Hello, I am working on the role of Peri biliary glands(PBG's). attached is one image of bile duct with PBG on the side, bile duct cells and PBG cells are marked by Cytokeratin (Purple). I want to know any methodologies to segment these PBG for quantification of number and geometrical parameters. I would be happy to share raw data files or try out any methodologies that you might think helps.
Thanks
Dear Dr. Gupta,
One solution is to manually annotate a number of images in your dataset and then train a supervised deep learning approach (such as U-Net). You will find an easy implementation of U-Net in the following link:
https://github.com/jocicmarko/ultrasound-nerve-segmentation
It is for another application but you could use it for your case as well.
Hope it helps,
All the best,
Amirreza
you can also try attention models/instance segmentation such as Mask RCNN
Amirreza Mahbod thank you for the code. This did not work that well with these images. If you can refer me to any other similar method that would be very helpful. thanks.
Dear Kapish Gupta , U-net is a pretty much standard method for deep-learning-based image segmentation. If you have too much color variation in your images you may apply strong augmentation techniques or normalize the images before sending them to the network.
Of course, there are other advanced CNN-based solutions for image segmentation but if U-net does not work at all for your dataset, I do not think others will do. You will find other easy-to-use implemented approaches (such as LinkNet or FPN ) from here: https://github.com/qubvel/segmentation_models
- Badges
- Science method