Lesson finished
Up next: Defining Image Segmentation and Comparing Segmentation Approaches
Unlock all 48 lessons of Advanced Medical Image Processing in Clinical ApplicationsFull course ₹40,000 · paid
Medical Image Segmentation: Motivation and Clinical Applications
A free lesson from Skill-Lync's Advanced Medical Image Processing in Clinical Applications course, on Grow by Skill-Lync · Last updated 16 Nov 2026 · Session: “Module 7 Session 1”
What will you learn in this free Advanced Medical Image Processing in Clinical Applications lesson?
The module opens with its outline and defines image segmentation as finding object boundaries. It contrasts natural and medical images, explains why manual annotation needs clinical training, and presents applications such as 3D bone visualization, vasculature analysis and radiation therapy planning.
- What image segmentation is and why it matters
- Natural versus medical image acquisition
- Why manual annotation is hard and costly
- Applications: bone 3D views, vasculature and radiation therapy
What comes after lesson 23?
The next free lesson is lesson 24, “Defining Image Segmentation and Comparing Segmentation Approaches” (12 min). Later in Skill-Lync's Advanced Medical Image Processing in Clinical Applications course (paid): lesson 26, Module 7 Session 4; lesson 27, Module 7 Session 5; lesson 28, Module 8 Session 1. The full course has 48 lessons (21 h 26 min of video); see the course overview or its price and syllabus.
Lesson transcript
Machine transcript of the lesson audio (lightly cleaned). Technical terms may contain errors; the video is the reference.
Hello everyone. Welcome to the course on advanced medical image processing. So in this series of lectures, we are going to discuss about the first module on image segmentation, where we will cover some classical image segmentation methods. This is a summary of the table of contents. So first we'll start out with the motivation and applications of image segmentation, especially in the context of medical image analysis. Then we will look at different image segmentation techniques and go deeper into a few selected methods, which are very popular and robust, even in this era of deep learning. Then finally, we will also discuss how we can evaluate segmentation accuracy of different methods. At the end of this lecture, we expect that we would have learned the following items.
We should have got an overview of classical image segmentation methods and also get a deeper understanding of selected methods. And we would also learn how to write our own algorithms for medical image segmentation. So starting out with motivation and applications. First of all, let us understand what exactly is image segmentation. Image segmentation is the process of identifying the boundaries of different objects in an image. The images can be of natural images or medical images. So natural images are those taken from in a natural setting like the room or the outdoor sceneries and all where we have the different objects that we encounter in our day to day activities.
Whereas in medical images, we get the scans of specific anatomies of the body. First of all, we need to understand there is a difference in the acquisition procedure for natural and medical images. For natural images, we are all familiar with the cameras that we have using CCD devices and all. But for acquiring medical images, there is a completely different physics involved here. There are specialized instruments or devices that have manufactured and which have specialized characteristics like they generate a specific magnetic field in the case of magnetic resonance images or in the case of X-rays or CT images.
They emit specific rays at a specified wavelength. Therefore, the characteristics of medical images is completely different from those of natural images. Now, medical images, the whole purpose of acquiring medical images is to get a better view of our internal working of different organs and if there are any abnormalities, etc. So, hence, when we receive a medical image, it is important to very accurately annotate or identify those regions which are abnormal or have some kind of disease activity, some tissue damage, anything, any kind of abnormal activity over there.
In order to annotate or identify the boundaries of objects in natural images, it does not require a lot of training because we are accustomed to seeing these objects in our day-to-day life and we can very easily identify which is a table, a chair or a car or a tree. However, for medical images, in order to identify those objects, the person who identifies has to have a high degree of clinical training. He should be able to distinguish what is normal and what is abnormal, where is the desired organ, which is background, which is muscle, which is tissue and all that.
So, even after this detailed training, a manual annotation of medical images is very time consuming. Plus, as we discussed, it also requires a high degree of clinical expertise. For that purpose, doctors who have had a significant degree of experience with working with medical images, they have to do this task. However, when they give time to annotating medical images, it takes away a lot of that time, their valuable time, which can be devoted to doing something more critical. For example, analyzing or taking care of patients and all.
So, this is the motivation behind having an automated approach to image segmentation. Then there is another issue of reproducibility. If I give a doctor one image or a set of images to annotate on day one and after a few days or a few weeks, I go back to him and ask him to annotate the same set of images again. There will be differences between the annotation set one and annotation set two. This is called intra-rater variability, a phenomenon that is very common in clinical data annotation.
This is quite natural in one sense because as human beings, we don't exactly remember what we did a week ago or a few weeks ago. And there is always a certain level of subjectivity when annotating a particular image or a particular data set. While there may not be a very big difference in the two sets of annotations, when it comes to validating the accuracy of the annotations or understanding which is more accurate, it creates confusion. Should we consider annotation set one primary reference or should we consider annotation set two as the primary reference?
Hence, another requirement for automated algorithms is that they should be highly reproducible. which means that no matter how many times I run the algorithm on the same image, it should give me the same result. This is very, very important, especially in automated algorithms because we want to make sure that they are robust and they are consistent. Plus, another important motivation for image segmentation algorithms or applying image segmentation to medical images is to leverage the many algorithms that have been developed for computer vision applications, like localizing object boundaries, extracting the images, all these from digital imagery.
So going forward, what are the applications of image segmentation? If you look at image segmentation as a overall tool, it has many applications, not just in medical images, but also in the context of natural images. So focusing our discussion on the context of medical images, very important application is content based image retrieval. So for example, suppose we have a patient coming in with a specific set of diseases or a specific kind of disease stage and the doctor wants to look at previous cases which have a similar condition.
And in order to do that, there has to be a reference point against which to compare the previous records. So with the help of image segmentation, what we can do is we can identify the anatomies which have a particular problem and then use that as a reference metric to go back to the database and search where we have previous histories of similar patients. How this helps is it helps us to understand how the disease might progress, what are the effects of different medications and referring to host of other data.
Another important application is tumor delineation, which means that we would want given a medical image, we want to identify the exact boundaries of a particular tumor. And doing so has a lot of uses, it helps the doctors to look more closely, decide whether to go for a surgery and what kind of therapeutics to apply. Then a third application is of course disease detection, detecting the exact region where there is disease activity. So, we want to identify the 3D reconstruction. So, we want to identify the 3D reconstruction.
Going from having a set of 2D images and constructing a 3D visualization is very important because it helps to identify where there are specific organs, where there are specific abnormalities, etc. Finally, motion tracking. Some organs like the heart, it is important to detect and monitor the motion of the left ventricle in particular because that helps to determine whether the heart is a particular. If the heart is functioning normally or abnormally, if there is any abnormality, what are the potential causes and what can be done to address this problem.
Now, let us take a closer look at some of the clinical applications that we mentioned earlier. So, first is 3D visualization. The image A, it shows CT scan. So, now those who are not familiar with CT scans, what they do is they are used to take very high resolution pictures of the anatomical regions. And for bones, it highlights the bones very well because it comes up as a white structure in the CT scan. If we were to follow the complete a quick break, as we be able to Remember, the background that I mentioned earlier as the way for theoretum side....
I will be looking for for 10, really long. Now, this particular image A, it shows a slice, a specific slice of the CT scan. CT scans are 3D volumes, and these are slice-by-slice visualizations. Now, the different bone structures that have been segmented are highlighted in different colors. Now, if we look at segmentation in image B, so we see the 3D structure of the bones of the pelvic region, and it gives a very good idea of the location of different bones and other structures. Now, if a surgeon were to look at it, he can easily understand the location of different bones.
If there are any abnormalities, and especially if he is planning to do a surgery or insert a surgical instrument, he can know where exactly to insert, where to position his instrument, so that there is minimal collateral damage. If it involves drilling into a bone, it should not harm any other surrounding muscle and tissues. So, this is a very important application combined with segmentation and graphics visualization. This is a very helpful tool for surgeons. Another important application is estimation of organ volumes. For the estimation of organ volumes, the main use case is in understanding whether the size of the organ is normal or whether it has shrunk or it has enlarged due to some abnormalities.
So, based on statistical analysis, there is very reliable data on what should be the organ volume with respect to the body size for different patients at different ages. So, by doing a segmentation, we can actually not just determine the area that is covered, but we can also determine the actual volume in the real physical space of that particular organ. And then, the other example of segmentation is that once we have this organ volume or the segmented organ, we can monitor the progress or the blood flow in it across a specific time and then understand whether this is normal or abnormal.
So, this is what is referred to as identifying physiological functions and determining abnormalities. This also helps in monitoring response to therapy and also understand whether the specific therapy is helping in checking the disease or if there is some progression in the abnormality. To get a better understanding, we will refer to these diagrams. The top row shows three images corresponding to the same patient and acquired on three different days, are day 34, 55, and 68. The bottom row shows the response to a particular physiological signal. Now, on day 34, which we consider as the reference day, the tumor has a specific volume and then the associated physiological signals are shown in the first column for the second row.
Then, for day 55, we see there has been a regression or the shrinkage of the tumor and which corresponds to a similar change in the physiological function. As we can see, the blue line corresponding to pyruvate, that is, the magnitude has increased, while the yellow line corresponding to lactate, the magnitude has remained more or less the same. Now, going to day 68, the third image, we see there is a relapse. There has been a slight increase in the tumor volume and that is reflected in the magnitude of the blue and yellow signals.
It is not just the magnitude, but also the time at which the peak occurs. You can notice that in the three plots, the peak occurs at different times. So, this also gives an indication of the degree of abnormality of the particular organ. Another interesting application of segmentation is vascular analysis. In this example, we are looking at an intracranial vascular disease model, which basically is a 3D model of the brain vasculature. Image A is the original image, the MR image. Then, image B is the segmented vasculature system.
Then, finally, image C shows a 3D visualization or a 3D rendering of that particular segmentation in image B. And, finally, image D shows the point cloud representation. Now, what is the advantage of doing all this? The advantage is that we can get a better visualization of the blood flow. And, by monitoring the blood flow, we can understand where exactly there are blockages, if any, whether there are any constrictions, whether there are any other abnormalities. And, then, finally, the point cloud representation, it helps us to simulate many different things, like simulate blood flow, simulate some kind of abnormality to understand how things will turn out in the future.
And, finally, we also talk about radiation therapy. So, in this context, for glioblastoma patients, how to do surgery and radiation therapy planning. Now, before a doctor goes into the actual radiation procedure or the therapy procedure, he needs to know the exact location of the brain tumor. Where exactly is the brain tumor located? So, he can position the radiation or dosage devices accordingly. He should also know if there are any overlapping with neighboring tissues or neighboring structures. For example, the big structure here is the ventricle. And, here, as you can see, is the glioblastoma.
Whether there is any significant overlap between the glioblastoma and the ventricle. So, accordingly, he has to plan the strategy for the surgery and the radiation. And, in all these cases, image segmentation is particularly important. So, here, we come to the end of the first part of week one. To recap, what did we learn? We got an idea of the motivation for having image segmentation. Got an idea of the learning objectives for this module. And, then, we also looked at the different clinical applications for medical image segmentation.
Starting from, why is medical image segmentation necessary? What is the point of having automated algorithms? And, what are the essential requirements of automated algorithms? Then, we also looked at some specific applications that would help you to understand the relevance of these techniques. And, those applications included 3D visualization of bone structures and of blood flow monitoring, vasculature analysis, then, finally, radiation therapy. So, we come to the end of this part. And, in the next part, we will go deeper into the image segmentation techniques. Thank you. Thank you.