The U.S. Food and Drug Administration (FDA) has permitted drug testing without animal models because of ethical concerns and the translational gap between human and animal drug responses. Consequently, research using human-derived cells to build more ...
The U.S. Food and Drug Administration (FDA) has permitted drug testing without animal models because of ethical concerns and the translational gap between human and animal drug responses. Consequently, research using human-derived cells to build more accurate and precise disease models and drug-evaluation systems has accelerated, and such approaches are increasingly adopted as alternative test methods for assessing the efficacy and safety of cosmetics. The skin is the largest organ directly exposed to the external environment, and keratinocytes play a central role in its barrier function. HaCaT cells an immortalized human keratinocyte cell line can recapitulate diverse skin related responses and are widely used in drug and cosmetic research. This study aims to establish a criterion for determining appropriate time points for drug treatment and measurement by quantitatively analyzing cell numbers. HaCaT cells were cultured on transparent, electrically conductive ITO electrode chips (16 and 32 channel), and both electrical impedance analysis and Artificial Intelligence (AI) based image analysis were performed in parallel. Impedance measurements enabled real time monitoring of cell attachment, proliferation, and growth. In parallel, time-lapse images were analyzed using StarDist, a deep learning-based cell segmentation tool, to automatically count cells and visualize proliferation dynamics over time. The HaCaT image dataset was acquired at fixed time intervals and re-labeled according to a No Reference Image Quality Assessment (NR-IQA) criterion. Based on this, a pretrained StarDist model was fine tuned to improve predictive accuracy. Model performance was evaluated using Intersection over Union (IoU), cell count error analysis, and the coefficient of determination (R²) with the impedance data. The finetuned model showed improved recognition accuracy and reduced error compared with the pretrained baseline while maintaining a high correlation with impedance analysis. This integrated approach enabled rapid and objective assessment of both morphological and physiological changes in cells and compensated for unstable early time impedance values. Combining the two modalities demonstrated more precise interpretation of cellular responses. Overall, by integrating noninvasive, continuous impedance monitoring with AI-based analysis to quantify the optimal timing for drug treatment, this work suggests technical implications for future precision diagnostics, drug development, and cosmetic safety evaluation.