🔬 Malaria Cell Detection Using Convolutional Neural Networks

St Mary's University Twickenham London
CSO7013 Machine Learning — Mid-Term Module Assessment

---

📚 About This Project

This repository contains my Mid-Term Module Assessment for the Machine Learning (CSO7013) module at St Mary's University Twickenham London.

Student	Nevin Tom
Student ID	2517238
Module	CSO7013 Machine Learning
Assessment	Mid-Term Module Assessment
University	St Mary's University Twickenham London

---

🎯 Project Overview

This project implements a Convolutional Neural Network (CNN) to detect malaria-infected blood cells from microscope images. The model classifies thin blood smear images as either:

• 🦠 Parasitized —> Infected with malaria parasites
• ✅ Uninfected —> Healthy cells

Why This Matters

Malaria causes over 600,000 deaths annually (WHO, 2023). Diagnosis requires trained microscopists to manually examine blood samples — a significant bottleneck in resource-limited areas. This CNN achieves 96.42% accuracy and can assist in automated screening.

---

📊 Results

Metric	Value
Accuracy	96.42%
Precision	0.9560
Recall	0.9749
F1 Score	0.9654
ROC-AUC	0.9936

Baseline Comparison

Model	Accuracy	Improvement
Logistic Regression (Baseline)	61.96%	—
CNN	96.42%	+34.46%

Training & Evaluation Visualizations

Training History	Confusion Matrix	ROC Curve

---

🚀 Quick Start

1. Clone the Repository

git clone https://github.com/Nevvyboi/MalariaCellDetection.git
cd MalariaCellDetection

2. Set Up Environment

# Create virtual environment
python -m venv venv

# Activate it
venv\Scripts\activate        # Windows
source venv/bin/activate     # Mac/Linux

# Install dependencies
pip install -r requirements.txt

3. Download Dataset

1. Download from: Kaggle - Malaria Cell Images
2. Extract and place in data/cellImages/:

data/
└── cellImages/
    ├── Parasitized/     ← 13,779 infected images
    └── Uninfected/      ← 13,779 healthy images

4. Run the Model

# Quick test (3 epochs, ~2 minutes)
python run.py --quick

# Full training (35 epochs)
python run.py

# Predict on random image
python run.py --predict-random

---

📁 Project Structure

MalariaCellDetection/
│
├── run.py                  # 🚀 Main entry point
├── README.md               # 📖 This file
├── requirements.txt        # 📦 Dependencies
├── LICENSE                 # ⚖️ MIT License
│
├── src/                    # 📁 Source code
│   ├── config.py           # ⚙️ Hyperparameters & settings
│   ├── dataset.py          # 📊 Data loading & preprocessing
│   ├── model.py            # 🧠 CNN architecture
│   ├── train.py            # 🏋️ Training loop
│   ├── evaluate.py         # 📈 Evaluation & visualization
│   └── baseline.py         # 📉 Baseline model
│
├── data/                   # 📂 Dataset (download separately)
│   └── cellImages/
│
├── models/                 # 💾 Saved model weights
│   └── bestModel.pth
│
└── outputs/                # 📊 Generated visualizations
    ├── trainingHistory.png
    ├── confusionMatrix.png
    └── rocCurve.png

---

🧠 Model Architecture

Input: RGB Image (128×128×3)
         │
         ▼
┌─────────────────────────────┐
│  Conv Block 1 (32 filters)  │
│  Conv → BatchNorm → ReLU    │
│  MaxPool(2×2)               │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Conv Block 2 (64 filters)  │
│  Conv → BatchNorm → ReLU    │
│  MaxPool → Dropout(0.25)    │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Conv Block 3 (128 filters) │
│  Conv → BatchNorm → ReLU    │
│  MaxPool(2×2)               │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Conv Block 4 (256 filters) │
│  Conv → BatchNorm → ReLU    │
│  MaxPool → Dropout(0.25)    │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Fully Connected (512)      │
│  ReLU → Dropout(0.5)        │
└─────────────────────────────┘
         │
         ▼
┌─────────────────────────────┐
│  Output (2 classes)         │
│  Parasitized / Uninfected   │
└─────────────────────────────┘

Hyperparameters

Parameter	Value
Input Size	128×128
Batch Size	64
Epochs	35
Learning Rate	0.001
Optimizer	Adam
Loss Function	CrossEntropyLoss
LR Scheduler	ReduceLROnPlateau
Early Stopping	7 epochs
Random Seed	42

---

💻 Command Line Options

Command	Description
python run.py	Full training (35 epochs)
python run.py --quick	Quick test (3 epochs, 2000 images)
python run.py --epochs 50	Custom epochs
python run.py --evaluate-only	Evaluate saved model
python run.py --predict "image.png"	Predict single image
python run.py --predict-random	Predict random test image

---

📋 Requirements

torch>=2.0.0
torchvision>=0.15.0
numpy>=1.24.0
matplotlib>=3.7.0
seaborn>=0.12.0
scikit-learn>=1.2.0
tqdm>=4.65.0
Pillow>=9.5.0

---

📖 Dataset

NIH Malaria Cell Images Dataset

Property	Value
Source	National Institutes of Health (NIH)
Platform	Kaggle
License	Public Domain (US Government Work)
Total Images	27,558
Parasitized	13,779 (50%)
Uninfected	13,779 (50%)

---

🔧 Troubleshooting

❌ "Dataset not found" error

Ensure the folder structure is correct:

data/
└── cellImages/
    ├── Parasitized/
    └── Uninfected/

❌ "3 classes detected" error

Delete any extra folders inside cellImages/. Only Parasitized and Uninfected should exist.

❌ CUDA out of memory

python run.py --batch-size 32

❌ Training too slow

Use quick mode for testing:

python run.py --quick

---

📚 References

1. Dataset: National Library of Medicine. (2018). Malaria Cell Images Dataset. Kaggle. https://www.kaggle.com/datasets/iarunava/cell-images-for-detecting-malaria

2. Paper: Rajaraman, S., et al. (2018). Pre-trained convolutional neural networks as feature extractors toward improved malaria parasite detection in thin blood smear images. PeerJ, 6, e4568.

3. Statistics: World Health Organization. (2023). World Malaria Report 2023.

---

📄 License

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

---

Made with ❤️ for CSO7013 Machine Learning

St Mary's University Twickenham London | 2025