Research On Diagnosis Method Of Abnormal ECG Based On Improved Random K Label Items

With the increasing pressure of fierce social competition and the acceleration of population aging,the prevalence of cardiovascular disease continues to rise.The electrocardiogram records the changes of patient’s heartbeat,which is the most commonly used method for diagnosing cardiovascular disease.At present,the difficulties of automatic diagnosis of ECG are as follows.1)The manual annotation of ECG data is time-consuming and costly,and it is difficult to obtain sufficient samples with complete and correct abnormality labels.2)The transition of ECG signal often indicates a variety of cardiovascular diseases,and the intricate correlations among these diseases make diagnosis more difficult.In order to solve the above problems,the main research work of this paper is as follows.(1)For the weak label datasets with missing and mislabeled labels,we propose to clean them to become the high-quality datasets that can be used for training,assisted by a small number of standard datasets that are fully and correctly labeled.In the first step,the positive and negative samples of different labels are clustered,and then,according to the distance from weak label data to the positive and negative clusters and the label similarity of the k-nearest samples,the mislabeled labels are removed.The second step is to mine the association rules between clusters according to the association between labels,and then fill in the missing labels for the weakly-labelled samples according to the rules.The third step is to clean iteratively until meeting the ending conditions.(2)An improved Random k labelsets(RAKEL)algorithm is proposed based on the correlations among the ECG labels.Firstly,based on the Bayesian network,the correlation between labels is mined to form the subset of candidate labels.Then,based on the similarity of optimal feature space of label subset,the correlation degree of candidate label subset is detected,and the label subset with high correlation degree is retained.Finally,based on Label Powerset(LP),classifiers are trained and predicted on the optimal feature space of the retained label subset.The main contributions are as follows.Firstly,an iterative cleaning algorithm for weak label ECG data is proposed.It solves the time-consuming and labor-consuming problem of manual labeling.According to the standard data set,weakly-labelled samples are used to enrich the training data set,which is is conducive to improving the prediction performance of the model.Secondly,an improved RAKEL algorithm is proposed based on the correlation of ECG data labels.According to Bayesian network and the similarity of optimal feature space,the label subset is determined,which avoids the drawback that RAKEL does not fully consider the correlation between labels.Experiments on real and simulated datasets demonstrate the effectiveness of the proposed method.