Sleep Apnea Detection using 1D CNN-BiLSTM with Attention

An AI-driven pipeline that detects sleep apnea from single-lead ECG signals using 1D CNN and BiLSTM layers enhanced with an attention mechanism.

🧠 Abstract

Sleep apnea is a disorder that leads to disrupted breathing during sleep. This project introduces a deep learning architecture combining 1D CNN + BiLSTM with an attention layer to detect apnea events from ECG signals.

---

👨‍💻 Team Members

• Sampurnaa Nag (12024052016005)
• Sylvia Barick (12024052016009)
• Debojyoti De Majumder (12024052020010)

---

📈 Architecture

• CNN Layers: Extract spatial features from raw ECG.
• BiLSTM: Capture temporal dependencies from past & future.
• Attention Mechanism: Focus on apnea-relevant time steps.
• Fully Connected + Sigmoid: Classify apnea or normal.

---

📊 Performance

• Training Accuracy: 95.79%
• Validation Accuracy: 93.62%
• Test Accuracy: 97.18%

---

📁 Dataset

• Source: PhysioNet Apnea-ECG Database
• Preprocessing:
  • Chebyshev bandpass filter (0.5–48 Hz)
  • 1-minute segmenting
  • Autocorrelation-based noise removal

---

🚀 Future Scope

• Incorporate SpO₂ and respiratory signals
• Use Transformer/Ensemble architectures
• Improve explainability for clinical use

---

📄 References

See references.md

---

🛠️ Setup

git clone https://github.com/<your-username>/Sleep-Apnea-Detection-CNN-BiLSTM.git
cd Sleep-Apnea-Detection-CNN-BiLSTM
pip install -r requirements.txt