 Medical imaging really makes visible what otherwise would remain hidden, and it has already become a key technology in medicine. With the modern 3D ultrasound, for example, I was able to see my daughter six weeks before she was born, and actually I got very well prepared for her mood. But the way doctors look at these images is quite different from how you and I would look at them. Doctors get very long and hard training to become incredibly skilled in seeing critical patterns in these images, to tell the difference between healthy development and disease, and to see whether a treatment works or not. But some diseases are so complex that looking at a single image is not enough to fully understand what's going on. And this is where it gets really tricky. So a brain injury, for instance, might show in one image. But to really understand, to get the full picture of the disease, you need to take multiple images, multiple scans with different techniques so doctors can make a decision. But there's a problem, because reading those complex images is not only very time-consuming, but it really pushes human abilities to the limit. And patterns of disease can often be so complex and subtle that it's really difficult to get it from the images. Brain tumors, for example, they can appear almost anywhere in the brain with very different sizes and structures and shapes. It's very difficult to understand exactly what's going on. And often it's not enough to make a tumor visible, but we need to measure what's going on. For instance, tumor size is very important to make a decision whether you do surgery or other options. But tumor size alone is not enough. We need to understand what's happening inside the tumor. For instance, do active parts become inactive when you give a particular drug? Now, this is really important information that doctors need to make the right decision at the right time. And without that information, there is a real risk that critical patterns of disease go undetected. How do we help doctors and make best use of these images? And putting it all together is really challenging, but this is the challenge that we are taking on in my lab at Imperial College. We ask, what if we could train a machine to help us read those images, to detect patterns that humans find difficult to see? And can we even go beyond this? Often we don't know what we are looking for in those images. What are the patterns that describe a disease? Can we build an intelligent machine that can help us find those patterns? Just a couple of months ago, we have all seen a machine beat yet another human world champion in yet another board game. But what was really exciting about this achievement was that the machine's skills were not acquired by programming, but by learning. So the go-playing machine has learned by playing millions of games to detect patterns of great gameplay. And our machine learns by viewing thousands of medical scans. And one of our first big achievements was being able to train a machine to recognize all major organs with an accuracy of a human expert. Now that's really important because now the machine understands what is visible in the image. So we took this further. We then trained a machine that could search through multiple scans and detect patterns of abnormality. We could highlight suspicious regions that might, for instance, indicate a cancer. And that's important because now we can help doctors not miss important patterns that otherwise by just human reading might have been missed. Recently we even went one step further. So we now trained a machine that can detect the most complex patterns of brain lesions and putting them all together to build a 3D map of the brain that would show those lesions with an unprecedented accuracy. Now these maps are so precise that we can start making the important measurements and give it to the surgeon to support their decision. But going from the lab to clinical practice comes with great challenges and it's about trust. Doctors, understandably, are very reluctant in taking advice from a machine that behaves like a black box. So understanding how a machine comes to a conclusion can be as important as what it decides. So for us as researchers that's really important. And we were curious how does our machine understand what a brain lesion is. So we designed it from the beginning so we could lift the lid and look inside and what we found was really exciting because that's what we didn't expect. When we looked how our machine did this the interesting thing to note is that we never gave any advice how to detect a brain lesion, how the brain even looks like but our machine figured out by itself that the brain consists of different structures like grey matter and white matter. We know that mapping those structures is important but the machine has figured that out by itself. This is what really convinces me that artificial intelligence has so much to offer in medicine. It will not only help us to routinely find critical patterns in very complex data it will also help us to learn something new to reveal insights about the most complex diseases and make the right decision when it really matters. Thank you.