抄録
A learning machine "adaline neuron" was employed for automatic diagnosis of the vectorcardiogram and the electrocardiogram. The frontal circle and the horizontal circle were divided into 480 meshes. The features were expressed by a binary digit, whether the vector loops passed through each mesh or not. In a part of the trials, 5 sets of binary digits were applied in addition to QRS duration and direction of inscription of QRS loops and T loops. In this trial a total of 490 meshes were used. Vectorcardiograms were taken by FRANK'S method in 235 cases. Several methods of adaline usage were tried. The best result was obtained so far by successive dichotomies based on the principle of the logical decision tree. First the normal patterns and the abnormal patterns were divided. The correct ratio was 96.5% when the 490 meshes were employed, cases of an output value within ±10 units being regarded as undecided. Next the abnormal cases were divided into two groups depending on whether the QRS duration was more than 0.12 seconds or less. The group of cases with a QRS duration of less than 0.12 seconds was divided into right ventricular hypertrophy and others. The correct ratio was 98.6%. The remaining cases were divided into left ventricular hypertrophy and myocardial infarction, the correct ratio being 88.8%. The group of cases with a QRS duration of more than 0.12 seconds was easily divided into complete left and right bundle branch block in all cases. Here the number of meshes could be decreased to 59 meshes without changing the accuracy appreciably. This attempt showed that the application of the adaline for automatic diagnosis of the vectorcardiogram and the electrocardiogram is promising.
