🧠 BCI Motor Imagery Classification using CNN-LSTM

Decoding brain signals for motor imagery classification using hybrid deep learning

🧠 Brain-Computer Interface | 🎯 Motor Imagery | 🤖 Deep Learning | 📊 EEG Analysis

---

🌟 What is This Project About?

This project tackles one of the most fascinating challenges in Brain-Computer Interface (BCI) technology: Motor Imagery (MI) classification. By combining the spatial feature extraction power of Convolutional Neural Networks (CNNs) with the temporal sequence modeling capabilities of Long Short-Term Memory (LSTM) networks, we can decode brain signals and classify different motor imagery tasks.

Motor Imagery refers to the mental rehearsal of motor actions without actual movement - imagine moving your left hand, right hand, feet, or tongue. Our hybrid CNN-LSTM architecture can distinguish between these different imagined movements by analyzing EEG brain signals.

🔬 Key Features:

• 🧠 EEG Signal Processing: Advanced preprocessing of brain signals
• 🎨 Spatial Feature Extraction: CNN layers capture spatial patterns in EEG data
• 🔄 Temporal Sequence Modeling: LSTM networks learn temporal dependencies
• 🎯 Multi-class Classification: Classify different motor imagery tasks
• 📊 Comprehensive Analysis: Detailed visualization and performance metrics
• 🔧 Production-Ready: Clean, modular, and scalable implementation

🚀 Quick Start

# Clone the repository
git clone https://github.com/VeerajSai/BCI-MI-using-CNN-LSTM.git
cd BCI-MI-using-CNN-LSTM

# Install dependencies
pip install -r requirements.txt

# Run the model training
python train.py

🏗️ Architecture Overview

EEG Data → Preprocessing → CNN Layers → Feature Maps → LSTM Layers → Dense → MI Classification
    ↓           ↓            ↓             ↓            ↓          ↓           ↓
Raw EEG    Filtering/    Spatial      Temporal     Sequential  Final     Left/Right/
Signals    Normalization  Features     Features     Memory     Dense     Feet/Tongue

🔍 Model Components

Component	Purpose	Key Features
Preprocessing	EEG signal cleaning	Bandpass filtering, normalization, epoching
CNN Block	Spatial pattern extraction	Conv1D/2D, MaxPooling, Dropout
LSTM Block	Temporal dependency modeling	Bidirectional LSTM, sequence learning
Classification	Motor imagery prediction	Dense layers, softmax activation

📊 Performance Metrics

Metric	Description	Status
Accuracy	Multi-class classification accuracy	✅ Optimized
Precision	Class-wise prediction precision	✅ High
Recall	True positive detection rate	✅ Excellent
F1-Score	Balanced precision-recall metric	✅ Robust
Kappa Score	Agreement beyond chance	✅ Strong
Training Time	Model convergence time	⚡ Efficient

Performance metrics are dataset-dependent and will vary based on subject and session

🛠️ Implementation Details

EEG Data Preprocessing

def preprocess_eeg(raw_data, fs=250):
    # Bandpass filtering (8-30 Hz for motor imagery)
    filtered_data = bandpass_filter(raw_data, low=8, high=30, fs=fs)
    
    # Common Average Reference (CAR)
    car_data = apply_car(filtered_data)
    
    # Epoching and normalization
    epochs = create_epochs(car_data, events, tmin=0, tmax=4)
    normalized_epochs = normalize_epochs(epochs)
    
    return normalized_epochs

Hybrid CNN-LSTM Architecture

def create_cnn_lstm_model(input_shape, num_classes=4):
    model = Sequential([
        # CNN for spatial feature extraction
        Conv2D(32, (1, 3), activation='relu', input_shape=input_shape),
        BatchNormalization(),
        MaxPooling2D((1, 2)),
        
        Conv2D(64, (1, 3), activation='relu'),
        BatchNormalization(),
        MaxPooling2D((1, 2)),
        Dropout(0.25),
        
        # Reshape for LSTM
        Reshape((timesteps, features)),
        
        # LSTM for temporal modeling
        LSTM(128, return_sequences=True, dropout=0.2),
        LSTM(64, dropout=0.2),
        
        # Classification layers
        Dense(50, activation='relu'),
        Dropout(0.5),
        Dense(num_classes, activation='softmax')
    ])
    return model

🎯 Motor Imagery Classes

🤚 Classification Tasks:

• Left Hand: Imagining left hand movement
• Right Hand: Imagining right hand movement
• Feet: Imagining foot/feet movement
• Tongue: Imagining tongue movement

📈 Applications:

• Assistive Technology: Control devices for paralyzed patients
• Rehabilitation: Motor function recovery training
• Gaming: Mind-controlled gaming interfaces
• Research: Understanding brain motor control mechanisms

🔧 Installation & Setup

Prerequisites

• Python 3.8+
• CUDA-compatible GPU (recommended for faster training)
• 8GB+ RAM
• EEG datasets (BCI Competition IV, etc.)

Dependencies Required

pip install tensorflow>=2.8.0
pip install keras>=2.8.0
pip install numpy>=1.21.0
pip install scipy>=1.7.0
pip install mne>=0.24.0        # For EEG data handling
pip install scikit-learn>=1.0.0
pip install matplotlib>=3.5.0
pip install seaborn>=0.11.0
pip install pandas>=1.3.0

📚 Usage Guide

Training the Model

# Train on BCI Competition dataset
python train.py --dataset bci_iv --subject 1 --epochs 100 --batch_size 32

# Train with custom EEG data
python train.py --data_path ./data/custom/ --model_type cnn_lstm --lr 0.001

Evaluating Performance

# Evaluate trained model
python evaluate.py --model_path ./results/models/best_model.h5 --test_data ./data/test/

# Cross-subject evaluation
python evaluate.py --cross_subject --subjects 1,2,3,4 --cv_folds 5

Real-time Classification

# Real-time BCI classification
python real_time_classify.py --model ./results/models/subject1_model.h5 --stream_data

🎨 Visualization & Analysis

EEG Signal Visualization

• Raw and filtered EEG signals
• Topographic maps of brain activity
• Time-frequency analysis (spectrograms)
• Event-related potentials (ERPs)

Model Analysis

• Training/validation curves
• Confusion matrices per class
• Feature importance visualization
• Classification accuracy per subject

Performance Metrics

• Subject-specific accuracy analysis
• Cross-session validation results
• Statistical significance testing
• Comparison with baseline methods

🤝 Contributing

We welcome contributions to improve BCI motor imagery classification! Here's how you can help:

1. 🍴 Fork the repository
2. 🌿 Create your feature branch (git checkout -b feature/NewPreprocessing)
3. 💻 Commit your changes (git commit -m 'Add advanced filtering')
4. 🚀 Push to the branch (git push origin feature/NewPreprocessing)
5. 📬 Open a Pull Request

Areas for Contribution

• Support for more EEG datasets
• Advanced preprocessing techniques
• Attention mechanisms for interpretability
• Real-time optimization
• Better Architecture
• Cross-subject transfer learning
• Web deployment

📊 Benchmarks & Comparisons

Method	Approach	Advantages	Best Use Case
CNN-LSTM (This Project)	Hybrid Spatial-Temporal	Best of both worlds	Motor Imagery BCI
CSP + SVM	Classical ML	Fast, interpretable	Simple motor tasks
Pure CNN	Deep spatial	Good for spatial patterns	Image-like EEG data
Pure LSTM/RNN	Sequential modeling	Temporal dependencies	Time series EEG
Transformer	Attention-based	Long-range dependencies	Complex sequences

🏆 Project Highlights

• 🧠 BCI Focus: Specifically designed for brain-computer interface applications
• 🎯 Motor Imagery: Specialized for classifying imagined movements
• 📊 EEG Expertise: Advanced EEG signal processing and analysis
• 🔬 Research Quality: Based on established BCI research methodologies
• 🚀 Production Ready: Scalable architecture for real-world deployment
• 📈 Comprehensive: End-to-end pipeline from raw EEG to classification

🔮 Future Roadmap

• Real-time Optimization: Low-latency classification for online BCI
• Advanced Architectures: Attention mechanisms and transformer variants
• Multi-modal Integration: Combine EEG with other biosignals
• Edge Deployment: Optimize for embedded BCI systems
• Clinical Validation: Extensive testing with patient populations

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

⭐ Star this repo if you found it helpful for BCI research!

Made with ❤️ by VeerajSai

"The mind is not a vessel to be filled, but a fire to be kindled." - Plutarch