Feasibility of artificial intelligence-based rapid on-site evaluation for the diagnosis of pulmonary disease.

Introduction: This study aimed to investigate the clinical feasibility of artificial intelligence-rapid on-site evaluation (AI-ROSE) based on exfoliated cell blotting from percutaneous puncture biopsy specimens for diagnosing pulmonary lesions and to provide a reference for intraoperative rapid diagnosis. Methods: A total of 266 patients with pulmonary lesions who underwent computed tomography (CT)-guided percutaneous core needle biopsy between June 11 and November 27, 2024, were enrolled. Exfoliated cell prints from biopsy specimens were stained with Diff-Quik, followed by diagnosis using AI-ROSE. Using the final histopathological diagnosis as the standard, the accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of AI-ROSE and conventional cytological diagnoses were compared. The consistency between the methods and histopathological diagnosis was analyzed. The diagnostic times for AI-ROSE, cytology, and histopathology were recorded and compared. Postoperative complications were documented, and the correlation between lesion characteristics and complications was analyzed. Results: Compared with histopathological results, AI-ROSE achieved a sensitivity of 95.67%, specificity of 79.31%, diagnostic accuracy of 92.11%, PPV of 94.31%, and NPV of 83.64% in diagnosing pulmonary lesions, with good consistency (κ = 0.764, P < 0.001). No significant difference in the AI-ROSE diagnostic accuracy was observed among the different pathological types (P > 0.05). Conventional cytological diagnosis showed a sensitivity of 87.50%, specificity of 85.71%, and accuracy of 87.10%, with lower consistency and histopathology (κ = 0.665, P < 0.001) than those of AI-ROSE. The mean diagnostic time of AI-ROSE was 254.60 ± 13.88 s, which was significantly shorter than that of cytology (1.48 ± 0.86 days) and histopathology (2.37 ± 1.42 days). The overall incidence of postoperative complications was 22.93%, including pneumothorax (12.78 %) and needle tract bleeding/mild hemoptysis (9.77%), with no fatal complications. A smaller nodule volume was associated with a higher risk of puncture bleeding (P = 0.004), whereas lesion size and puncture path length were not significantly correlated with pneumothorax risk (P > 0.05). Discussion: AI-ROSE enables rapid intraoperative diagnosis of pulmonary lesions in percutaneous biopsy with high diagnostic performance and consistency with histopathological results, demonstrating its favorable value for clinical application and popularization.