The IoT-enabled CNN classification system achieved 99% precision for atrial fibrillation and normal sinus rhythm and 98% precision for congestive heart failure on benchmark ECG datasets with 32 seconds latency per diagnosis.
An IoT-enabled, PSO-optimized wavelet denoising and CNN classification system deployed on a low-cost microcontroller provides highly accurate and low-latency automated ECG analysis for real-time monitoring.
Effect estimate: Precision 0.99 for ARR and NSR; 0.98 for CHF
Electrocardiogram (ECG) signal analysis is fundamental for the diagnosis of cardiac disorders and the detection of arrhythmias. However, manual interpretation remains labor-intensive and prone to variability, emphasizing the necessity for automated diagnostic systems. This study presents a novel methodology that combines particle swarm optimization (PSO) with wavelet transform (WT) to enhance ECG signal quality through effective noise suppression. The proposed denoising framework achieves a notable signal-to-noise ratio (SNR) of 18.24 at an input SNR of 10 dB, preserving vital signal features. Subsequently, the denoised signals are converted into two-dimensional representations via the 2D discrete wavelet transform (2D DWT), enabling robust feature extraction for classification using convolutional neural networks (CNNs). The classification model achieves exceptional performance, with precision values of 0.99 for atrial fibrillation (ARR) and normal sinus rhythm (NSR), and 0.98 for congestive heart failure (CHF), accompanied by high recall and F1-scores. In addition, the study introduces a fully automated arrhythmia detection and classification system, integrating an optimized deep learning architecture with real-time ECG acquisition using the AD8232 sensor. Data transmission is facilitated through a secure Internet of Things (IoT) framework employing the Node-RED IBM platform and the Message Queuing Telemetry Transport (MQTT) protocol, ensuring efficient and reliable communication with analytical modules. The system is deployed on an ESP32 microcontroller, demonstrating low-latency processing with the complete diagnostic workflow spanning acquisition to classification executed within 32 s. These results underscore the system’s efficacy in delivering accurate and timely cardiac anomaly detection, with promising implications for real-time healthcare monitoring and smart medical diagnostics.
Azzouz et al. (2026) studied ECG signal classification including Cardiac Arrhythmia, Congestive Heart Failure, and Normal Sinus Rhythm (n=162). Particle Swarm Optimization-tuned Wavelet Transform for ECG denoising combined with 2D DWT-based feature extraction and lightweight CNN classification on an IoT-enabled embedded ESP32 system vs. No treatment or manual ECG interpretation was evaluated on Classification precision for ECG arrhythmia types (ARR, CHF, NSR) (Precision 0.99 for ARR and NSR; 0.98 for CHF). The IoT-enabled CNN classification system achieved 99% precision for atrial fibrillation and normal sinus rhythm and 98% precision for congestive heart failure on benchmark ECG datasets with 32 seconds latency per diagnosis.