A beginners guide to ImageJ (and Fiji) transcript

Hi, welcome to this channel which is not about playing the guitar. Unfortunately, my day job is bioiming teaching physiology and teaching pharmarmacology and I teach a bioiming undergraduate course and also a post-graduate course and last week um I was doing a walkthrough of image J for the bio imaging class and I thought at the end of it since we'd recorded the zoom I thought maybe this actually might be useful as a as a as a YouTube video.

Now, there are references to examples of work that the students need to do and some lectures that they're going to to have in the in the coming weeks. If you're watching this, just ignore that. And hopefully just this walk through of all the basic parts of image J might be of some use to you. Hope you enjoy this. [Music] [Laughter] [Music] Good. Okay. So, I wanted to start off then with um well, what's the difference between image J and Fiji and why would you why would you choose one over the other?

So, as you can see here, both image J um on the left hand side and FG on the right hand side. They look essentially the same um because FG really is just image J. It's just packaged up a bit differently, you know. So if we if we look at the file menu for both image J and for FG, there's maybe one or two more options in the Fiji with the edit that looks the same for the image menu. quite a few options here in MHJ and a lot more in Fiji.

And the biggest difference is probably within the plugins folder. You'll see here are this is pretty much the standard plugins that you would get with image J um with a couple of addition additional ones that I've put in that are necessary for the course. But in Fiji, this all comes as standard massive list of of plugins. So effectively Fiji is just image J but loaded with uh a ton of plugins. Now so what are the plugins?

Well the plugins are extensions to the basic program. And if you come to the um to the MageJ homepage which is this space here where you would download the the most recent versions of MageJ. Fiji has a a different um homepage, but we'll use the MSJ page for now. If you look in at the plugins for MSG, which are here, uh you'll see these are all of the available plugins at the moment. And there are hundreds. I haven't counted them.

So pretty much anything you could any task any image analysis task you may have there is the likelihood that there will be a plugin. Now in order to install these plugins you would simply um select one at random maybe this one here. Um and you'll see that there's a a Java file and a class. And normally you would just follow the instructions and download probably in this case the the Java. You would download that and simply put it in the plugins folder in image.j.

So if you have an installation of image J on your computer then you will have these folders and within the fold there'll be a plugins folder and you basically just drop those plugins into the plugins folder and then the next time you run image J that plugin that you've installed in the plugins folder should appear here in this menu. So that's basically that's relatively straightforward. Um it's the main difference between Fiji and MageJ about the plugins.

So let me I'm going to do the rest of this demo in image J rather than Fiji. But if you're choosing to use Fiji, then it's pretty much pretty much the same. So let me just get rid of Fiji for the moment. Uh okay. So that was image J and Fiji. Uh where we would and where we would put the plugin. So opening an image. Um let me just bring this folder in here. Now normally the way I would open an image would be simply just to grab an image and a data set and drag it on top of the image. uh the image icon the the program there are different ways of opening the files.

You can choose open here. Um often we might choose import and as you can see here there are a variety of different file formats that we could import. Um quite often we might open an image sequence and that's what we'll do in in just a few minutes. But first of all, let me oops, let me just open a a single a single file to begin with. So, I'll take a an image at random. Let's bring image J back. Uh, let's just say this image here.

Okay. So, I've got one single digital image opened. First thing we want to do is to talk about pixels and lookup tables. Now, I'd like you to look at the top of the image here because what this tells us is some very important information about the image. It tells us in this case that this image is 219. Let's call it 220 220x 220 microns. There are 512x 512 pixels in the image. It's an 8 bit image and the total size of this image is 256K.

If we zoom in, then we start to see the pixels. Now because this lookup table uh that I've chosen is quite dark, it's difficult to see these individual pixels. So one thing we could do is to change the lookup table. And basically that is just the the map of colors that are assigned depending on the particular intensity of a pixel. So let me just quickly go to the image menu and down to the lookup tables. And you see I have a a whole selection of of lookup tables.

Um I have some favorites in here, but the most dramatic one is possibly the spectrum lookup table. Now really what that's doing is it's looking at the intensity of a given pixel and assigning a particular color to that. So, if I move my mouse around the image, you'll see if you look up in in this window here, you'll see that there's a value. There's an X and Y coordinate and a value. So, every pixel has an XY coordinate and has a value which is its intensity, the intensity of the of the pixel.

And you see here that this particular pixel has a value of 49. If you can see that the darker one has a pixel value of five. If I was to use a lookup table which used a which a gray lookup table and let me just zoom out by selecting the magnifying glass and holding down the alt key. Okay. So sometimes the gray I find the gray lookup table actually probably the most useful. So every single pixel in this image has a value that ranges between zero and 255.

And the brighter pixels will have a value close to 255. And you see the value the pixel I'm pointing at or I'm at at the moment is 253. And if I go over to a dark one, it's got a value of 10. So the lookup tables, regardless of which one I choose, don't change the values. All they do is they change the way in which colors are mapped to that lookup table. There's an edge one, which is quite my favorite one is the smart because it's quite nice at bringing out bright objects, but it also keeps the grayscale.

So, it's worthwhile uh if you're new to this to have a look at the different lookup tables and just get a feel for which ones or what they do basically and and how and what the different colors are. So, that is pixels and lookup tables. I'm going to go back to just a regular um uh regular gray lookup table. And now we'll look at some filters. Now, are there any questions so far at this point? I'll just stop for a second.

Okay. Nothing in the chat box. So, filters, you'll find them up here. And we've got sharpen, find edges, and down here we have a few more to gauian blur, median, mean, minimum, and and so on. So when might we use these? Well, let's just try them and see what happens. But what I will do first of all is I will duplicate this image so that I always have an original. One thing to be aware of in MHJ is that the the undo facility is not great.

Sometimes there's an undo option. Often there isn't an undo option. So I do tend to to duplicate images quite a bit. So you can see that to duplicate an image is control shift D or we just choose duplicate. So I'm just going to duplicate that image. Um it duplicates it at its original scale. I just put it off to the side so that I've I've got a an image that I can return to. In fact, actually, I might duplicate a couple of them just in case.

Okay. So, what might the filters do? Well, first of all, we would have to look at this image and think, well, what do we think this image might benefit from? Um, it's a little bit noisy. You can see a lot of sort of grainy speckles here. perhaps um a lowass filter might work. Now, rather than lowass the whole image, for the purposes of this demonstration, I'm just going to try to lowass or do a filter in a small area of the image.

So, to do that, I'll take a rectangular box icon here, draw a box, and process. And let's uh smooth that. Okay. So you can see that what's happened if I zoom in and you can you can maybe get a nice a nicer view of how that smooth filter has actually worked. And I explained in the one of the first two lectures how the smoothing filter worked. If you remembered it was to do with in that in in the example I used a 9 by9 grid of pixels, the area of pixels and you you calculated the the mean intensity of pixels around the perimeter of one individual pixel and then you change that pixels the central pixels value to the average value of the pixels around about it. and you can see what happens.

Let's try a a different one. Uh maybe maybe a sharpen filter. Oh yeah. So just take another box and we'll sharpen that bit. Okay. So it's quite a dramatic difference. Maybe down here we might try a maybe a detect edges. Let's try that. See if it see if it catches these nuclei here. Find edges. It's not too bad. Possibly using what's known as a soel filter. I won't go into the details here, but okay. So you can see here from what I've done is I have I've sharpened this area of the image.

I've smoothed this area of the image and I've done a a fine edge on on this particular part. So these can be quite useful. The filters can be very useful. But the key thing to remember about the filters is that it changes the numbers quite dramatically. So if you were to then if your intention for the image analysis was to measure intensity then you should be very wary about using these types of image filters but they could be useful for identifying objects which you then wanted to measure the shape of.

So that's um filters uh taken care of you. There are a whole host of different filters which you know we probably won't even touch on during the course but you know by all means in in your own time if you have any spare time um have a look at these yourself. Um the FIA transform we will look at next week and I probably don't want to get into that just now. Okay, let me just bring back one of my original images that I'm not going to I'm going to not save that.

Uh, and just bring back here was the original image. And I'll just zoom up a little bit. Okay, so measurements. What what kind of measurements could we make on this image? Well, the simplest measurement that we could make would be perhaps a pointto-oint. Now if the image is calibrated then the measurements that image J will give you back um will be in microns. Um so if I was to but if the image is not calibrated then the measurements that are made will be in pixels.

Now we can tell if the image is calibrated in two ways. One would be to look up here to see whether or not it gives us the the dimensions of the image in microns. And the other way would be to see well is there a does it give us a scale bar? And this is the bit that I can never find. Uh where is the scale bar? Um the tools here tools scale bar. There we go. So, analyze tool scale bar asks me um maybe I wanted a 20 micron scale bar.

Um I can say do I want it? You see it down here in the image. Um what color do I want it to be? Yeah. So on. Okay. Um the fact that the scale bar is reporting in microns tells me that my image is is calibrated or has some calibration. Right. All right. So, if I wanted to measure the distance between two nuclei here, I might just take my line tool, uh, click and drag. And it tells me, if you look up here, oh, it's gone.

If you looked up here, I'll do it again. Look up in this area here, you'll see the measurements. 26.74 microns between those two nuclei and 57.03 between those two and so on. Um, this tool here allows us to measure angles. I maybe for some reason might want to measure the angle between let's say these three points and it tells us the the angle where it says there angle 46.56 is the angle. We could also measure brightness.

Now before we measure brightness, uh what I would need to do is to is probably to go into the measurement options and tell image J exactly what it is that I want to measure. So I would set my measurements. So that was let me just show you that was analyze set measurements and here is a list of um classifiers parameters that I could want to measure. Um, so yes, if I was measuring intensity, I would probably want to know the minimum and maximum values in an area.

I might want to know the total uh intensity in an area. Uh, yeah, the mean value and the modal gray value uh and the median. Uh, these would all be quite useful to to know. Might want to limit it to the threshold. Talk about that in just a just a minute. Let's just go with that for the moment. If I was to draw an area, let's say this area here, and then do measure, you see it pops up a results window that gives me the area of the the region that I've measured, the mean intensity value, the standard deviation of those intensity values, and the other intensity information that we asked for.

Another way that we might uh choose to measure would be to do a plot profile. So again I would take a a line here. So let me draw a line through some bright objects. So you see that my line is going through four bright objects here. Let me look at the plot profile for these which will be uh plot profile. So analyze plot profile and here we see each of these peaks. So this graph represents the intensity along this yellow line.

So as we move along this yellow line, what is plotted here is the intensity of the pixels underneath the line. So when we get to the very bright objects, we get a big peak and here and here and here. So you could rather than measure the distance between the objects using the line tool as I showed you at first, you could measure the distance between the peaks which would give you give you the same result I guess. So that can be quite useful uh as a 2D plot profile.

And finally on this we might want to go for a surface plot which is basically a 3D plot of do the wireframe as well. Basically a 3D plot of the intensities. So get the XY coordinates of each pixel and the height of each column. So each pixel is represented by a column and the height of that column is indicative of the intensity. So that's just quickly how we might do some measurements. Now thresholding and segmentation is one of the most important uh functions of pretty much any image analysis uh tool.

Before going on, is are there any questions at all? Are you still with me? Good. Okay. Nothing in the box. And um let's crack on then. So thresholding and segmentation. What do we mean by that? Well, it's about selecting areas of the image that we're interested in. And normally we would be interested in the bright areas, but we could equally be interested in the dark areas. Perhaps if our image looked something like that, we might be interested in the dark images.

But uh I'm going to invert this again. The invert is quite useful sometimes because thresholding algorithms more often than not I would say would be expecting bright objects. So if if the image that you've collected happens to be reversed and it's the dark blobs or the dark staining which is the thing that you're interested in, just inverting the image sometimes can be quite useful. All right, we would need to adjust the threshold.

Now initially image J will make a guess at the objects. So this this window just popped up on another screen. This is the window here that's that's important. It's made a guess. It's showing me the histogram uh to have a lot of pixels which are very dark and it automatically assumes that you're not interested in the dark stuff that you're probably interested in the upper range of intensities. And so it has automatically selected for me every pixel that falls within the range the brightness range of 120 to 255.

Now I can change that just by moving the slider. I could also choose a different algorithm which will estimate something different. Okay, you try these yourself. Now the reason for having these um is would be if you wanted to make to do a batch process. If you wanted to automate this process by creating a macro that um that analyzed a whole bunch of images uh like hundreds of images maybe um in sequence then you wouldn't want to be automatic.

You wouldn't want to manually select the the threshold for each of these. So what could be useful would be that you take a sample of your images and you find one of these automated um settings which might work for your image and then you can build that into the macro. So worthwhile having a look at these and in exercise one you're asked to this is the practical exercise you're asked to to try a few of these. So, we'll go with the default and I'll just uh adjust this.

I don't need to bring the top down because I'm I'm really looking for the brightest. Could change the color of this if um if I needed, but we'll stick with red. Uh okay, let's just go with maybe something like that. If I apply this, I end up with a binary image. I have removed all of the intensity information and I'm basically left with shapes. The first part of that process we refer to as thresholding. That's where I was choosing the range of data that I wanted to measure.

Segmentation was that point where I chose apply. I have segmented the data into this binary image where each pixel is either white or black. Okay? Either has a value and you'll see here of zero or 255. Now that's the opposite for what we'd expect normally white values to be 255 and black values to be zero. But in the segmented image it's it's reversed because it knows that that was the bright images. Okay. If we want to measure this, you'll see that there's quite a lot of little specks of pixels here that we probably don't want.

So, what I would probably do would be try to remove these single pixels or maybe just do an erosion. Now, if I go to my binary options, these are the options that I have to process a binary image. So, I go process binary. Um, let me do an erosion. Now, an erosion will erode. Let me just zoom up. Actually, might make it a bit clearer. Okay. So, here we've got a big blob and some little tiny spots. If I do an erosion, it will take one pixel off of the perimeter of each object and will remove single pixels.

And I think you'd agree much nicer, cleaner image. I could then analyze particles. Now I probably don't want anything which is smaller than say 20 pixels. Um yeah, let's just do I want to exclude it. Display the results. Include the holes. We'll show the outlines of the objects. Um you have options here of what you want to show. Let's just show the outlines for now. Um click okay. And there you see what we've got is a a map showing me the objects, the number of the object.

And if I go to my results window, it will tell me here for each of those objects which are numbered. So let's let's see here. Um let's look at image six there. Uh so um object six. Here's object six here. tells us the area of it. The intensity values are not of much use. Now I see I made a mistake here because what I should have done in my measurements is I should have set my measurements and I should have asked for the shapes factor which is there shape description.

I should have clicked the shape description and that would have given me a value for each object between zero and one where one is a perfect circle, zero is a straight line and this object here would have probably tended um towards a value of one uh more than say this object here at 13 which may have given me a value of something like about 0.5 or thereabouts. Okay, so that is thresholding, segmentation, a little bit of filtering.

Um, let me Any questions on that so far? Because I'm going to move on now to stacks and then then probably wind up looking in the chat box. Okay. Nothing. Right. Okay. So that was all single images. Let me finish off by looking at a series of images, a stack of images or a Z series. Um I pulled out the image that we were looking at was one of a set of images which are shown here. Now I have in in this folder in this folder here there are 50 images.

Now the same data set is also represented here. Now this picked image is a biorad file format which I can lift and just drop straight on to image J and it will automatically open it automatically opens the image and within this image as you will see it's a series of images. Now if I had difficulty opening that what I would do is I would need to import Oh, sorry. I need to I would It's um open a binary file. It's input output input output bio reader.

If I had any difficulty opening, I would use this biorad reader. And there are options to open different types of files from Leica microscopes, from Zeiss microscopes, um perhaps diccom readers if you were using medical images from MRI scans and so on. So it's important uh to know what type of data you're trying to open and to make sure the image J andor Fiji has the has the means to import that file. And both of both MSG and FG can pretty much open all of the standard files.

Let me open it as an image series though just for just for fun. Uh I would do file um import an image sequence. I would take the first image in the folder and just click open. It will detect that there are 39 images in there and yeah I'll just take them all. So, it opens up a Z series of images. Now, I don't like the blue lookup table. So, I will change this to a gray lookup table. There we go. Uh, looks a lot nicer, I think you would agree.

Um, bit of a mess there in the adventure. So, this is the advantia here. Um, this is the smooth muscle cells. These are the endothelial cell nuclei. Now, normally when I'm scanning blood vessels, I would scan from the outside in. But in this particular case, this is from the inside out, whereas this is the bottom section. Sorry, this is the top section, and this is the bottom section. I'm just going to reverse that. Um, if I can remember where that option is.

Um. Oh, stacks. Under stacks, won't it reverse? Okay. Stacks, tools, reverse. And it just flips the whole thing. So, this is now the top section. This is the bottom section. So, what can we do with a stack? Well, one of the things we could do is we could make a substack. So, that would be maybe I don't want these these very brightest points, but I might just want maybe just the smooth, no actually the endthelial cells because they're quite nice.

So if I look up at the where I am in the Z scan, it tells me so this is if you look up here, this is image 30 of a stack of 39. So probably what I want starts about 33 and goes to 39. So maybe make a small stack of just No, let's make it a bit thicker. Let's do it from 27 to 39. Slice 27 to slice 39. I'm going to make a substack. Um, where am I going to make a substack? Should have rehearsed this before doing this. Make a substack.

So, stacks tools. make a substack. And I want you to do I think we said 29 to was it 22 to 39? I can't remember. Doesn't matter. 22 to 39. Let's do that. Okay. So, I've got a smaller stack. Now, this one is only 18 slices. Here's the original stack uh which had 39 images. So, I've made a substack. Maybe I might want to actually make it even smaller, but maybe just crop out a little bit of the image that I'm interested in.

Maybe just that bit there. Let's just crop that. So, I'll go image. Uh, and again, where did they put these things? Uh, so if you're doing this in Fiji, things might be in different places. Oh, image crop. There we go. Right. So, I've cropped it. I've got a smaller data set. Now, this can be very useful if you don't have very much memory in your computer. This could be very useful particularly for when you get to the 3D volumes.

Okay. So, I've got a nice little substack here which I'll just zoom up a little bit. Um, I've cut my data down to just a few slices. Smaller amount of data. Um, let me just duplicate this because I'm going to do a few things here and I don't want to lose it. So, just going to duplicate duplicate the entire stack. Okay, I'll keep this down here. I'll actually do two just in case I get into difficulty. Okay, always make some duplicates.

You can never go wrong, right? Last couple of things. Then I've got a 3D volume, but you know, how do I how do I how do I display this in a in a 2D image? Well, I could make a Z projection. So, I'm going to go to stacks and Z project. Now, if I do an average, if I do a maximum intensity Z projection, what this will do is it will look at the XY location of every pixel and it will look at the intensity of oops, I can't do it will look at the intensity of every pixel.

So, let me just zoom in. Let's say on let's say this location here. Um what the Z projection will do is it will look at the intensity of try and keep your eye on that pixel. You'll see that as I go down the Z series. I'm looking at this one here that that pixel location changes in intensity. If you were to take the average intensity of all of the pixels at that one single location down through the Z axis and then make a new image which had only the mean values then oh only the maximum value then you would get a maximum intensity projection.

If you chose to take the average value, oops, stack required, of course. If you were to take the average value, then that would be the image you would get. And if you were to take the standard deviation of the average of every pixel at that location, then you get the standard deviation. And you can guess what the other ones would be. Sum, mean, minimum, and so on. So, you know, which one of these looks the best tend to be always you kind of gravitate towards the brightest image, but actually sometimes the standard deviation can work very well.

And there's a few occasions where I've used the standard deviation. It's worked really nicely. So, that Z projections, those can be very useful for displaying all of the images in a stack in the one image, in a one 2D image. Finally, what about 3D? Now, we're not going to talk about 3D until the probably the last the last week or so. The last two weeks, we'll talk about 3D. Uh oh, no, we'll do a bit next week. We'll talk about point spread functions and correcting 3D volumes.

So, I think actually it might come next week. Let me just briefly trailer that then by using the volume viewer plugin. So here we have that this data set here. I've loaded up. I went to the plugins. I've gone 3D volume viewer. And so here we have a a volume. Let me zoom out here. Okay. So there's our volume of data as a 3D volutric rendering. And we can choose we can use this this curve here to change the opacity of the pixels.

Cut out all the dark stuff. Maybe switch on the light. Very nice. See, there's a little bit of noise in here. I would maybe maybe go back and maybe smooth the image here. Let's do that actually just for just for a bit of fun. Um process smooth. Oh, let's actually do a little bit finer grain. We'll go to filters and we'll do a gaussium blur. And we can preview this, which is useful. So, that's too much, isn't it? Let's maybe do 0.5.

Yeah, that's a wee bit better. Maybe one is too much. Yeah, one is too much. 0.5. Okay. Yeah, I'll do it to all the slices. So, that's me. smoothed the the data. If I go back now to my 3D volume viewer, hopefully if I change this, it's a we bit cleaner. I think you would agree. Okay. Now, this also offers orthogonal slicing and so on, but I think I will leave that for now. Okay, that was yeah, probably 40 minutes or so of a quick whiz through MHJ.

There was a lot of topics there. Um, but hopefully hopefully that was useful as a just as a quick introduction to how you would use MageJ and the the kind of functionality that has. And as you saw from all of those plugins, we have really only scratched the surface of what of what MageJ is capable of. Okay, that's all I've got for you today. And if you enjoyed that, then please think about subscribing to the channel.

Ring the bell for notifications. Give it a thumbs up if you like the video. And it really helps because it it lets me know that there's somebody out there that really is interested in seeing these types of videos. See you next time.