AI-driven quality control of cell-based ATMPs: automated, accurate, and affordable.

Abstract

The production of neuroepithelial stem (NES) cells, a promising therapeutic candidate for neurological conditions such as stroke and spinal cord injuries, faces significant manufacturing challenges, particularly in quality control (QC). Specifically, current QC methods rely on labor-intensive manual assessments that are costly, time-consuming, and subject to operator bias, limiting scalability and reproducibility. The aim of this study is to develop an automated, accurate and affordable QC method for NES cells. This study addresses the above mentioned challenges by developing a Convolutional Neural Network (CNN)-based model for automating morphological QC during NES cell expansion. The model achieved an area under the receiver operating characteristic curve (AUROC) of > 0.997 when applied to test data, with robustness validated via image rotation and segmentation. The model successfully distinguished key morphological features, such as rosette-like structures in high-quality cultures, and detected pronounced heterogeneity in low-quality samples. Complementary flow cytometry validation confirmed the biological relevance of the model's predictions. By integrating this CNN-based QC system, we offer a scalable and cost-effective solution for real-time monitoring, batch consistency, and reduced process variability. This approach supports the clinical translation of NES cell therapies and holds broader applicability across other cell-based advanced therapy medicinal products.