Neural network model for recognizing long and short QT syndrome on ECG

Automatic detection of long and short QT integral syndrome is an extremely urgent task. Long and short QT syndromes are predictors of life-threatening cardiac arrhythmias. Manual and automatic (based on classical signal accumulation algorithms) determination of the QT interval length does not always allow determining the presence of pathology. A hybrid model for determining long and short QT syndrome is proposed, which combines the use of classical algorithms for automatic detection of the expected location of the QT complex with subsequent classification of these complexes to search for pathologies using artificial neural networks. We proposed 2 architectures (convolutional and recurrent) of neural networks and achieved accuracy for neural network models by the metric F1 = 0.98-0.99. When classifying ECGs, we used an approach called Single Lead First. It consists in the fact that the neural network does not accept all 12 leads at once, as a single image, but analyzes each complex in each lead independently. This allows to reduce the dimensionality of the problem, simplifying the training, and also to increase the volume of the training sample several times. Further research will be aimed at analyzing the late potentials of the ventricles on high-resolution ECG, which are a predictor of a number of serious heart disorders, such as sudden death syndrome, myocardial infarction, ventricular tachyarrhythmia.