ECG Signal Classification and Anomaly Detection using LSTM and HMM Models

Accurate detection and classification of electrocardiogram (ECG) signals is required for cardiac condition diagnosis and timely medical intervention. However, capturing the complex temporal dependencies in ECG data is difficult and hence may provide less effective performance comparatively. The project proposes an innovative approach to detecting and classifying the anomalies in ECG using LSTM and HMM. Preprocessing and normalization of the ECG data are done. After that, the sequence is prepared for input in LSTM, which builds a strong LSTM model with reinforced regularization techniques designed for feature learning from the ECG data. The features extracted by LSTM were concatenated with the original features and applied to train the HMM for anomaly detection based on estimating the likelihood of the log probabilities. Thresholding the log probabilities will locate anomalies. The next step is the implementation of HMM to classify ECG signals. The proposed method shows the framework comprehensively for effective detection of anomaly and classification, combining the strong points of both LSTM and HMM. The results obtained are employed to illustrate and prove the efficiency of the proposed combined approach and emphasise the importance of the combination of deep learning and statistical models for further analysis of ECG signals.