I am studying the first deep learning project. I need to counting tubules and cells. How to do this? If you suggest a tutorial (or another thing) related to this subject, I would love to.
Firstly, I must detect and count tubules (black ones) and after must detect and count cells (yellow ones).
I am planning to use UNET but I don't understand how to make detect and count operations.
Thanks
This paper is on a related topic and it may help for collecting training data. But it is not a deep learning approach.
Article Detection of tubule boundaries based on circular shortest pa...
Different variant of wavelet transform in image processing technique help to count the tubules and cells . After the find out the morphological feature like area perimeter ,color, etc parameter of tubles and cell.
Basically, you need to divide the task into steps:
1. Segmenting the regions:
This could be done with Unet, which you need to train first. You would need an annotated dataset(the histopathology images with corresponding ground truth - there are public datasets).
Check this link: https://towardsdatascience.com/nucleus-segmentation-using-u-net-eceb14a9ced4
2. After that you would need to make your predictions binary and label the connected components. Then you can count them. You can use open-cv python library for this step.
I'm so thankful you all answers. They help with my project. I preferred the YOLO algorithm for tubul and cells detection. I am working on this.
Due to cell contiguity, a nonuniform background, shape and size variations of cells, and different techniques of image acquisition, the task becomes even more difficult . Deep leaning base Faster-RCNN architecture combined with a U-Net neural network, for cell segmentation and also used in cell counting.
http://biometrics.uniss.it/vps_files/posters/poster11.pdf
Counting them will be easy just raster over the image and count connected components of the output of the Unet
- Badges
- Science topic