Abstract
PURPOSE: Bronchiolar adenoma (BA) is a rare benign pulmonary neoplasm originating from the bronchial mucosal epithelium and mimics lung adenocarcinoma (LAC) both radiographically and microscopically. This study aimed to develop a nomogram for distinguishing BA from LAC by integrating clinical characteristics and artificial intelligence (AI)-derived histogram parameters across two medical centers. METHODS: This retrospective study included 215 patients with diagnoses confirmed by postoperative pathology from two medical centers. Medical center 1 provided 151 patients (68 BA and 83 LAC nodules) as the training cohort, while medical center 2 contributed 64 patients (28 BA and 36 LAC nodules) as the external validation cohort. Risk predictors and the nomogram were developed using clinical characteristics and AI-derived histogram parameters. RESULTS: Nodule density (solid, ground glass, and subsolid) exhibited a statistically significant difference between the BA and LAC groups (p < 0.01). The following parameters were significantly higher in the LAC group compared to the BA group (all p < 0.05): 2D long diameter, 2D short diameter, 2D average diameter, 2D maximum surface area, 3D long diameter, 3D surface area, 3D volume, and entropy. In contrast, CT value variance was significantly lower in the LAC group than in the BA group (p < 0.01). A nomogram was constructed incorporating density, 2D short diameter, and CT value variance. The area under the curve (AUC) of the nomogram in the training and validation cohorts were 0.821, 0.811. CONCLUSION: The AI-based nomogram, as a non-invasive preoperative tool, had the potential to enhance diagnostic accuracy for distinguishing BA from LAC.