fine tuning segmentation step

[skim245]

Dear author,

I've tried VistoSeg following your tutorial with my data and found that I need to fine-tune the segmentation step.
I would like to know if we can modify threshold and also split overlapping nuclei.
Here is an example image from spotcheck showing poor counts (counting a single nuclei as multiple fragments; may be we need max&min nuclei size threshold?) and also counting overlapping nuclei as one instead of two.
thank you in advance!

[madhavitippani]

Hello skim245,

Can you share your raw image, mask.mat file from VNS function and nuclei.mat from refineVNS function.

[skim245]

Hi @madhavitippani
I've uploaded three files in google drive and below is the link!

https://drive.google.com/drive/folders/1zNLtGC0MujHsqYiEZNXL_2LnHbd51FZT?usp=sharing

[madhavitippani]

Requested access

[madhavitippani]

There are a couple of things you need to change in the code for your images.

1. in the VNS script that is located here ~/VistoSeg/code/VNS.m comment out line 11 and change line 12 as following
   %Img1_smooth = imgaussfilt(Img1,4); %smooth image
he = imadjust(Img1, [.2 .3 0; .6 .7 1],[]); %adjust contrast in image

because smoothening is not working for your images, and you need contrast adjustment on the raw image itself

2. in refineVNS script here ~/VistoSeg/code/refineVNS.m add the following line after line 19
   mask_dark_blue = bwareafilt(mask_dark_blue,[5 2000]);
   in this line [5 2000] is the size threshold on segmented nuclei. Meaning nuclei with less than 5-pixel area and greater than 2000-pixel area are dropped from the segmentation.
   With these changes, I was able to get the following results

without any size threshold

With size threshold [5 2000]