Evaluating cell growth and hypoxic regions of 3D spheroids via a machine learning approach

Abstract

This study investigated the applicability of the area of spheroids and hypoxic regions for efficient evaluation of drug efficacy using machine learning (ML). We initially developed a high-throughput detection method to obtain the area of spheroids and hypoxic regions that can handle over 10 000 images per hour with an error rate of 2%-3%. The ML models were trained using cell growth of six cell lines (i.e. HepG2, A549, Hep3B, BEAS-2B, HT-29, and HCT116) and hypoxic region variations of two cell lines (i.e. HepG2 and BEAS-2B); our model can predict the area of spheroids and hypoxic region of certain growth date with high precision. To demonstrate the applicability, HepG2 spheroids were treated with sorafenib, and the efficacy of the drug was evaluated through a comparison of differences in areas of cell size and hypoxic regions with the predicted results. Furthermore, our ML approach has been shown to be applicable to provide the model-driven evaluative criterion for toxicity and drug efficacy using spheroids.