A quick and reliable image-based AI algorithm for evaluating cellular senescence of gastric organoids

Objective: Organoids are a powerful tool with broad application prospects in biomedicine. Notably, they provide alternatives to animal models for testing potential drugs before clinical trials. However, the number of passages for which organoids maintain cellular vitality ex vivo remains unclear. Methods: Herein, we constructed 55 gastric organoids from 35 individuals, serially passaged the organoids, and captured microscopic images for phenotypic evaluation. Senescence-associated β-galactosidase (SA-β-Gal), cell diameter in suspension, and gene expression reflecting cell cycle regulation were examined. The YOLOv3 object detection algorithm integrated with a convolutional block attention module (CBAM) was used to evaluate organoid vitality. Results: SA-β-Gal staining intensity; single-cell diameter; and expression of p15, p16, p21, CCNA2, CCNE2, and LMNB1 reflected the progression of aging in organoids during passaging. The CBAM-YOLOv3 algorithm precisely evaluated aging organoids on the basis of organoid average diameter, organoid number, and number × diameter, and the findings positively correlated with SA-β-Gal staining and single-cell diameter. Organoids derived from normal gastric mucosa had limited passaging ability (passages 1–5), before aging, whereas tumor organoids showed unlimited passaging potential for more than 45 passages (511 days) without showing clear senescence. Conclusions: Given the lack of indicators for evaluating organoid growth status, we established a reliable approach for integrated analysis of phenotypic parameters that uses an artificial intelligence algorithm to indicate organoid vitality. This method enables precise evaluation of organoid status in biomedical studies and monitoring of living biobanks.