SpAi: A machine-learning supported experimental workflow for high-throughput spheroid production and analysis

Abstract

Three-dimensional (3D) cell cultures, especially spheroids, provide a physiologically accurate model for cancer research in comparison to conventional two-dimensional (2D) cultures. Nevertheless, the difficulties in producing and analyzing spheroids have impeded their extensive use in high-throughput screening—a critical process for drug discovery. This study presents a simplified process for the effective creation and examination of spheroids using a 3D-printed mold casted polydimethylsiloxane (PDMS) microwells. The utilization of our specially constructed mold facilitated the creation of consistent spheroids, which were subsequently exposed to doxorubicin for the purpose of anticancer medication treatment. We herein improved spheroid analysis by including a convolutional neural network (CNN) model, specifically U-Net, into a graphical user interface (GUI). This integration allows for automated detection and measurement of spheroid size from microscope pictures. The performance of this system surpassed that of conventional image analysis methods in terms of both accuracy and efficiency. The implemented workflow offers a scalable and cost-efficient platform for conducting high-throughput drug screening, which has the potential to enhance the success rates of cancer therapies in clinical trials.