Analysis of Cell Viability in Microfluidic Spheroid Arrays by Image Analysis and Neural Networks

Background

3D microfluidic spheroid arrays have the potential to overcome current limitations in drug development and clinical trials. These limitations include high costs of time and money in clinical trials mainly due to the lack of predictability of preclinical studies (Hirschhaeuser et al., 2010). 3D spheroid arrays offer higher predictability, due to their three-dimensional structure. The cells’ composition is closer to a natural cell composition within the body, compared to traditional two-dimensional in vitro models like petri dishes (Cui et al., 2017). With the use of spheroid arrays more realistic cell contacts can be simulated early on during preclinical studies. It allows a simplified and improved drug screening process, which results in reduced time and cost of the trial (Begley & Ioannidis, 2015). The procedures’ lack of practicability due to missing automation and standardization is caused by high complexity (Eilenberger et al., 2021).

The proposed approach contributes to the automatization of the evaluation and analysis of such experiments. Therefore, an easy-to-use algorithm for the automated analysis of 3D spheroid images was developed. To create the images used in this study, cell nuclei (colored with Hoechst 3342, blue) and dead cells (colored with Ethidium-homodimer-1, red) of spheroids were stained using DAPI (excitation: 390nm, emission: 460nm) and TRITC filters (excitation: 530 nm, emission: 645 nm). Images of the spheroids were taken from an above point of view using a fluorescence microscope (Figure 1). These images can be used for the extraction of important measurements to evaluate and assess the quality of a cell. Important metrics like cell viability can be extracted since the color difference allows the differentiation of living and dead cells in a cell aggregate. The developed algorithm allows the extraction of metrics of high interest for further analyses, which most importantly includes viability and the area covered by single-cell aggregates (Friedrich et al., 2009; Ivanov et al., 2014).

Figure 1.

Rendering of the microfluidic spheroid array and an example image of the microfluidic spheroid array with 15 spheroids. Adapted from Schurr et al. (2021)