Model Deteksi Obstruktive Sleep Apnea Berbasis Integrasi Multimodal Sinyal ECG-EMG Menggunakan Deep Learning
YOSSI HASANAH PUTRI, Prof. Ir. Hanung Adi Nugroho, S.T., M.Eng., Ph.D., IPM., SMIEEE ; Ahmad Ataka Awwalur Rizqi , S.T., Ph.D
2026 | Tesis | S2 Teknik Elektro
Obstructive respiratory disorders such as sleep apnea are conditions that can lead to cardiovascular and metabolic complications if not detected early. Early detection is therefore essential to prevent long-term risks and to support the development of non-invasive health monitoring systems. This study aims to develop a cardiorespiratory monitoring method based on the combination of electrocardiogram (ECG) and electromyogram (EMG) signals using a Convolutional Neural Network Long Short-Term Memory (CNN–LSTM) architecture for automatic detection of sleep apnea events.
The dataset used in this research was obtained from the Multi-Ethnic Study of Atherosclerosis (MESA) Sleep Dataset, which contains synchronized ECG and EMG recordings during sleep along with clinically validated apnea labels. The preprocessing stages included filtering, denoising, segmentation, and normalization to reduce noise while preserving relevant physiological characteristics. The CNN architecture was employed to extract spatial features from ECG and EMG segments, while the LSTM network captured temporal dependencies between epochs, enabling the model to identify dynamic patterns related to apnea occurrences.
Training and testing were performed using a balanced dataset to ensure robust generalization. Experimental results demonstrate that the proposed multimodal CNN–LSTM model achieved a high detection performance, with an accuracy of 90.3%, precision of 88.7%, sensitivity (recall) of 91.5%, F1-Score of 90.1%, and an AUC of 0.895. These results outperform several previous studies using single-modal ECG inputs by approximately 2–5%. Furthermore, the model exhibits stable convergence without overfitting and maintains computational efficiency with 382.416 parameters, making it suitable for implementation in wearable or edge computing-based sleep monitoring systems.
This study confirms that integrating ECG and EMG signals through the CNN–LSTM architecture can produce an accurate, efficient, and adaptive automatic apnea detection system. The proposed approach contributes to advancing smart healthcare technology based on physiological signal analysis and provides a foundation for future telemedicine applications.
Kata Kunci : CNN–LSTM, ECG, EMG, Pemantauan Kardiorespiratori, Sleep Apnea / CNN–LSTM, ECG, EMG, Cardiorespiratory Monitoring, Sleep Apnea