Ajay007001/Brain-Stroke-Prediction

🧠 Brain Stroke Classification using VGG19 Transfer Learning

A high-accuracy image classification model trained on medical brain scans using VGG19 (ImageNet pretrained) to distinguish different stroke types. Achieves 89% accuracy on the Teknofest 2021 Brain Stroke dataset.

🎯 Training Configuration

• Image Size: 250x250 pixels (optimal resolution)
• Epochs: 15 (best performance achieved)
• Batch Size: Standard training configuration
• Data Augmentation: Applied for robust training

---

🌟 Model Performance

Confusion Matrix

### Model Training Performance ### ROC curve

🌟 Live Streamlit Demo

👉 🚀 Launch the Interactive Web App

Experience the power of AI-driven medical imaging analysis:

• 📤 Drag & Drop your MRI scan (PNG, JPEG)
• ⚡ Instant Predictions with confidence scores
• 📊 Interactive Visualizations and risk analysis
• 📱 Mobile & Desktop responsive interface
• 🎯 Real-time Classification of stroke types

No installation required - just upload and predict!

---

🧪 Testing Images

👉 📁 Download Test Images from Google Drive

Ready-to-use brain MRI samples for testing this model:

• 🩸 Hemorrhagic Stroke samples
• 🧠 Ischemic Stroke samples
• ✅ No Stroke samples
• 📋 Organized by category for easy identification

Use these test images to evaluate the model's performance and capabilities!

---

🚀 Highlights

• ✅ Transfer learning with VGG19
• 🧠 Medical image classification for stroke diagnosis
• 📊 Accuracy: 89%, F1-Score: 88%
• 🔬 Trained on colorized brain stroke CT/MRI images
• 📁 Dataset: Teknofest 2021, Kaggle

---

📂 Dataset

• Source: Teknofest 2021 Brain Stroke Dataset
• Type: Colorized medical images
• Split: train/, test/ with multiple stroke classes

---

🏗️ Model Details

• Base: VGG19 with frozen early layers
• Custom Head: Dense, Dropout, Softmax
• Optimizer: Adam | Loss: Sparse Categorical Crossentropy
• Input Size: 250x250 | Framework: Keras + TensorFlow 2.x

---

📈 Performance Metrics

Metric	Value
Accuracy	89%
F1-Score	88%
Precision	87%
Recall	88%

---

🧪 Usage

This model is ready for integration into clinical AI pipelines or academic research.
To load the .h5 file and run predictions, refer to colab notebook ↗ (see predict.py or notebooks/).

Quick Links:

• 🌐 Live Streamlit Demo - Test the model instantly
• 📁 Testing Images - Sample MRI scans for evaluation

---

📋 License

MIT License – free to use, modify, and distribute.

👤 Author

Ajay Vasan S
Machine Learning Engineer
🔗 GitHub | 📧 LinkedIn

📂 GitHub Project: AjayVasan/Brain-Stroke-Predictor

⭐ Star this repo if it helped you — and feel free to open issues for feedback!