Classification of Heart Beats Using LVQ Neural Networks after Detection from Continuous ECG Signal, Followed by Feature Extraction Using PCA: A Case Study

The ECG (electrocardiogram) signal provides a lot of valuable information about the condition of the heart. Many cardiac problems are visible as distortions in the ECG. The analysis of heart beat cycles in an ECG signal is vital for long-term monitoring of heart patients. However, in long-term monitoring, it is a very tedious and timeconsuming task to analyze the ECG recording beat by beat. This is because the abnormal heart beats can occur randomly and a long-term ECG record, say 24 hours, may contain hundreds of thousands of heart beats. Hence, it is highly desirable to automate the entire process of heart beat classification by developing a computer-assisted technique to annotate the heart beats in order to facilitate review by a medical expert. Since several arrhythmia are potentially dangerous and life threatening, if not detected within a few seconds to a few minutes of their onset, their fast and accurate detection on a real-time basis is required. The present work proposes a technique for classification of a heart beat into one of the three categories: Normal, PVC (Premature Ventricular Contraction) and Fusion First, the individual beats belonging to each category are extracted from the MIT-BIH arrhythmia database and after preprocessing, features are extracted from the beats using Principal Component Analysis (PCA) .This is done in order to reduce the dimensionality of the vectors to be input to the neural network classifier so as to avoid issues like increased computational complexity and overtraining of the net. The feature vectors thus obtained are used to train an LVQ based neural network classifier to classify them in the aforementioned categories. After the network is trained, its performance in terms of its generalizing ability is tested on a separate test dataset which was not used during training. Key Words-PVC, ECG beat classification, PCA, neural network