Related Posts
Stay in touch with our newsletter!
APEER is now ZEISS arivis Cloud. The Machine Learning toolkit is now arivis AI. Learn More >
Cell counting is a common task in many life science applications and for over 100 years the process has been primarily manual. These manual processes can be time-consuming and unrepeatable due to lack of statistical robustness and subjectivity of cell definition among researchers. Automating the cell counting process is the key factor to obtain reproducible results across various samples and users.
While today’s microscopes are designed for automated imaging, the process of extracting information from these micrographs can be challenging. There are excellent commercial and open source software packages that offer useful tools for image processing, but most researchers barely scratch their surface. This is partly due to the bulky nature of these software packages that come with steep learning curve. In addition, most life scientists are skilled at understanding the biological challenges but not trained at coding image processing algorithms, especially with machine learning and artificial intelligence. A good portion of code is available in the open source domain, but a typical researcher is not comfortable working with code. Therefore, valuable research time gets wasted in performing manual tasks or dealing with multiple fragmented software packages.
arivis Cloud addresses these issues by providing an easy to use online platform arivis Cloud for the researcher to build custom applications (Workflows) using individual pieces of software (Modules). For example, Supervised Segmentation Trainer and Segmentation Mask Generator (Predictor) modules on arivis Cloud can be used to build cell counter workflows. These modules are based on U-Net, a convolutional neural network that was developed for biomedical image segmentation. They are designed to scale up computational resources by leveraging GPU acceleration in the cloud.
Although U-Net has proven to be ideal for many applications, its accuracy depends on the amount of training data, as it is the case with any deep learning approach. Therefore, a traditional image processing approach is preferred for single images or smaller datasets e.g. single 3D stack or a 96 well plate. Traditional image processing involves a handful of image processing operations that progressively extract information from images. For example, cell counting using multichannel images often involves extraction of the blue (DAPI) channel as a first step. This image then gets thresholded to detect cells, refined to fill missing pixels & separate overlapping cells (e.g. Watershed) and analyzed to report numbers of cells and their individual measurements, respectively.
Each of the above tasks can be performed individually in most image processing software packages but the researcher needs to carefully document settings for each step during execution in order to repeat the experiment on multiple images. A minor error is setting up parameters for a single step may yield statistically varying results due to propagation of errors through each step in the process. These human errors can be easily overcome by automating the process. An end to end automated workflow to perform these tasks can be easily put together on arivis Cloud using just two versatile modules, Advanced Threshold and Particle Analyzer 2D. The results from this workflow can be seen here.
Alternatively, the researcher can use one of the preconfigured public workflows like the Cell Counter and Analyzer developed by the arivis Cloud team. This approach provides the researcher more control over individual modules that make up the workflow. The researcher can further customize the workflow by swapping, deleting or adding modules. The results from this workflow can be seen here.
In the next step the results can be published or shared with selected individuals who can repeat the same experiment using the same settings without the worry of defining every parameter in the workflow.
In summary, with arivis Cloud, applications can be customized by researchers for their specific jobs to be done. These customized applications can then be automated where the results can be shared for download or online visualization by their peers. arivis Cloud not only saves valuable research time but also provides a platform for highly repeatable results by researchers and their peers.
