Brain MRI Age Classification using Deep Learning (Course Integration Version)

This exercise has been modified from the original example in order to be able to integrate it into course/s at higher education institutions.

It shows how to work with an MRI brain image dataset and how to use transfer learning to modify and retrain ResNet-18, a pretrained convolutional neural network, to perform image classification on that dataset. The MRI scans used in this example were obtained during a study [1] of social brain development conducted by researchers at the Massachusetts Institute of Technology (MIT), and are available for download via the OpenNEURO platform: https://openneuro.org/datasets/ds000228/versions/1.1.0

Concept covered

Teaching end-to-end AI workflow for image classification
Applying transfer learning to a real-world data
Modifying a pre-trained network interactively and programmatically
Training the modified network to classify MRI images
Evaluating the model
Understanding network predictions using occlusion sensitivity maps

Suggested Audience

Instructors teaching any course in neuroscience and/or bio-medical disciplines which uses MRI data. The exercises can be used with students who have very little programming knowledge (use DND version) or with students who are familiar with programming (use non-DND version).

Time

The exercise can be run within 30-45 minutes.

How to Use

The package contains of 2 versions of the same exercise: one provides an interactive, app-based workflow (using Deep Network Designer=DND) and the other one a programmatic workflow. Both versions consist of working exercises where parts of the code have to be filled by students. The exercises are self-explanatory and contain all information required to execute it. If the students have very little programming knowledge, the app-based workflow is recommended. If the students have some programming knowledge, the programmatic workflow can be suitable. The exercise can be used as part of a course/module/lecture or as self-directed and self-paced exercise/homework/assignment.

The package contains a small subset of the original dataset. Using the original dataset (20+GB) would result in longer run times.

The solution files are available upon instructor request. If you would like to request solutions or have a question, contact the MathWorks online teaching team.

Products:

Release last tested

R2022a

References

[1] Richardson, H., Lisandrelli, G., Riobueno-Naylor, A., & Saxe, R. (2018). Development of the social brain from age three to twelve years. Nature Communications, 9(1), 1027. https://doi.org/10.1038/s41467-018-03399-2

w-boom/Brain-MRI-Age-Classification-using-Deep-Learning