Automated Stereology using Artificial Intelligence
Only CELLAIRUS provides automatic, accurate, unbiased, high-throughput 3D cell quantification in histological specimens.
Unbiased Stereology is recognized as the gold standard for accurate quantification because it is a rigorous and unbiased methodology for quantifying features of biological tissues such as the size, shape, distribution, and quantity of objects. Unfortunately, stereological methods are tedious and time consuming which increase the cost and the amount of time needed to complete thorough scientific research investigator.
Current manual stereological cell counting studies can take months and can be expensive to conduct. Cellairus has been developed to preserve the integrity and value of stereology while substantially reducing the amount of time needed to produce cell population estimates.
CELLAIRUS uses state of the art machine learning software integrated with cell segmentation, stereological counting rules, and a sophisticated framework using unbiased stereological sampling to identify, classify, and quantify cells in 3D microscope images along with automating the most widely used unbiased stereological probe, the optical fractionator.
- Faster than manual stereology—our testing indicates that a typical study is up to 35 times faster than manual stereology
- Unbiased and more accurate than other 3D detection methods.
Cellairus Workflow Operation
A workflow guides the user through the process of setting up a study: drawing contours, and defining counting frame, grid sizes and disector parameters.
CELLAIRUS estimates the section thickness at every site by evaluating a focus metric on all planes in the subvolume and selecting the top and bottom planes for which the focus metric rises above a stack-adapted level. Using this approach, the computed thickness variation is consistent with that obtained by manual measurement by multiple users.
3D Image Segmentation
To avoid bias from manual threshold selection, we developed an automated parameter estimation process. This algorithm analyses indivudal subvoulmes with high numbers of objects and optimizes key segmentation parameters to best match expected object structure.
Train Machine Learning Classifier
The classifier is trained by manually marking the center and diameter of cells within small subregions. For this study, the training process presented ~90 subregions across the full rostro-caudal extent requiring ~1 hour of time.
Apply Counting Rules & Calculate Population Estimates Each
Each counting frame site, with its automatically detected objects, is filtered to only retain the detected cells that fall within the disector.
Entire studies are rapidly processed. The nine sections of the current study were processed in ~6 minutes, while manual methods required ~6 hours.
CELLAIRUS takes advantage of the high-resolution image data that is acquired by our Vesalius high-speed, confocal slide scanner to perform fast, accurate analysis of cell number in fluorescent tissue sections.
- Delineate anatomical regions of interest in which to count cells
- Train to recognize the cells in your specimens
- Deep learning automatically counts cells
- Computes the results and coefficients of error using established stereological formulae
CELLAIRUS also contains features that allow for comprehensive data auditing
- Visually inspect the results using a dynamic 3D visualization environment
- Manually edit any of the automatic results
- Retain your image data and detection results in an easy to access digital format to review at a later time, even years later
CELLAIRUS was supported by NIMH grant R44-MH105091 to MBF Bioscience, Inc.