A Study on the Computer Diagnosis of Electrocardiograms:Differentiation of electrocardiographic patterns by means of linear discriminant functions

概要

論文の詳細を見る
The pattern diagnosis of electrocardiograms (EKG) by means of linear discriminant functions was undertaken with a digital computer. With X, Y and Z leads of Frank's scalar EKG, 27 parameters composed of amplitudes at 8 points of QRS apart at an interval of 10 msec and the maximal T amplitudes were manually measured. Forty-five EKG's consisting of three groups, Normal, left and right ventricular hypertrophy (LVH & RVH), -were used as material to derive a set of linear discriminant functions. In order to differentiate two groups from each other by means of a linear discriminant function : F=L1X1+L2X2+………+LnXn+C F : discriminant value X1, X2, …………Xn : amplitudes of QRS and T L1, L2, ……… Ln : discriminant coefficients C : constant A summation of the products of the measurements of the amplitudes, from X1 to Xn, and the discriminant coefficients, from L1 to Ln, respectively is made, then a constant C is added. The decision as to which group an EKG pattern belongs to is made according to whether the sum, that is, the discriminant value F is positive or negative. The discriminant coefficients, from L1 to Ln, and the constant C were derived by the computer in such a way that the formula will give the best discrimination. The derived formulae were fed with 79 EKG's which did not include the material-EKG's and consisted of the 3 groups of Normal, LVH and RVH, so that one could see whether formulae were able to make differential diagnosis of these 3 types of EKG patterns when applied to clinical series of EKG, S. A set of 27 dimensional linear discriminant functions derived from all the 27 parameters discriminated well any of two combinations out of the three groups from each other when applied to the series of material-EKG's, but not so satisfactorily when applied to the series of other 79 EKG's. This may have been due to the insufficiency of the number (about 15 for each group) of the material-EKG's in comparison with the size of the dimension (27 of the formulae. In order to overcome this crucial difficulty, a set of 8 dimensional linear discriminant functions was derived from the 8 parameters of each QRS of the 3 leads of X, Y and Z. Then, the three discriminant values calculated by this set of formulae with three QRS's of the three leads of the material-EKG's and the three maximal T amplitudes were used as parameters to derive another set of 6 dimensional linear discriminant functions. When the series of 79 EKG's were put into this set of formulae, there were given correct diagnoses to 16 out of 22 Normal EKG's, 17 out of 25 LVH ones and 26 out of 32 RVH ones, 59 out of 79 in total (75%). Based on these figures, it can be stated that the scheme of linear discrimiant functions works almost as good as average physicians in EKG diagnosis though not beating experienced specialists. Because of the simplicity of the computation, this scheme spends less computer time than other complicated but not much more capable programs for EKG diagnosis.
一般社団法人日本生体医工学会の論文

A Study on the Computer Diagnosis of Electrocardiograms:Differentiation of electrocardiographic patterns by means of linear discriminant functions

スポンサーリンク

概要

一般社団法人日本生体医工学会 | 論文

スポンサーリンク

A Study on the Computer Diagnosis of Electrocardiograms:Differentiation of electrocardiographic patterns by means of linear discriminant functions

スポンサーリンク

概要

一般社団法人 日本生体医工学会 | 論文

スポンサーリンク

一般社団法人日本生体医工学会 | 論文