Automating diagnosis from ECG signals
Congestive heart failure is the condition where the heart loses its ability to pump blood. The usual approach for diagnosis is to examine ECG signals. But that is time consuming and often inaccurate if the required expertise is not available.
Recently, Manish Sharma from the Institute of Infrastructure, Technology, Research and Management, Ahmedabad collaborated with researchers from IIIT Nagpur and Singapore to come up with an alternative: to detect the ECG peaks that signal congestive heart failure, using automated machine learning techniques.
The team collected ECG data from existing databases such as Fantasia, the Normal Sinus Rhythm Database and Boston’s Beth Israel Hospital. The data from these databases were arranged as balanced and unbalanced datasets to train the system.
First, the ECG signals were decomposed using a filtering machine. Then the wavelet coefficients obtained were input into a feature extraction engine. The output was fed into a support vector machine to further classify the peaks.
The researchers then checked overall accuracy, specificity, sensitivity and error rate, against data they had not used while training the system. They report that the method had a very low error rate and more than 99% accuracy.
Besides detecting congestive heart failure at early stages, this strategy can be used to detect coronary artery disease and myocardial infarction, says Manish Sharma.
Cognitive systems Research, 55: 82-94 (2019)