Real-Time Monitoring of CAR T Cell Dynamics in Tumor Patient-Derived Organoids using the OrganoIDNet algorithm

Background Patient-derived organoids (PDOs) provide physiologically relevant 3D tumor models for preclinical drug testing, yet robust, scalable methods to quantify dynamic responses to immunotherapies remain limited. OrganoIDNet is a deep learning–based image analysis framework that enables automated, label-free segmentation and longitudinal quantification of organoid morphology. Here, we extend the application of OrganoIDNet to evaluate chimeric antigen receptor (CAR) T cell activity against pancreatic ductal adenocarcinoma (PDAC) PDOs targeting the tumor-associated antigen CD318. Results CD318-directed CAR T cells were co-cultured with PDAC PDOs using a Matrigel-based sandwich system and monitored by time-lapse bright-field imaging. OrganoIDNet enabled accurate single-organoid segmentation and continuous quantification of organoid number and area across multiple effector-to-target ratios. CAR-318 T cells induced robust, antigen-dependent cytotoxicity, characterized by progressive reductions in organoid number and size that correlated with increased T cell activation and reduced exhaustion markers. Dynamic imaging further revealed rapid T cell–organoid interactions and early tumor cell elimination, capturing killing kinetics and spatial patterns not accessible by conventional endpoint assays. Conclusions By integrating organoid–immune co-cultures with OrganoIDNet-driven live-cell imaging, we established a scalable, automated, and high-content platform for real-time assessment of CAR T cell efficacy in solid tumors. This approach surpasses traditional 2D and bulk cytotoxicity assays by providing continuous, spatially resolved, and patient-specific readouts of therapeutic response, supporting its utility as a preclinical framework for CAR T cell development and personalized immunotherapy evaluation.