/UTokyo_CNN_MRI

This repository is for MRI and MNIST data CNN controller and related experimental scripts. Some are out of date and are strictly for informational and historical records.

Primary LanguagePythonMIT LicenseMIT

UTokyo_CNN_MRI

MRI data CNN controller follow-up procedures.

  • Adding more convolutional layers to the CNN.
  • Incorporating Batch Normalization for better convergence.
  • Using Callbacks for early stopping and model checkpoints.
  • Adding visualization for feature maps after pooling.
  • Visualizing gradient updates during training.
  • Implementing a custom callback to visualize optimization progress.

Modular CNN for MRI and MNIST Data

This repository contains a modular Convolutional Neural Network (CNN) designed to test different pooling techniques, gradient functions, and feature classification strategies for MRI and MNIST data.

Table of Contents

Features

  • Modular Design: Easily swap out different components of the CNN, such as pooling layers or optimizers.
  • Support for Multiple Pooling Techniques: Test with Max Pooling or Average Pooling.
  • Customizable Optimizers: Choose between Adam and SGD optimizers.
  • Feature Classification: Currently supports dense layers for feature classification.

Installation

  1. Clone the repository:

    git clone https://github.com/Austin-TheTrueShinobi/UTokyo_CNN_MRI.git
cd UTokyo_CNN_MRI

  2. Install the required packages:

    pip install tensorflow

VSCode Setup

  1. Installation: If you haven't installed VSCode, download it from Visual Studio Code.

  2. Recommended Extensions:

    • Python: Provides enhanced support for Python, including debugging, linting, IntelliSense, etc.
    • TensorFlow Snippets: Offers useful snippets for TensorFlow.
    • Markdown All in One: Essential for editing README.md files.
    • GitLens: Supercharge the Git capabilities built into Visual Studio Code.

  3. Environment Setup:

    • Ensure you have Python 3.x installed.
    • Set up a virtual environment: 
      python -m venv venv
    • Activate the virtual environment:
      • Windows: .\venv\Scripts\activate
      • MacOS/Linux: source venv/bin/activate
    • Open the project folder in VSCode.
    • When prompted, select the Python interpreter located inside the venv directory.

MRI Data Setup from IDA

The Image and Data Archive (IDA) at the Laboratory of Neuro Imaging (LONI) is a data repository that hosts a variety of neuroimaging datasets. To utilize MRI data from IDA for our project:

Registration and Data Access:

  1. Create an Account:

    • Visit the IDA database and click on Register.
    • Fill in the required details and complete the registration process.
    • Note: Some datasets may require additional permissions or agreements to access.

  2. Browse and Select Dataset:

    • Once logged in, navigate to the Collections tab.
    • Browse through available datasets or use the search functionality to find specific MRI datasets.
    • Click on the desired dataset to view more details.

  3. Download Data:

    • Within the dataset page, you'll find download options. Depending on the dataset, you might find raw MRI scans, pre-processed data, or both.
    • Download the MRI data files, which are typically in NIfTI format (.nii or .nii.gz).

Preparing Data for the Project:

  1. Organize Downloaded Data:

    • Create a folder named data within the project directory.
    • Extract and place the downloaded MRI data into the data folder. Ensure you maintain a structured directory if there are multiple categories or types of scans.

  2. Data Loading in the Script:

    • The main.py script should be set up to load data from the data directory. If the MRI data has multiple categories or is divided into training and testing sets, ensure the script's data loading section reflects this structure.
    • For NIfTI files, you might need additional libraries like nibabel to read the data: 
      pip install nibabel

  3. Pre-processing:

    • MRI data often requires pre-processing before it can be used for training. This might include normalization, resizing, or augmentation.
    • Ensure the main.py script or any other pre-processing script you use is tailored to handle the specific characteristics of the MRI data from IDA.

Usage:

  1. Run the Model:

    • With the data in place, you can proceed to run the model: 
      python main.py

  2. Model Training:

    • The script will load the MRI data, preprocess it if necessary, and then train the CNN model on it.
    • Monitor the training process, and once completed, evaluate the model's performance on any test or validation data you've set aside.

Contributing

  1. Fork the repository on GitHub.
  2. Clone the forked repository to your machine.
  3. Make your changes and commit them to your fork.
  4. Push your changes to GitHub.
  5. Submit a pull request to the main repository.

License

This project is licensed under the MIT License. See the LICENSE file for more details.

Note: Replace your_username with your actual GitHub username in the clone URL.