Deep Learning-Guided Holotomography Reveals Early Structural Remodelling During Pluripotency Exit

Real-time assessment of human pluripotent stem cell (hPSC) quality is critical for reproducibility and safety in regenerative medicine, yet current methods are invasive, labor-intensive, or highly operator-dependent. We present DeepHOPE (Deep-learning-guided Holotomography for Pluripotency Evaluation), a non-invasive, automatizable, and statistics-driven platform that integrates three-dimensional (3D) refractive index imaging with deep learning to assess pluripotency. DeepHOPE performs robustly across diverse contexts, including germ-layer differentiation, retinoic acid-induced differentiation, and mid-reprogramming cultures, enabling streamlined cell production workflows and improving the efficiency of midbrain dopaminergic neuron differentiation through informed colony selection. Mechanistically, DeepHOPE detects minute colony-scale topological changes that precede molecular loss of pluripotency. These early changes are associated with rapid F-actin remodeling, including apical-to-basal redistribution during early differentiation. Consistent with a functional role for cytoskeletal regulation in state transitions, sustained reduction of actomyosin tension decreases pluripotency, identifying cytoskeletal dynamics as an upstream determinant of early pluripotency exit.