🦠 PlasmoScan

Detect malaria-infected and uninfected cells using deep learning (custom CNN and transfer learning with VGG16).

📝 Project Overview

This project classifies cell images as Parasitized (malaria-infected) or Uninfected using:

• 🧠 A custom Convolutional Neural Network (CNN)
• 🏆 Transfer learning with VGG16 (feature extraction and fine-tuning)

🔄 Workflow

1. Dataset Download
   • ⬇️ Download the dataset from Kaggle: cell-images-for-detecting-malaria
   • 🗂️ Unzip to cell_images/ with subfolders Parasitized/ and Uninfected/
2. Data Preparation
   • 🧹 Images are loaded and preprocessed using ImageDataGenerator with augmentation (rescale, zoom, flip)
   • 🔀 80/20 train/validation split
3. Modeling Approaches
   • Custom CNN:
     • 🏗️ 3 Conv2D blocks (with BatchNorm, MaxPooling, Dropout)
     • 🧩 Dense layers for binary classification
     • ⏳ Trained for 10 epochs (Adam optimizer, learning rate 0.0003)
   • Transfer Learning (VGG16):
     • 🏛️ VGG16 base (pretrained on ImageNet, no top)
     • 🧩 Custom dense layers on top
     • 🧊 Feature Extraction: VGG16 base is frozen, only new layers are trained
     • 🔓 Fine-Tuning: Last convolutional block of VGG16 is unfrozen and trained along with the new layers for improved accuracy
     • ⏳ Trained for 10 epochs (RMSprop optimizer, learning rate 1e-5 for fine-tuning)
4. Evaluation
   • 📈 Training and validation accuracy/loss plotted for each approach
   • 🥇 Typical results: >96% validation accuracy after fine-tuning
5. Model Saving
   • 💾 Models saved as .h5 files (e.g., malaria_model.h5, malaria_model_finetuning.h5)
6. Prediction & Visualization
   • 🖼️ Random images from both classes are predicted and visualized with confidence scores

🔬 Fine-Tuning (VGG16)

Fine-tuning is a crucial step that allows the model to adapt the deeper layers of VGG16 to the malaria cell dataset:

• 🧊 Feature Extraction: Initially, all VGG16 layers are frozen and only the custom classifier is trained.
• 🔓 Fine-Tuning: The last convolutional block of VGG16 is unfrozen, allowing its weights to be updated during training. This helps the model learn dataset-specific features and improves accuracy.
• ▶️ How to Run:
  • In the notebook, follow the section labeled Fine Tuning.
  • The code will unfreeze the last block, recompile the model with a low learning rate, and continue training.
• 🏅 Results:
  • Fine-tuning typically increases validation accuracy to ~96% or higher.
  • The notebook provides plots to compare training/validation accuracy and loss before and after fine-tuning.

🚀 Usage

1. Install dependencies:

   pip install -r requirements.txt

2. Download and extract the dataset from Kaggle as described above.
3. Run the notebook:

   jupyter notebook notebook.ipynb

   • ▶️ Follow the notebook cells to train, evaluate, and visualize predictions.
   • 🔬 For fine-tuning, execute the cells in the Fine Tuning section after initial transfer learning training.

🐳 Docker

A Dockerfile is provided for containerized deployment:

1. Build the Docker image:

   docker build -t malaria-cell-detection .

2. Run the container:

   docker run -p 8888:8888 -v $(pwd)/cell_images:/app/cell_images malaria-cell-detection

   • 🌐 Access Jupyter at http://localhost:8888
   • 📦 Mount the dataset as a volume for large data

☁️ Deployment

• Local: Use Docker or run the notebook directly
• Cloud: Deploy the Docker image to any cloud provider supporting containers (AWS, GCP, Azure, etc.)
• Note: Ensure the dataset is available in the container or mounted as a volume

📦 Requirements

• Python 3.8+
• See requirements.txt for all Python dependencies:
  • tensorflow
  • numpy
  • matplotlib
  • scikit-learn
  • pillow
  • opencv-python
  • tqdm

📊 Results

• Custom CNN: ~93% validation accuracy
• VGG16 Transfer Learning: Up to ~96% validation accuracy after fine-tuning
• Visualization: The notebook includes code to visualize predictions and model performance
• Fine-Tuning: Provides a clear boost in accuracy and generalization for the malaria cell dataset

📄 License

MIT License

---

Keywords: 🐳 Docker, 🚀 Deployment, 🦠 Malaria, 🤖 Deep Learning, 🧠 TensorFlow, 🏛️ VGG16, 🔓 Transfer Learning, 🔬 Fine-Tuning, 🧬 Cell Classification