1
00:00:02,02 --> 00:00:06,18
In this second video of our two-part Deep-Learning Technology video series,

2
00:00:06,18 --> 00:00:10,04
we’ll show you how to use the training data we created in part 1

3
00:00:10,04 --> 00:00:14,18
to setup your software for label-free nucleus detection.

4	
00:00:14,18 --> 00:00:17,16
To start, go to the Deep Learning layout tab

5
00:00:17,16 --> 00:00:21,03
and click ‘New Training’ in the deep learning tool window.

6
00:00:21,03 --> 00:00:24,03
Change the name of your new training and select the data set

7
00:00:24,03 --> 00:00:28,01
you created while watching the first video.

8
00:00:42,18 --> 00:00:47,20
Click OK on the dialog box and wait a few seconds.

9
00:00:55,09 --> 00:00:57,22
Then, choose the input channel.

10	
00:00:57,22 --> 00:01:02,23
For label-free nucleus detection, you’ll select the
transmitted light channel.

11
00:01:02,23 --> 00:01:05,14
However, since we’re going to use the training data

12
00:01:05,14 --> 00:01:08,09
to detect nuclei from phase contrast images,

13	
00:01:08,09 --> 00:01:12,06
select PH instead of transmitted light.

14	
00:01:12,06 --> 00:01:18,05
The training label class is nuclei only,
and we defined it in the previous video.

15
00:01:18,05 --> 00:01:21,15
Now we can select the training configuration.

16	
00:01:21,15 --> 00:01:25,04
In many cases, the standard network is OK.

17
00:01:25,04 --> 00:01:27,22	
We can also select the training duration;

18	
00:01:27,22 --> 00:01:30,02	
If we select “Iteration limit,”

19	
00:01:30,02 --> 00:01:35,06
the training process stops automatically
when the required number of iterations is reached.

20
00:01:35,06 --> 00:01:39,11
If necessary, we can increase it later.

21
00:01:39,11 --> 00:01:44,20
Typically, the deep learning training phase
requires a powerful computer processor.

22
00:01:44,20 --> 00:01:50,19
We used an NVIDIA Quadro P2200,
and the process took about 50 minutes.

23
00:01:50,19 --> 00:01:53,11
There are several ways to check accuracy.

24
00:01:53,11 --> 00:01:56,09
The similarity between the training layer and the layers

25
00:01:56,09 --> 00:02:02,00
where the neural network identified the nuclei,
called inferred layers, is shown by default.

26
00:02:02,00 --> 00:02:05,12
The minimum value is 0, and the max is 1.

27
00:02:05,12 --> 00:02:09,17
Note that the similarity increases during the training process.

28
00:02:16,15 --> 00:02:20,22
When each iteration finishes, a check point is created.

29
00:02:20,22 --> 00:02:25,20
You can check the quality of the neural network
by validating the data at each check point.

30
00:02:25,20 --> 00:02:29,06
Validation data are selected from the training data set.

31
00:02:29,06 --> 00:02:35,01
The red areas are the inferred nuclei from
the validation data’s phase contrast channel.

32	
00:02:35,01 --> 00:02:39,16
The red circle is the training label showing the correct nuclei area.

33
00:02:39,16 --> 00:02:49,08
You can save the neural network at a check point
and end the training, but let’s continue.

34
00:02:49,08 --> 00:02:56,11
In this example, the training process is finished
after five check points have been created.

35
00:02:56,11 --> 00:03:01,15
Now that the training is complete,
we can see that the detection is accurate.

36
00:03:01,15 --> 00:03:13,19
Save the neural network at checkpoint five
so that we can use it for other images.

37
00:03:13,19 --> 00:03:16,06
The neural network training is complete.

38
00:03:16,06 --> 00:03:21,03
Go to the Count and Measure layout tab
to apply the neural network to other images.

39
00:03:21,03 --> 00:03:26,05
The images we’re going to use in this example
are available in the installer guide.

40
00:03:26,05 --> 00:03:30,07
Open one of the images.

41
00:03:30,07 --> 00:03:41,07
Apply the neural network to the phase contrast channel
to infer nuclei and count them.

42
00:03:41,07 --> 00:03:50,08
In the Automatic Threshold menu,
select Neural Network Segmentation.

43
00:03:50,08 --> 00:03:56,15
Select the PH channel since the neural network
was trained using the phase contrast channel.

44
00:03:56,15 --> 00:04:00,05
A detection threshold will be applied to the nuclei.

45
00:04:00,05 --> 00:04:04,02
Set the threshold so that two close nuclei are properly split

46
00:04:04,02 --> 00:04:07,19
and so that all nuclei are properly detected.

47
00:04:07,19 --> 00:04:10,14
Then, click the Count and Measure button.

48
00:04:10,14 --> 00:04:14,10
The number of nuclei can be counted in the PH image.

49
00:04:14,10 --> 00:04:18,07
In addition to being able to identify nuclei without staining,

50
00:04:18,07 --> 00:04:21,10
the deep-learning technology helps keep your cells healthier

51	
00:04:21,10 --> 00:04:26,02
by eliminating the phototoxicity caused
by fluorescence excitation.

52
00:04:26,02 --> 00:04:27,14
Thanks for watching.

53
00:04:27,14 --> 00:04:32,06
If you’d like more information about cellSens software’s
deep-learning technology,

54
00:04:32,06 --> 00:04:35,21
visit olympus-lifescience.com.