Hello and welcome back to a new week on image and video processing. The topic of this week is medical imaging. Medical imaging is a fascinating area. The challenges are incredible and so are the rewards. When we solve problems in medical imaging, we are basically helping humanity and that's a great reward. Now, medical imaging is very, very different than many other areas of image processing. One of the differences is that very often, we start the investigation, we started working medical imaging with a problem presented to us. Let's see a couple of examples of that. Here, we see images of skin. Basically, just regular pictures of the skin with potential lesions, potential cancer lesions. So, maybe a doctor knocks the door in our office and says, could you please help me? What I need to do is I need to automatically segment out the lesion. I want to measure it. I want to measure its area. I want to measure some of its characteristics inside and sometimes I even want to track it. I want the patient to come again, let's say, after a year or after a few months or after a certain period and control how that lesion, how that region is changing. but the problem is very specific. So, you think, and you say, I know I have learned how to do image segmentation. One of those techniques is active contours. So, you can start with the contour and you can let that curve deform towards the boundary of the object, this yellow curve, and all these three images were actually segmented out with active contours. You might think, oh, but this image is actually extremely noisy. These are hertz. But these hertz are, from the point of view of image processing, noise. But you also know, know how to do image denoising. So maybe before you apply active contours, you basically do image denoising. So, this is a very important application. And a particular example where we are applying our knowledge in image processing to a particular problem. Let's see another example. Here is another example in the area of cystic fibrosis. And without going into too many details, this is a patch that is basically put on the skin and it basically measures the amount of sweat and the concentration of chloride in the sweat. And the basic idea, the challenge here is to detect this curve and detect this curve and compute the ratio of the areas that are inside this curve and inside this curve. So, once again, the scenario is that a doctor knocks your door and asks you, explains to you the problem, and asks you a particular question. And says, you need to detect this curve and you need to detect this curve for me. Once we detect the curves, we can compute the area and that area gives us information about the concentration of chloride in the sweat. So, you think for a while and you say, this curve looks very elliptical to me. Can I assume that that's an ellipse or a circle? You ask the doctor, this is a process of communication between two different disciplines. If the answer is yes, you say, I know how to detect those type of curves, they're half transformed, the half transformed is great to detect parametric curves. So, you apply a half transform not to the image but to the edges of the image and you get this curve, perfect. Now, this is non-parametric so you can either apply active contours or you can try obtuse threshold algorithm to see if you can get this region. Once you get this region, you compute the ratio of areas and you have solved an extremely important problem. Now often, in medical imaging, we don't really attack problems like these ones that I just mentioned, but we basically address basic research issues with data that comes from medical data or biological data. An example of that is here. These are basically surfaces that represent the gray matter in our brain. In the brain of the subject that went into an MRI scanner. And here, we see what are called the sulci. So basically, our brain, our gray matter is folded. And these are basically the bottom of the folds. And from, for many reasons in neuroscience, these curves are extremely important. So, here is a challenge for image processing in 3 dimensions, how to automatically compute these sulci curves not for a particular medical problem, or medical application, but for a basic research, basic science, basic neuroscience question. Once you compute these, you can create very interesting networks of these sulci, which again, have a very important neuroscience problem, is a very important information for many neuroscience applications. Some of them actually related to particular problems, but some of them are related to research. So, these are just some examples of medical imaging. We're going to see much more in the next videos. But one of the fascinating areas of medical imaging is that it's a constant learning experience. This is not done just by people that are experts in image processing, you have to work very close with doctors, with people in Biology, with neuroscientists, with people that basically have the medical imaging challenges, and that basically forces you to learn new problems all the time, and forces them to learn new techniques, and basically forces both groups to work close with each other, which actually is a lot of fun because as I say, it's a constant learning experiance. I'm going to try to teach you some of the challeges in medical imaging. It's a huge area so I'm going to pick a few examples in the next videos to illustrate some of the problems, some of the challenges, and some of the rewards when you work in medical imaging. See you in the next videos. Thank you very much.
