Abstract
Endocytoscopy enables real-time observation of lesions at ultra-magnification. In the gastrointestinal and respiratory fields, endocytoscopic images are similar to hematoxylin-eosin-stained images. This study aimed to compare the nuclear features of pulmonary lesions in endocytoscopic and hematoxylin-eosin-stained images. We performed an endocytoscopy to observe resected specimens of normal lung tissue and lesions. Nuclear features were extracted using ImageJ. We analyzed five nuclear features: nuclear number per area, mean nucleus area, median circularity, coefficient of variation of roundness, and median Voronoi area. We conducted dimensionality reduction analyses for these features, followed by assessments of the inter-observer agreement among two pathologists and two pulmonologists to evaluate endocytoscopic videos. We analyzed the nuclear features of hematoxylin-eosin-stained and endocytoscopic images from 40 and 33 cases, respectively. Endocytoscopic and hematoxylin-eosin-stained images displayed a similar tendency for each feature, despite there being no correlation. Conversely, the dimensionality reduction analyses demonstrated similar distributions of normal lung and malignant clusters in both images, thus differentiating between the clusters. The diagnostic accuracy of the pathologists was 58.3% and 52.8% (κ-value 0.38, fair), and that of the pulmonologists was 50% and 47.2% (κ-value 0.33, fair). The five nuclear features of pulmonary lesions were similar in the endocytoscopic and hematoxylin-eosin-stained images.