Complete separation was also obtained even when the number of dimension were reduced to seven by using the initial seven Fourier coefficients of X lead alone. This suggested that the important diagnostic information was mostly contained in X lead, and also that the lower harmonic components had the greater part of the information of ECG. Nonlinear discriminant function played an important role in the low dimensional information space.
