vishal815 / Team-LogicLegends-Proxmed-Hackathon-Hypodense-Segmentation-AI-Project-

Project Overview:- The objective of this hackathon challenge is to develop a robust and efficient algorithm or AI model capable of accurately segmenting the hypodense region from Brain Non-Contrast Computed Tomography (NCCT) images. The primary goal is to automate and streamline the identification of early ischemic changes in acute stroke patients.

Team LogicLegends Proxmed Hackathon Hypodense Segmentation AI Project.

👆click on this link to open Project in Google Colab

download.mp4

Team Members

• Vishal Lazrus
• Ritesh Kumar Singh

Project Overview

The objective of this hackathon challenge is to develop a robust and efficient algorithm or AI model capable of accurately segmenting the hypodense region from Brain Non-Contrast Computed Tomography (NCCT) images. The primary goal is to automate and streamline the identification of early ischemic changes in acute stroke patients.

Project Structure

The project data has been organized in the following format:

Data Preprocessing Steps

1. Extracted zip data and converted it into the specified structure.
2. Preprocessed data to dimensions (128, 128, 128).
3. Performed additional data preprocessing.

Data Visualization Steps

1. Conducted 2D visualization with different axes.
2. Combined image and label for visualizing hypodense regions.
3. Utilized 3D visualization techniques and HTML5 video visualization.

Models Implemented

1. 3D U-Net Model (CNN)

   • Developed a convolutional neural network model for 3D data.
   • Training, testing, and evaluation were carried out.

2. V-Net CNN Model for 3D Data

   • Implemented an alternative CNN model for 3D data.
   • Conducted training, testing, and evaluation.

Additional Steps

• Explored different visualization techniques for better understanding.
• Used Kaggle to host and share data: Kaggle Data Link

Usage

Run the main script or notebooks for training and testing.

Install dependencies:

```bash
pip install -r requirements.txt
```

Prerequisites

Ensure you have the following libraries/modules installed:

• TensorFlow
• Seaborn
• Pandas
• html5lib
• NumPy
• Matplotlib
• nibabel
• scikit-image
• os

Contributions

1. Vishal Lazrus
2. Ritesh Kumar Singh

The results of this project are good, and Team Proxmed has provided a clear explanation of the problem. Our continuous research, experimentation, and tireless efforts, often extending into the late hours of the night, have been instrumental in this project. We have learned a lot from this project. Thank you 😃.

Clone/Download

To clone or download this project, you can use the following commands:

git clone https://github.com/vishal815/Team-LogicLegends-Proxmed-Hackathon-Hypodense-Segmentation-AI-Project-.git