Customization methods for ECG beat classifiers using SOM and LVQ achieved classification accuracies of 91.3% and 90.3% without requiring annotated patient-specific data.
Do customization methods for ECG beat classifiers using SOM and LVQ achieve high classification accuracy without requiring annotated patient-specific data?
Customization methods for ECG beat classifiers using SOM and LVQ can achieve >90% accuracy without requiring annotated patient-specific data, offering potential for commercial ECG analysis systems.
Tasa de eventos absoluta: 91.3% vs 98.8%
In this study, a Self-Organizing Map (SOM), an unsupervised clustering algorithm and Learning Vector Quantization (LVQ), were employed to develop an ECG beat classifier. To improve the performance of the classifiers, we investigated two methods of customization requiring minimum human intervention and preliminary results are reported. The MIT/BIH database was used to develop and test the above methods. The classifier developed using SOM and LVQ provided a classification accuracy of 98.8% on 33 files, which assumed the availability of a small set of annotated patient specific data. The two customization methods that were developed to overcome the need of annotated patient specific data reported accuracies of 91.3% and 90.3% respectively. The customization techniques seem to offer a great potential in improving the dynamic performance of commercial ECG analysis systems.
Palreddy et al. (Tue,) conducted a other in ECG beat classification. Customization methods for ECG beat classifier using SOM and LVQ vs. Classifier requiring annotated patient-specific data was evaluated on Classification accuracy. Customization methods for ECG beat classifiers using SOM and LVQ achieved classification accuracies of 91.3% and 90.3% without requiring annotated patient-specific data.