Deep Learning for Medical Imaging module (BE41003)

This module covers the fundamental concepts and advanced methodologies of machine learning for imaging and relates those to real-world problems in medical imaging and computer vision. Students will experience different approaches to machine learning including supervised and unsupervised techniques with an emphasis on optimization and deep learning methods as convolutional neural networks and generative adversarial networks. Applications include X-ray CT, MRI and ultrasound image reconstruction and analysis such as classification, segmentation, and registration applied to datasets in healthcare. The aim of the module is to equip students with the skills needed to work in, and conduct research into, image computing and applied machine learning in medicine. During practical sessions, the students will be presented with medical applications and will be working in a team. Moreover, group projects under the guidance of the instructor help to assess the ability to transfer theoretical knowledge into algorithmic development and implementations.



Indicative content:



Part 1 Introduction to Linear Regression and Optimization

• Introduction to linear regression
• Convex optimization and regularization
• Stochastic Gradient descent
• Iterative numerical solvers

Part 2 Deep Learning

• Machine learning basics
• Deep feedforward networks
• Convolutional Neural Networks (CNN)
• Generative Adversarial Network (GAN)

Part 3 Advanced Methods for image analysis

• Deep recurrent neural networks
• Equivariant learning
• Transformers
• Support vector machines (SVM)
• Future trends, limitations, and challenges

Part 4 Deep Learning for Image reconstruction and segmentation

• Introduction to CT, MRI reconstruction
• Unrolling deep iterative methods for image reconstruction
• Supervised and unsupervised learning
• Generalization and overfitting with medical images
• SVM for image segmentation and analysis.