We developed an inclusive crash course for the introduction to Deep Learning (DL) and its medical applications. The goals of the lecture are:
- Demystify AI
- present applications to the medical domain
- show the power of DL
- introduce the challenges and current limitations of DL
The modularity of the lecture allows us to handle crowds of diverse, and even unknown, scientific backgrounds, and to adapt it to different formats.
The repo includes:
-
Teachable_machine tutorial: meant for learners with no or little technical knowledge. The goal is to make them aware of the power of a neural network in tasks like image classification, and the importance of a well structure dataset.
-
TensorFlow Playground tutorial: meant for learners with a basic knowledge of a neural network. The goal is to show the difference between linear and non-linear problems, how different features are learnt in the hidden layers and what proper weights initialisation means.
-
DL_Classification tutorial: this tutorial teaches how to deal with imbalanced data and understand the importance of considering not only global accuracy but also other metrics (accuracy per class, specificity, sensitivity).
-
check_flie.py: file with hints and solutions
-
utils.py: module with more complex functions, not directly included in the notebooks, but available for students with higher technical skills interested in the details.
-
requirements.txt: during the course the learners were asked to execute all the notebooks in Colab, but a requirement.txt file has been included in the repo.
The slides used to introduce the theoretical parts and the applications are available at https://zenodo.org/record/7053457#.Yx3RWRMzZJU.