Abstract The increasing use of precious, patient-derived cells has driven the need for non-perturbing and label-free cell measurements, particularly in the oncology field. To address this we developed the Incucyte® AI Cell Health Analysis Software Module, which uses two pre-trained deep neural networks to perform automated, unbiased analysis of Phase contrast images to segment individual cells and perform label-free Live/Dead cell classification. The neural networks which perform cell instance segmentation and infer cell viability were trained on a wide diversity of cell types with varied morphologies, ensuring that the analysis is applicable across a variety of adherent and suspension tumor cell types. Here, we demonstrate the application of this analysis across diverse and commonly used biological models of breast cancer, glioblastoma, and B-cell lymphoma. In each case, cells were treated with chemotherapeutic compounds or biosimilar antibodies and Phase contrast images were acquired at regular intervals over 3 - 4 days using the Incucyte® Live-Cell Analysis System. Using the Incucyte® AI Cell Health Analysis cells were accurately segmented and the percentage of dead cells were quantified over time without the requirement for a fluorescent reporter or other exogenous label, and with limited user input. Four breast cancer cell lines were treated with a panel of chemotherapeutics designed to target specific expression patterns. AI Cell Health analysis showed that Estrogen receptor (ER) inhibitor Tamoxifen selectively induced >60% cell death only in ER positive cell lines BT474 and MCF7; dual epidermal growth factor receptor (EGFR/HER2) inhibitor Lapatinib induced cell death in AU565, BT474 and MCF7 which express these surface markers. In contrast, Lapatinib and Tamoxifen induced morphological change - but minimal cell death - in triple negative MDA-MB-231 cells. Three glioblastoma cell lines A172, U87 and T98G were treated with a larger panel of chemotherapeutic compounds and for four of the active compounds, efficacy was also determined. Cisplatin, doxorubicin, vinblastine and taxol induced concentration-dependent cell death in A172 and T98G cells; U87 cells displayed resistance to each of these compounds with a maximal 46.5% cell death induced by doxorubicin. Ramos B-cell lymphoma cells were exposed to increasing concentrations of monoclonal antibody Rituximab and the biosimilar Truxima®. The antibodies induced specific cell death via the surface marker CD20 in a time and concentration-dependent manner with similar efficacy (IC50 Rituximab 94.7 ng/mL; Truxima® 110.3 ng/mL), while antibody control IgG1 remained non-perturbing to cells. These results demonstrate that the Incucyte® AI Cell Health Analysis is applicable to a broad range of cancer types cultured in 2D monolayer. This unbiased method enables accurate, label-free quantification of cytotoxic effects induced by clinically relevant therapeutics. Citation Format: Gillian Lovell, Daniel A. Porto, Jasmine Trigg, Nevine Holtz, Nicola Bevan, Timothy Dale, Daniel Appledorn. AI-driven image analysis enables simplified, label-free cytotoxicity screening. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5418.
Read full abstract