1
00:00:01,480 --> 00:00:05,840
In this two-part video series, we’ll show you how to use
cellSens software’s

2
00:00:05,840 --> 00:00:10,000
deep-learning technology for label-free nucleus detection.

3
00:00:10,000 --> 00:00:13,840
Rather than using fluorescence images
to detect and count nuclei,

4
00:00:13,840 --> 00:00:18,200
deep-learning technology can count the nuclei
using simple transmitted images

5
00:00:18,200 --> 00:00:23,920
with no staining, eliminating the phototoxicity
caused by fluorescence excitation.

6
00:00:23,920 --> 00:00:27,360
Deep learning uses a neural network
that’s trained by an operator

7
00:00:27,360 --> 00:00:30,880
who inputs images with the nuclei already identified.

8
00:00:30,880 --> 00:00:36,400
When this training is complete, the system will be able
to recognize nuclei on its own.

9
00:00:36,400 --> 00:00:40,400
In this video, we’ll show you how to
prepare your training data.

10
00:00:40,400 --> 00:00:43,520
The remaining steps in label-free nucleus detection

11
00:00:43,520 --> 00:00:46,000
are covered in the second video.

12
00:00:46,000 --> 00:00:50,360
To begin, go cellSens software’s Count and Measure tab.

13
00:00:50,360 --> 00:00:54,280
Open a training image located in the cellSens installer.

14
00:00:54,280 --> 00:00:59,560
To access it, go to cellSens > example images >

15
00:00:59,560 --> 00:01:05,960
deep learning > training images. Open the first image.

16
00:01:05,960 --> 00:01:10,400
These fluorescence training images’ nuclei have
been stained with DAPI.

17
00:01:10,400 --> 00:01:14,320
The brightfield and phase contrast channels are shown.

18
00:01:14,320 --> 00:01:17,400
Use the DAPI channel to create binary training data

19
00:01:17,400 --> 00:01:20,440
in the software’s Count and Measure module.

20
00:01:20,440 --> 00:01:25,120
Since we have to do this for multiple images,
record it as a macro.

21
00:01:25,120 --> 00:01:33,720
To do this, open the Macro Manager tool window
on the menu bar.

22
00:01:33,720 --> 00:01:47,040
Click Create Macro and name it to create the training label.

23
00:01:47,040 --> 00:01:51,480
Then select the DAPI channel from Dimension Selector.

24
00:01:51,480 --> 00:01:55,720
Select Automatic Threshold to automatically
apply an intensity threshold

25
00:01:55,720 --> 00:02:01,400
on the DAPI channel to detect nuclei.

26
00:02:01,400 --> 00:02:06,000
Then click DAPI channel.

27
00:02:06,000 --> 00:02:09,200
Double click on the phase name to modify it.

28
00:02:09,200 --> 00:02:14,200
Later, we’ll use this as the name of the training label.

29
00:02:14,200 --> 00:02:17,200
Select the color from the drop down.

30
00:02:17,200 --> 00:02:21,160
Use a different color from the DAPI channel and other phases.

31
00:02:21,160 --> 00:02:23,240
We chose red.

32
00:02:23,240 --> 00:02:26,360
Then click Count and Measure.

33
00:02:26,360 --> 00:02:29,200
The nuclei have been detected in red.

34
00:02:29,200 --> 00:02:31,800
If two nuclei are located close together,

35
00:02:31,800 --> 00:02:35,600
they are sometimes detected as a single nucleus.

36
00:02:35,600 --> 00:02:40,920
When this happens, you can tell the system
to automatically split these into two objects.

37
00:02:40,920 --> 00:02:43,200
Click select all detected objects

38
00:02:43,200 --> 00:02:47,640
and then click Auto split selected objects.

39
00:03:02,600 --> 00:03:05,320
Confirm that the objects have been split properly,

40
00:03:05,320 --> 00:03:08,440
and then stop recording the macro.

41
00:03:19,720 --> 00:03:24,440
Now, let’s apply the macro we recorded to multiple images.

42
00:03:24,440 --> 00:03:28,240
Close the image you have open.

43
00:03:28,240 --> 00:03:36,520
Activate toggle batch mode, and then click run macro.

44
00:03:36,520 --> 00:03:41,280
We’re going to use images in the installer guide
for this example.

45
00:03:41,280 --> 00:03:45,360
To define these input images, select documents in file system,

46
00:03:45,360 --> 00:03:56,240
and select the 10 training images.

47
00:03:56,240 --> 00:03:58,560
Then click next.

48
00:03:58,560 --> 00:04:02,400
Choose the location where you want to save the files.

49
00:04:12,320 --> 00:04:16,560
Here, we rename the batch training label.

50
00:04:31,720 --> 00:04:37,240
The training data must be in .vsi format.

51
00:04:37,400 --> 00:04:40,600
Normally, we recommend you select close documents

52
00:04:40,600 --> 00:04:44,880
in case the batch macro is applied to a large number of images.

53
00:04:44,880 --> 00:04:50,360
Since we’re only using 10 images in this example,
we’ll leave it unselected.

54
00:04:50,360 --> 00:04:58,520
Run the batch macro to create training labels for the 10 images.

55
00:04:58,520 --> 00:05:02,360
Check to confirm that the process
was successful for all 10 images

56
00:05:02,360 --> 00:05:09,480
and that the data were saved in .vsi format.

57
00:05:09,480 --> 00:05:15,040
It’s good practice to quickly verify the quality
of the nucleus identification and splitting.

58
00:05:15,040 --> 00:05:21,640
If an object is not properly split, please do so manually.

59
00:05:29,120 --> 00:05:33,160
Don’t forget to save the image if you modify it.

60
00:05:33,160 --> 00:05:35,960
Your training data is now ready.

61
00:05:35,960 --> 00:05:41,720
Check out part 2 of this video series where
we’ll show you how to use this dataset.

62
00:05:41,720 --> 00:05:45,600
If you’d like more information about
Olympus deep-learning technology,

63
00:05:45,600 --> 00:05:49,880
visit
olympus-lifescience.com.