A Multi-Resolution Deep Learning Approach for IoT-Integrated Biomedical Signal Processing using CNN-LSTM

The rapid advancement of Internet of Things (IoT) technology and deep learning has enabled real-time health monitoring for elderly individuals and patients. This work proposes an IoT-based health monitoring system integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) netw...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2025 5th International Conference on Trends in Material Science and Inventive Materials (ICTMIM) S. 1362 - 1369
Hauptverfasser: Priyadharshini, K., Mathias, Ajisha, Krishnan, R. Santhana, Kanthimathi, N, Francis Raj, J. Relin, Malar, P. Stella Rose
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 07.04.2025
Schlagworte:
Anomaly detection
Biomedical monitoring
Biomedical Sensors
Biosensors
CNN-LSTM
Convolutional neural networks
Deep learning
Feature extraction
Health Monitoring
IoT
Medical diagnostic imaging
Medical services
Monitoring
Physiological Signals
Real-Time Analysis
Real-time systems
Signal Denoising
Wavelet Transform
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The rapid advancement of Internet of Things (IoT) technology and deep learning has enabled real-time health monitoring for elderly individuals and patients. This work proposes an IoT-based health monitoring system integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to analyze physiological signals and detect anomalies effectively. A network of biomedical sensors, including ECG, EMG, EEG, heart rate, blood pressure, SpO2, temperature, accelerometer, gyroscope, and respiratory rate sensors, continuously collects health data. These sensors transmit real-time physiological signals using wireless communication protocols such as Wi-Fi, Bluetooth, LoRa, and NB-IoT, ensuring seamless data acquisition and remote monitoring. To enhance signal quality, the preprocessing stage employs Wavelet Transform (WT) for denoising biomedical signals, mitigating motion artifacts and environmental noise. Discrete Wavelet Transform (DWT) is applied to extract meaningful features while preserving critical health information. Additional preprocessing steps, including feature normalization, segmentation, and feature extraction, improve deep learning model efficiency. The proposed CNN-LSTM hybrid model leverages CNN for spatial feature extraction and LSTM for capturing temporal dependencies in time-series biomedical data. The architecture includes multiple convolutional layers with ReLU activation, batch normalization, max-pooling, bidirectional LSTM layers, dropout regularization, and an attention mechanism. Model training utilizes labeled medical datasets with an Adam optimizer and binary cross-entropy loss function. Performance evaluation considers accuracy, precision, recall, F1-score, and AUC-ROC metrics. Results demonstrate improved anomaly detection for conditions such as arrhythmias, respiratory irregularities, and early stroke indicators. Deployment in real-time health monitoring applications facilitates continuous patient assessment, fall detection, and predictive medical alerts. This system enhances proactive healthcare solutions by integrating IoT and deep learning for real-time, intelligent, and automated health monitoring.
DOI:10.1109/ICTMIM65579.2025.10988087