Objective  To develop a deep learning-based plug-in for automated, multidimensional quantitative analysis of tight junctions (TJs), addressing methodological bottlenecks in epithelial barrier morphological analysis, such as strong subjectivity and low throughput, and to validate the tool using an in vitro model of allergic rhinitis (AR).

Methods Based on the ImageJ platform, an automated multidimensional image-analysis workflow was established by integrating Cellpose-based single-cell segmentation, fluorescence quantification within region of interest, and Ridge Detection algorithms for linear-structure detection. Primary human nasal epithelial cells (hNECs) were used as the model and were stimulated with interlukin-13 (IL-13) and interlukin-17A (IL-17A), respectively, followed by immunofluorescence staining and transmission electron microscopy. The consistency between the automated plug-in and conventional manual analysis was evaluated, and the extent of TJ disruption induced by inflammatory cytokines was quantitatively assessed.

Results The plug-in showed strong correlations with manual analysis across four core dimensions, including fluorescence intensity (r = 0.9954, P ＜ 0.0001), network coverage area (r = 0.9964, P ＜ 0.0001), cell perimeter, and intercellular gap width (r = 0.9979, P ＜ 0.0001), and intercellular gap width (r = 0.972, P ＜ 0.0001). Both IL-13 and IL-17A significantly disrupted the TJ network structure of hNECs, manifested by increased cell perimeter and abnormally widened intercellular gaps, and the plug-in accurately captured and quantified these pathological changes.

Conclusion The automated multidimensional TJ quantification tool developed in this study provides a reliable methodological support for standardized and automated assessment of the physical barrier function of the respiratory epithelium.