Blood Cell Classification using Biomedical Image Processing

This repository presents a biomedical engineering project focused on blood smear microscopy image classification using image processing and deep learning.

The project classifies white blood cell microscopy images into four categories and also supports a secondary binary classification task based on nuclear morphology.

---

Important Note

This project is intended for educational and portfolio purposes. It is not a medical diagnostic tool and should not be used for clinical decision-making.

---

Project Overview

The objective of this project is to support automated blood cell analysis through biomedical image processing and convolutional neural networks.

Classification Tasks

1. Four-class white blood cell classification

• Eosinophil
• Lymphocyte
• Monocyte
• Neutrophil

2. Binary nuclear morphology classification

• Mononuclear cells
• Polynuclear cells

This project is relevant to biomedical engineering because it addresses automated microscopy image interpretation, a task associated with hematology support systems, digital pathology, and AI-assisted clinical workflows.

---

Dataset

This project uses the Blood Cell Images dataset by Paul Mooney, available on Kaggle.

The dataset contains approximately 12,500 augmented blood cell images distributed across four white blood cell classes:

• EOSINOPHIL
• LYMPHOCYTE
• MONOCYTE
• NEUTROPHIL

Expected local dataset structure:

data/
└── dataset2-master/
    └── dataset2-master/
        └── images/
            ├── TRAIN/
            │   ├── EOSINOPHIL/
            │   ├── LYMPHOCYTE/
            │   ├── MONOCYTE/
            │   └── NEUTROPHIL/
            └── TEST/
                ├── EOSINOPHIL/
                ├── LYMPHOCYTE/
                ├── MONOCYTE/
                └── NEUTROPHIL/

The full dataset is not included in this repository. Please download it from the original dataset source and place it locally inside the data/ folder.

---

Verified Dataset Counts

The dataset structure was verified with the following image counts:

Split	Eosinophil	Lymphocyte	Monocyte	Neutrophil	Total
Train	2497	2483	2478	2499	9957
Test	623	620	620	624	2487
Total	3120	3103	3098	3123	12444

---

Methodology

The implemented workflow includes:

1. Loading microscopy images from the dataset folders.
2. Resizing images to 80 x 60 pixels.
3. Converting images from BGR to RGB.
4. Normalizing pixel values to the range [0, 1].
5. Encoding labels for multiclass or binary classification.
6. Training a convolutional neural network using TensorFlow/Keras.
7. Evaluating performance with accuracy, loss curves, classification reports, and confusion matrices.

---

Model Architecture

The CNN model includes:

• Convolutional layers
• Batch normalization
• Max pooling
• Dropout regularization
• Dense classification layers
• Softmax output

The model is implemented in:

src/model.py

---

Repository Structure

.
├── README.md
├── LICENSE
├── requirements.txt
├── .gitignore
├── assets/
│   ├── figures/
│   └── results/
├── data/
│   └── README.md
├── docs/
│   ├── dataset_notes.md
│   ├── methodology.md
│   └── project_summary.md
├── models/
├── notebooks/
├── outputs/
│   └── README.md
└── src/
    ├── README.md
    ├── __init__.py
    ├── check_dataset.py
    ├── config.py
    ├── data_loader.py
    ├── dataset_utils.py
    ├── main.py
    ├── model.py
    ├── predict.py
    ├── smoke_test.py
    ├── train.py
    └── visualization.py

---

Installation

Clone the repository and install the required dependencies:

pip install -r requirements.txt

If OpenCV fails on Windows, reinstall it with:

pip uninstall opencv-python opencv-contrib-python -y
pip install opencv-python

---

Usage

1. Check dataset structure

python src/check_dataset.py --data-dir data/dataset2-master

2. Run a fast smoke test

python src/smoke_test.py --data-dir data/dataset2-master --task multiclass --samples-per-class 5

3. Train the four-class model

python src/main.py --data-dir data/dataset2-master --task multiclass --epochs 20

4. Train the binary nuclear morphology model

python src/main.py --data-dir data/dataset2-master --task nuclear --epochs 20

5. Predict a single image

python src/predict.py --model outputs/trained_model.keras --image path/to/image.jpeg

---

Outputs

After training, the project generates:

outputs/
├── accuracy_curve.png
├── loss_curve.png
├── confusion_matrix.png
├── classification_report.txt
├── dataset_summary.json
├── training_logs.npy
└── trained_model.keras

---

Visual Results

Example Cell Classes

Preprocessing Example

Classification Output Example

---

Technical Skills Demonstrated

• Biomedical image processing
• Microscopy image analysis
• Deep learning with CNNs
• TensorFlow/Keras model training
• Image preprocessing with OpenCV
• Model evaluation with confusion matrices and classification reports
• Healthcare-oriented AI workflow

---

Reproducibility Status

• Dataset structure verified.
• Image loading and preprocessing pipeline verified.
• CNN training pipeline implemented.
• Output generation configured for learning curves, classification reports, and confusion matrices.

---

Future Work

• Add Grad-CAM or saliency map visualizations.
• Compare CNN results with transfer learning architectures.
• Add a lightweight web demo for image inference.
• Improve model reproducibility with fixed seeds and experiment tracking.
• Evaluate model performance on external blood smear datasets.

---

License

This repository is licensed under the MIT License.

Dataset rights belong to the original dataset creators. Please refer to the original dataset page for usage rights and redistribution conditions.