Early Heart Disease Detection Based on Anomaly Behavior in ECG Data Using Cross-Correlation and Machine Learning

Abstract

The importance of early detection of electrocardiogram anomalies primarily lies in its potential to facilitate timely diagnosis of heart diseases and improve patient outcomes. Existing detection approaches continue to adopt intricate feature selection, often leading to increased complexity of the model without significant improvement in the detection performance. In addressing the challenge, the current study seeks to develop a hybrid model that incorporates SVM with cross-correlation for feature extraction and ECG signal classification, namely, Normal Sinus Rhythm and Atrial Fibrillation. The developed model was trained and validated with a dataset of 36 ECG recordings that had been carefully segmented and preprocessed to enhance the focus and precision of the features. The developed model was able to cross correlate for feature extraction over the impulse response and pattern recognition for classification and demonstrated excellent performance in differentiating NS from AF ECG signals. The study attains the important benchmarking of 100% accuracy, sensitivity and specificity on the dataset, thereby, demonstrating its potential to reliably detect rhythm disturbance. This study represents a considerable contribution to the ECG analysis by proposing a diagnostic model that integrates cross-correlation based feature extraction and SVM classification. The model's precise functioning with a well-annotated dataset and the slight need for preprocessing and minimal data shows the opportunity for the model to be scaled and implemented more objectively for practical use in clinical settings for the detection of heart disease in early stages.