IoT-based ECG arrhythmia detection and classification model using an optimization based hybrid deep capsule stacked autoencoder
Related Experiment Videos
Vijay A Kotkar1, Makarand Shahade2, Dhanraj S Jadhav3
1Department of Computer Engineering, PCET's Pimpri Chinchwad College of Engineering and Research, Ravet, Pune, Maharashtra 412101, India.
Insights
A novel deep learning method accurately identifies cardiac arrhythmia using ECG and vital signs. This approach overcomes challenges like noise and data limitations, achieving 99.6% accuracy for real-world application.
Area of Science:
- Cardiology and Medical Informatics
- Artificial Intelligence in Healthcare
- Signal Processing
Background:
- Heart disease, particularly cardiac arrhythmia, poses significant risks, especially in adults over 40.
- Current arrhythmia classification methods face challenges including signal quality issues, class imbalance, and generalization limitations.
- Existing methodologies often fail to address these critical issues comprehensively.
Purpose of the Study:
- To introduce a novel deep learning-based framework for accurate identification and classification of cardiac arrhythmias.
- To address limitations of existing methods by incorporating enhanced pre-processing and feature extraction techniques.
- To improve the reliability and applicability of arrhythmia detection in real-world clinical settings.
Main Methods:
- A pre-processing step involving min-max normalization and an enhanced wavelet Wiener filtering technique (UP_WWF).
- Feature extraction using an adaptive dense convolutional osprey network optimized by the osprey optimization algorithm.
- Classification of arrhythmias using an attention-based deep capsule stacked autoencoder (Att_DCSAE).
Main Results:
- The proposed framework achieved an overall accuracy of 99.6% on a combined ECG dataset (including body temperature, heart rate, and blood oxygen levels).
- Evaluated on the MIT-BIH dataset, the approach yielded high precision (99.98%), recall (99.68%), and F1-score (99.86%).
- Demonstrated significant performance improvements over the latest techniques in the literature.
Conclusions:
- The developed deep learning framework offers a robust solution for cardiac arrhythmia classification.
- The method's high accuracy and ability to handle real-world data complexities suggest strong clinical applicability.
- This approach represents a significant advancement in automated arrhythmia detection and diagnosis.
Abstract:
Heart disease is one of the most dangerous diseases, and it is getting worse in adults over the age of 40. Cardiac arrhythmia is defined as an irregular heartbeat, and quick diagnosis and treatment are critical for saving the lives of a huge population. Currently, the arrhythmia classification involves diverse challenges that limit the existing methodologies from attaining the desired results. Some of the major challenges involve signal quality issues, including noise and motion artefacts, class imbalance and dataset limitations, generalization issues due to context variability across patients, real-world deployment and clinical trust. Several methodologies have been introduced to date to focus on these critical issues, but most of them are unable to cover and address the issues altogether. This study offers a novel deep learning-based method for identifying and classifying arrhythmias as a remedy. First, the pre-processing step receives the input data, which includes body temperature, heart rate, blood oxygen level, and ECG signal. Pre-processing is done with min-max normalization and an enhanced wavelet Wiener filtering technique (UP_WWF) in this case. After pre-processing, the input is submitted to the feature extraction stage, which extracts significant features. Features are extracted using an adaptive dense convolutional osprey network where the osprey optimization algorithm is used to tune the parameters of the network for better feature discrimination. Finally, with the help of the extracted features, the various arrhythmia classes are classified using an attention-based deep capsule stacked autoencoder (Att_DCSAE). The proposed framework is evaluated using the MIT-BIH dataset for benchmarking, whereas the constructed combined ECG dataset includes ECG, heart rate, blood oxygen level, and body temperature measurements. The proposed approach resulted in an overall accuracy of 99.6% on the combined ECG dataset, signifying that it can be applied in hard real-world conditions. The study also shows significant improvements when evaluated on the MIT-BIH dataset with a total precision, recall, and F1-score values of 99.98%, 99.68% and 99.86%, resulting in the overall improvement of 0.6% precision, 0.08% recall, and 0.7% F1-score when compared with the latest techniques in the literature.
