This talk focuses on deep learning applied to 3D structural Magnetic Resonance Images (MRIs) of the human brain. It starts with a broad overview of deep learning for medical imaging including the challenges faced when working with 3D images. Subsequently, two ongoing research projects will be introduced. In the first (more theoretical) project, a novel weight-sharing technique is discussed that builds on reflection symmetry. The technique has been implemented in Tensorflow as a novel convolutional layer type. In the second (more applied) project, I will present our "noise-to-brain" model, a Generative Adversarial Network (GAN) that generates MRIs of the brain.

How do we insure the AI systems we design have clear intent in mind whilst providing consistent value to users and being mindful of ethical considerations? Some of the Systems AI team will cover design thinking and prototyping approaches taken when beginning to design an AI system alongside a range of our favourite examples across industry, research and arts.