Artificial Intelligence in Diagnostic Medicine Educational Modules

Overview: There are a multitude of online resources to teach interested individuals about the role and application of AI methods in the healthcare space. In general, these resources either require substantial time and energy commitments (such as Coursera, multi-week courses) or are too light to provide understanding (such as individual articles on https://towardsdatascience.com/). In addition, there are limited resources geared for practicing physicians and medical trainees.

→ To meet this need, we have developed a set educational materials to provide hands on experience with using AI methods for healthcare applications.

 

Educational Modules:

  1. Intro to Evaluation of Classification Methods
    Summary: Explores a binary classification dataset (UCI Breast Cancer) and train a neural network. We will then explore various evaluation metrics for testing the prediction output from the trained model.

  2. Advanced topics in the evaluation of diagnostic machine learning methods
    Summary: Presents some key questions and reg flags when evaluating the performance of machine learning algorithms for classification, with a focus on clinical diagnosis.

  3. Radiomics and Conventional Machine Learning for Classification
    Summary: Employs hand-crafted radiomics features to classify images

  4. Intro to Deep Learning for Medical Image Classification using PyTorch
    Summary: Creates a simple classifier for MedMNIST chest radiograph images to identify pneumonia

  5. Intro to Conventional Segmentation Methods
    Summary: Implements conventional methods, such as Canny edge detection and threshold-based methods, for medical image segmentation Tasks

  6. Deep Learning for Medical Image Segmentation (Keras TensorFlow)
    Summary: Trains and evaluates a U-Net in Keras for chest segmentation

  7. Deep Learning for Medical Image Segmentation (PyTorch)
    Summary: Trains and evaluates a U-Net in PyTorch for chest segmentation

Instructions:

  1. Click on the links above to direct you to the module's IPython notebook file in the google drive environment.
  2. Select the "Open with Google Colaboratory" button at the top to launch a free instance on a server.
    NOTE: If you do not see "Open with Google Colaboratory" at the top, you will need to connect this free app to your Google account. See instructions below.
  3. This will allow you to work through the modules without installing custom python software on your local system. You can also download the *.ipynb files and run them locally in Jupyter Notebook.
  4. Run each cell with the "Run" button to the left of each cell or with "Shift+Enter" keys.
  5. Please send feedback and suggestions for improvement to Adam Alessio.

Connecting Google Colaboratory to Google Account:

1. After clicking one of the links to the modules, you should see a website with "Open with Google Colaboratory" button as shown to right:

2. If not, select the drop down arrow and click on "Connect more apps" as shown to right:

3. Search for "Colaboratory" and select install (this installs only to your online google account. There is no impact to your local machine):

4. Finally, click on one of the modules above to refresh and you should now see the "Open with Google Colaboratory". After opening, it should look something like the image to the right:

Copyright © 2021, Michigan State University
www.midilab.org

Use of these modules assumes agreement with the following software license. This license may evolve as we distribute these modules through other outlets.
license_AI_DiagnosticMedicine_EducationalModules_v1.txt

These modules were made possible through a grant from the ARS Foundation. Thank you to the foundation for supporting the education of current and future healthcare innovators.

 

All modules were authored by members of the MIDI Lab. If you have comments or suggestions for these modules, please contact Adam Alessio (aalessio [at] msu [dot] edu).