Electrocardiogram Diagnosis Using Hybrid Case-Based Reasoning

This paper proposes a hybrid architecture that combines correlation dimension and feature extraction techniques into the case-based reasoning (CBR) technique to do electrocardiogram (ECG) diagnosis. CBR is used in the system as the core problem-solving method, which applies the expert experience to analyze ECG and propose diagnosis. The correlation dimension values of ECGs are used for clustering of cases for CBR. A Min-Max Turning Points Selection (MTPS) algorithm is used to quickly extract turning points from an ECG waveform as wave features for retrieval of the most promising solution cases from a specific cluster of cases. Finally, the CBR can learn representative cases from the problem-solving process by an approximation-based training method. This hybrid architecture manifests several interesting features. First, high-quality representative cases can be effectively learned, thanks to the correlation dimension values-based clustering and approximation-based training methods. This feature not only reduces the number of cases needed to be stored in the case library, but improves the correctness of the proposed solutions. Second, the proposed MTPS algorithm for extraction of turning points is fast, simple, and noise-tolerant, which helps a lot in the calculation of similarity among cases. Finally, the clustering algorithm can automatically re-cluster the cases in the case library in response to new diseases, which prevents our system from degradation due to the increase of new cases. Our experiments with the MIT-BIH arrhythmia database demonstrate that the architecture has very high sensitivity and specificity, as well as speedy response.