AI based quantification of epidermal proliferation and apoptosis in human skin

Researchers developed and validated an artificial intelligence (AI)-based workflow to quantify epidermal proliferation and apoptosis in immunohistochemically stained human skin. Using human skin cultured on the chorioallantoic membrane (CAM) model, the study combined a convolutional neural network (CNN)-based cell-detection model with a semantic segmentation model that identifies the epidermis. This combined approach enabled automated quantification of Ki-67-positive proliferating cells and cleaved Caspase-3 (CASP3)-positive apoptotic cells specifically within the epidermis. Validation against manual cell counts demonstrated high agreement, with mean accuracies of 95.43% for Ki-67 and 97.0% for CASP3, while maintaining low false-positive rates across staining batches and experimental conditions.

The AI workflow was implemented using the IKOSA image analysis platform (KOLAIDO GmbH), which was used to create, train and deploy the AI models for epidermis segmentation and cell detection without coding. The platform enabled analysis of whole-slide images by combining epidermis recognition with cell classification, restricting quantification to the epidermal compartment. The resulting workflow generated standardized quantitative outputs and visual overlays, reducing subjective bias and providing reproducible measurements with relatively modest annotation effort. The authors also demonstrated that the cell-detection model trained on Ki-67 could be applied to CASP3-stained sections without retraining because both markers exhibited comparable nuclear DAB staining morphology.

To demonstrate the workflow in a biological application, the researchers evaluated the effects of topical hydroxytyrosol (HT) treatment on human skin cultured in the CAM model. AI-assisted analysis showed that six days of HT treatment significantly reduced epidermal cell proliferation, as measured by Ki-67, while apoptosis, measured by CASP3, remained unchanged. The authors conclude that the validated AI workflow provides an objective and scalable approach for quantifying epidermal regeneration in experimental human skin and may support future translational dermatological research and dermatopathology applications.