NuCLS

Hugging Face Integration and Nuclei detection in breast cancer images using YOLOv8.

Welcome to the NuCLS Project, where I delve into the integration of complex biomedical image datasets with great deep learning tools. This project focuses on leveraging Hugging Face for seamless data handling, conducting in-depth exploratory data analysis (EDA), and applying advanced deep learning techniques for detecting and classifying cell nuclei in breast cancer images. By optimizing model performance and utilizing precise evaluation metrics, I aim to contribute to improved accuracy in medical image analysis.

Explore the full project, code, and documentation on my GitHub page: NuCLS Project on GitHub

Project Overview

In this project, I:

  • Integrated a large, complex breast cancer image dataset with detailed nuclei annotations into Hugging Face’s data management system, enabling efficient access, preprocessing, and augmentation.

  • Performed exploratory data analysis (EDA) to understand the dataset’s structure and characteristics, including the distribution of cancerous nuclei across different hospitals.

  • Implemented a YOLOv8 model, a type of convolutional neural network (CNN), to detect and classify cell nuclei in biomedical images.

  • Developed custom training pipelines using PyTorch and Hugging Face’s datasets and transformers libraries.

  • Optimized model performance through hyperparameter tuning, data augmentation, and regularization techniques.

  • Evaluated model performance using metrics such as Intersection over Union (IoU), Precision, Recall, and F1 Score.

Check out the repository for more information, including code, datasets, and instructions for replicating the experiments.