Deep Learning for Cell Segmentation of Large Multi-Channels Time-Series Data in the Amira™ Software
- Abstract number
- 53
- Presentation Form
- Poster
- DOI
- 10.22443/rms.elmi2021.53
- Corresponding Email
- [email protected]
- Session
- Poster Session 2
- Authors
- Jan Giesebrecht (1), Sarawuth Sarawuth Wantha (1, 1), Rengarajan Pelapur (1)
- Affiliations
-
1. Thermo Fisher Scientific Materials & Structural Analysis Quai 8.2 , 39 rue Armagnac 33800 Bordeaux FRANCE
- Keywords
Amira software, living cells, time series data, deep learning, cell segmentation, image processing.
- Abstract text
Quantitative live cell imaging has been widely used to study various dynamical processes in cell biology. Deep Learning has been successfully applied in image segmentation by automatically learning hierarchical features directly from raw images data. Amira™ software offers an integrated Deep Learning environment providing an interface to automatically segment and extract cellular features from microscopy images. Advances in Deep Learning have positioned neural networks as a powerful alternative to traditional approaches such as manual and algorithmic-based segmentation. In particular, the development of the U-Net architecture provided a significant boost to segmentation performance and has now become the template for many modern segmentation models.
We have trained with the Deep Learning Training interface in Amira™ a generic U-Net model using synthetic light microscopy data generated using the SIMCEP simulation tool [Ref. 1]. Using a time series dataset of human breast carcinoma cells [Ref. 2], we applied the U-Net model to predicted the location of the stained cells directly on the raw image data. The prediction results from the pre-trained model on the data set of carcinoma cells were promising. Though this Deep Learning based workflow can be used directly on the raw data it could be part of a fully automated processing pipeline. The combination with the new Smart Multi-channels time-Series system in Amira™ well facilitates the visualization and quantitative analyses of challenging high-resolution data time-series.
- References
[Ref. 1: https://www.cs.tut.fi/sgn/csb/simcep/tool.html].
[Ref. 2: Dr. Kamm. Dept. of Biological Engineering, MIT, MA (USA)]