Deep Learning in Healthcare: 2023-2024
Practical Coordinator | |
Lecturer | |
Degrees | Schedule B1 (CS&P) — Computer Science and Philosophy Schedule A2 — Computer Science Schedule B1 — Computer Science |
Term | Hilary Term 2024 (16 lectures) |
Overview
Neuroscience research has inspired deep learning, and with the increasing digitisation of the medical domain, deep learning is set to advance healthcare. Deep learning has the potential to transform healthcare in areas ranging from medical imaging to electronic health records. This course aims to introduce students to recent advances in deep learning in healthcare, and the application of deep learning algorithms to medical data, including practical considerations for adapting models to the diversity of healthcare data. We will first introduce the biological basis for neural networks, the foundations of neural networks, and how the numerical operations are based on concepts from linear algebra, continuous mathematics, and probability, and how they are applied to train supervised models. We will then cover the computational components (e.g. convolutional neural networks) that form the backbone of powerful tools in deep learning, and how these can be deployed in the context of a variety of medical data (e.g. images). We will then cover networks that exploit the sequential (or temporal) structures in the data. Throughout the course, we will cover the practicalities of training neural networks, focusing particularly on applications in the healthcare domain, including discussion of optimisation, scalability, privacy, and fairness.Learning outcomes
The goal of this course is to provide an intuition for adapting deep learning algorithms to healthcare data and understanding the subtleties in applying these methods to real-world data. We will also discuss open challenges for future research.
In this course, students will:
- Broaden knowledge and fluency about state-of-art deep learning algorithms.
- Develop practical ability to design and train neural networks, and understand how to adapt models to diverse types of healthcare data.
- Understand practical considerations and domain-specific challenges associated with the use of medical data.
- Demonstrate how to systematically explore a basic deep learning problem.
Prerequisites
The emphasis of this introductory course will be on the application of deep learning to healthcare. Nonetheless, it is important to have good mathematical background in the following topics:
Essential (as covered in first-year Computer Science course)
- Linear Algebra
- Continuous Mathematics
Desirable
- Proficiency in Python programming or significant experience with an alternative programming language are essential for the practical sessions and the mini-project examination. Experience with Pytorch is desirable
Synopsis
Introduction
History of deep learning, and overview of healthcare applications
Connectionism, feedforward networks, activation functions
Backpropagation, optimisation techniques
From ANNs to CNNs
Sparse connectivity, weight-sharing, convolutional neural networks (CNNs)
Residual blocks, encoder-decoder architectures
Working with medical data
Practical considerations, data-handling
Techniques for low-data regimes (few-/low-shot learning, domain adaptation, augmentation, dropout)
Performance metrics (Dice, ROC, AUC, Hausdorff)
Sequence models, attention, and transformers
Recurrent neural networks, backpropagation through time (BPTT)
Long-short-term memory (LSTM) and Gated recurrent units (GRUs)
Attention, self-attention, transformers, and visual transformers
Explainability, privacy, and fairness
Saliency analysis, uncertainty, out-of-distribution detection
Federated learning, differential privacy
Continual learning
Reading list
The application of deep learning to healthcare data is still expanding, and there are no textbooks that cover this topic in great depth. As such, the material will be self-contained within the lecture slides, with several references to relevant academic articles.
Where relevant, the lecture slides may refer to the following textbooks:
- Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016
- Raul Rojas. Neural Networks: A systematic introduction. Springer-Verlag, 1996
Taking our courses
This form is not to be used by students studying for a degree in the Department of Computer Science, or for Visiting Students who are registered for Computer Science courses
Other matriculated University of Oxford students who are interested in taking this, or other, courses in the Department of Computer Science, must complete this online form by 17.00 on Friday of 0th week of term in which the course is taught. Late requests, and requests sent by email, will not be considered. All requests must be approved by the relevant Computer Science departmental committee and can only be submitted using this form.