AI-enabled label free monitoring of TGFβ1-induced remodeling in iPSC-derived alveolar organoids.

Pulmonary fibrosis (PF) is a progressive interstitial lung disease with poor prognosis and limited therapeutic options, largely due to the lack of physiologically relevant and dynamically traceable preclinical models. Here, we establish an induced pluripotent stem cell (iPSC)-derived alveolar organoid platform that faithfully recapitulates PF-like remodeling upon TGFβ1 stimulation. The organoids exhibit lineage commitment to the distal lung epithelium (NKX2.1, SFTPC/SFTPB) and, following TGFβ1 exposure, undergo hallmark fibrotic changes including organoid condensation and size reduction, collagen deposition (Masson's trichrome), myofibroblast activation (α-SMA), up-regulation of profibrotic genes (COL1A1, FN, VIM, ACTA2), and partial EMT-like reprogramming (↑CDH2, TWIST1, and ↓CDH1). To enable label-free, longitudinal readouts, we integrate a deep neural network (YOLOv8-nano) that detects subtle morphological cues directly from bright-field images and classifies treatment status with high fidelity. Across augmented datasets (8892 images), the model achieved strong performance on the original context-preserving images (mAP50-95 up to 0.95; high precision/recall and 98-99% true-positive rates), supporting robust discrimination of control versus TGFβ1-treated organoids. This AI-enhanced organoid system provides a quantitative, label-free platform for monitoring fibrotic remodeling and offers a scalable foundation for preclinical antifibrotic screening and mechanism-of-action studies.