Application of Convolutional Dendrite Net for Detection of Myocardial Infarction Using ECG Signals

Myocardial infarction (MI) is a sudden-onset medical emergency. Without timely diagnosis and treatment, mortality is very high. The mining of deep features increases the computational burden while improving the accuracy of MI detection. So, there is a trade-off between speed and accuracy. However, considering our own classification process, we are more inclined to integrate the logic between shallow features and deep features. Therefore, this article is dedicated to transforming the high computational cost deep features into low computational cost shallow feature logic, so as to improve the speed without affecting the accuracy. The method proposed in this article mainly includes artifact removal, waveform location, coding, and classification. Hilbert curve completes the coding, which efficiently converts the time-domain features of electrocardiogram (ECG) signals into image features through a double recursive algorithm. Convolutional Dendrite Net (CDD Net) realizes the classification of images by extracting the logical relationship between features. The method is verified by Physikalisch Technische Bundesanstalt (PTB) dataset. The results show that the average detection accuracy and time of one patient are 98.95% and 2.09 ms, respectively.