An ECG classification based on modified local binary patterns: a novel approach

The lack of a fast, reliable, and general electrocardiogram (ECG) classification algorithm remains a major challenge toward an ECG-only workflow for heart disease diagnosis. In this work, the feasibility of our proposed algorithm in classifying 11 different heartbeat arrhythmia is investigated. Eleven heartbeat classifications with a total of 30,790 heartbeats selected from 32 patients from MIT-BIH dataset were investigated to evaluate the proposed algorithm, which is based on Modified Local Binary Pattern (MLBP). The reference feature vector for each arrhythmia was extracted during the training phase by applying the LBP operator to all different ECG signals individually, and the log-likelihood operator is used to classify each signal MLBP vector and all reference feature vectors. To enhance the algorithm accuracy, two additional morphological features are investigated, which are variance and mean. The proposed Mean–Variance Modified-LBP (MV-MLBP) algorithm was applied, and the average accuracy of 99.76 was obtained. The MV-MLBP was found to be noise resistance, while the reported accuracy was obtained using no pre-processing, such as drift cancelation and noise reduction. In the arrhythmia classification process, the MV-MLBP algorithm has recorded a noticeably high accuracy rate. The proposed LBP-based approach has great potential to be transmitted to the clinic. No pre-processing necessity, combined with the low computational complexity, has changed it to a fast and robust ECG classification algorithm. However, additional patient studies are necessary to optimize and validate the workflow.