Using artificial intelligence to improve cell therapy assays: automated quantitative image analysis of cells on...

As the field of cell therapy continues to advance, the combination of cells and directed delivery methods (such as three-dimensional scaffolds, cell printing etc.) continues to grow. These technologies require methods to accurately determine cell numbers and viability to enhance process optimization and develop appropriate release tests. Current methods have limited dynamic range and require substantial manual effort to produce results. Here we describe a simple fluorescent imaging-based method for counting live and dead cells in scaffold cultures that is consistent, automated, and quantitative. First, we optimized labware to control the samples providing uniform imaging fields with standard methods for cell number and viability. We then used a traditional, nondynamic image quantitation software (Gen5, Winooski, VT) across a range of cell concentrations and compared it to the ability of a trained artificial intelligence software (Aiforia, Helsinki, Finland) to determine both cell counts and cell viability in image analysis. Using Aiforia, the live cell counts were highly correlative to seeding concentration (p = 0.0007; r²=0.96) across all tested ranges whereas Gen5 showed no correlation (p = 0.6; r²=0.09). Dead cell counts measured by the two methods were correlated to each other (p = 0.004; r²=0.90) indicating that both systems were equally capable using Propidium Iodide based detection. After completing proper training of the AI system, it provided a clear improvement in data accuracy from its ability to recognize cells amidst highly dynamic backgrounds typical of scaffold culture images. We believe that cell therapy will significantly benefit from AI based approaches.