Abstract: In order to improve the automatic diagnosis performance of left and right bundle branch block based on electrocardiogram(ECG)， an ensemble learning method was proposed，while a combination of multi-lead electrocardiogram and convolution neural network model served as the basic learner. Firstly， effective diagnostic lead data is extracted from the clinical 12-lead synchronous static electrocardiogram and divided into slices of multi-lead single heart beat data. Secondly， the bootstrapping method is used to extract multiple data subsets. Each subset would be perturbed and input to the base learner. Afterwards， the corresponding prediction models are obtained. Then， the Bayesian method is used as the combined strategy of ensemble learning to fuse multiple prediction models. Finally， the diagnosis is provided by voting combined with the classification results of multiple beats in an ECG. The experimental results show that the method has high sensitivity and specificity， which has clinical application value.

Key words: electrocardiogram(ECG), bundle branch block, ensemble learning, convolutional neural networks, Bayesian method