A Reproducible Microfluidic Platform Combined with Deep Learning Image Analysis Enables Quantitative Single-Cell Characterization of Keratinocyte Responses to Flow

Abstract Advanced in vitro models require physiologically relevant culture conditions together with objective and reproducible methods for quantitative cell analysis. Here, we present an integrated microfluidic tissue culture platform combining controlled recirculating flow with a Human-in-the-Loop (HITL) deep learning workflow for quantitative single-cell analysis of dense keratinocyte cultures. The platform comprises a custom 3D-printed microfluidic bioreactor with automated flow regulation and was experimentally validated for biocompatibility. Immortalized human keratinocytes (HaCaT) cultured under continuous flow ($50\,\mu\mathrm{L/min}$) were compared with static controls. Fluorescence microscopy images were analyzed using a computational pipeline based on Z-projections, spatial upsampling, and iterative HITL fine-tuning of the Cellpose (cyto3) model. The optimized workflow improved single-cell detection by more than $140\%$ compared with pretrained generalist segmentation models, enabling robust quantification of cell number, morphology, viability, and fluorescence marker expression in highly confluent cultures. Flow significantly increased metabolic activity, proliferation, cell area and diameter, and keratin 5 expression while preserving cell morphology. Under differentiation conditions, flow further enhanced keratin 10 (an early differentiation marker) expression while maintaining cell viability. This integrated platform provides a scalable framework for objective evaluation of advanced in vitro microphysiological systems, supporting biomedical research, Organ-on-a-Chip technologies, and New Approach Methodologies (NAMs).