OSCAR is an online ML-powered tool for organoid cell counting using bright-field images

TL;DR

OSCAR is an online tool that uses machine learning to estimate the number of cells in organoids from bright-field images, improving the reliability of organoid-based experiments.

Contribution

OSCAR introduces a new workflow combining Mask R-CNN and regression modeling to estimate organoid cell counts from bright-field images.

Findings

OSCAR estimates organoid cell numbers within ±16% of the actual count.

The tool uses a Mask R-CNN model for organoid segmentation and a regression model to estimate cell numbers.

OSCAR enhances the reliability of organoid-based assays, particularly in co-culture experiments.

Abstract

Numerous software tools have been published to aid organoid quantification. These tools generate estimates of total organoid number and morphological characteristics in images. However, there remains a need to estimate the number of organoid cells in a well for use in organoid-based experiments (e.g., co-cultures). We present OSCAR (organoid segmentation and cell number approximation using regression), a workflow for estimating organoid cell numbers from bright-field images. Step one is a Mask-R-CNN-based convolutional neural network for identifying organoids in bright-field images and estimating the area of each organoid. Step two is an empirical multiple linear regression model relating the number of cells in an organoid to its area. OSCAR’s estimate of the total number of cells in a well was within ±16% of the real number of organoid cells. OSCAR is an online tool capable of generating this key metric and will contribute to the increased reliability of organoid-based assays.

•A Mask R-CNN model for precise organoid segmentation and cell counting•Accurately estimates organoid cell numbers from bright-field images•Provides an additional tool to enhance reliability of organoid-based assays

A Mask R-CNN model for precise organoid segmentation and cell counting

Accurately estimates organoid cell numbers from bright-field images

Provides an additional tool to enhance reliability of organoid-based assays

Organoids are increasingly being used as in vitro models; however, methods for obtaining estimates of organoid cell numbers for use in experiments are limited. Obtaining such a metric could aid in streamlining organoid experiments while improving their reproducibility. Addressing this bottleneck, this study combines computer vision and mathematical modeling approaches, providing a method for estimating the number of organoid cells in a well from a bright-field microscopy image alone and showing its use in a co-culture assay setting.

Burnell et al. develop a computer-based method to count the number of cells in a well of organoids from a regular microscope image. They compare its use to standard counting techniques and test it in experiments where organoids are cultured with immune cells.

Figures