White Blood Cell Classification and t-SNE Visualization

Project Overview

This project focuses on classifying white blood cells (WBCs) using deep learning and visualizing their features using t-SNE (t-distributed Stochastic Neighbor Embedding). The white blood cells analyzed include:

• BA (Basophils): White blood cells involved in immune responses, particularly allergic reactions and parasitic infections.
• EO (Eosinophils): Combat parasitic infections and contribute to allergic reactions.
• LY (Lymphocytes): Provide immune defense. Different types include B cells, T cells, and NK (natural killer) cells.
• PLATELET (Thrombocytes): Vital for blood clotting and wound healing.
• SNE (Segmented Neutrophils): The most abundant mature neutrophils that ingest bacteria and fungi.

Sample Cell Images

t-SNE Visualization

Getting Started

Follow these steps to set up and execute the project:

1. Download the Dataset

You can access the blood cell dataset here.

2. Unzip the Dataset

Extract the contents of the downloaded .zip file to a designated folder.

unzip blood\ cells.v1i.coco.zip -d dataset_original

3. Preprocess the Dataset

Run the pre_process_dataset.ipynb Jupyter notebook to preprocess the data before training.

4. Train the Model

Use the transfer_learning_tutorial.ipynb notebook to train the classification model. This notebook is adapted from the PyTorch transfer learning tutorial.

A pre-trained model checkpoint is available here. Download it and place it in the ckpt/ directory.

5. Visualization

Use the wbc_resnet_tsne.ipynb or wbc_resnet_tsne_resnet34.ipynb notebook to visualize the extracted features using t-SNE. The code is adapted from the tutorial.

Examples of visualizations can be found in ./visualizations/