Reliability of Rapid On-Site Evaluation Achieved by Remote Sharing Systems (E-ROSE) and AI Algorithms (AI-ROSE) Compared With the Gold Standard in the Diagnosis of Lung Cancer.

Background and objective: In recent decades, artificial intelligence has seen significant development in various fields of medicine, including interventional pulmonology. The study aims to evaluate the diagnostic performance of innovative approaches to detect lung cancer on biopsy sample images (Rapid On-Site Evaluation, ROSE) compared to the diagnostic gold standard.

Methods: We conducted a multicentric study, comparing remote anatomopathological evaluation (E-ROSE) and machine learning algorithms (AI-ROSE) reliability in diagnosing lung cancer, evaluating 277 biopsy sample images, 25 of which were doubtful; to compare them with the definitive histological examination performed by the pathologist.

Results: E-ROSE achieved a diagnostic accuracy of 95.5%, with a sensitivity of 99.0% and specificity of 88.7%, including doubtful cases respectively 91.4%, 97.1%, and 81%. AI-ROSE showed a sensitivity of 96.4% and a specificity of 78.9%, with an accuracy of 92.5%. Including the doubtful cases, the best model achieved an accuracy of 85%, sensitivity of 97.4%, and specificity of 75.4%. The discriminative ability of the tests was compared both for positive/negative cases, showing Area Under ROC Curve (AUC) of 93.9% for E-ROSE and 87.6% for AI-ROSE; while including doubtful, AUC was 89.1% for E-ROSE and 86.4% for AI-ROSE.

Conclusions: The study suggests that the application of innovative methods such as E-ROSE and AI-ROSE could provide valuable support to interventional pulmonologists in the diagnostic process.