HuBMAP_Vasculature_2023

Table of Contents

• Overview
• Data Source
• Quick Summary and Key Findings
• Future Considerations

🌄 Overview:

The competition aims to segment microvascular structures, including capillaries, arterioles, and venules, from 2D PAS-stained kidney histology images. Human bodies depend on the intricate organization of 37 trillion cells. To understand this, the Vasculature Common Coordinate Framework (VCCF) maps cells using blood vasculature, ranging from the whole body down to individual cells. However, limited knowledge about microvasculature leaves gaps in the VCCF. Automation in microvasculature segmentation could fill these gaps. The hosting entity, the Human BioMolecular Atlas Program (HuBMAP), is working on a global platform to map all human cells. With advanced biological technologies, HuBMAP explores cellular interconnections. Despite the vast unknowns of microvasculature, machine learning can amplify our understanding, pushing forward the development of the VCCF and the Human Reference Atlas (HRA).

🔢 Data Source

The full dataset can be found here at https://www.kaggle.com/competitions/hubmap-hacking-the-human-vasculature.

🛠 Tools

• Jupyter Lab
• Python
• PyTorch

📄 Notes

Makes sure to have the following folders:

1. Create a main 'images' folder within your main project folder
2. Then make sure to have the following folders:

• output
• test (where your training images are)
• train (where the train images are)
• train_images
• train_masks
• val_images
• val_masks

3. Make sure to have a 'saved_images' folder within your main project folder