Application of artificial intelligence to digital-rapid on-site cytopathology evaluation during endoscopic ultrasound-guided fine needle aspiration: A proof-of-concept study

Abstract

Background: During endoscopic ultrasound-guided fine needle aspiration (EUS-FNA), cytopathology with rapid on-site evaluation (ROSE) can improve diagnostic yield and accuracy. However, ROSE is unavailable in most Asian and European institutions because of the shortage of cytopathologists. Therefore, developing computer-assisted diagnostic tools to replace manual ROSE is crucial. Herein, we reported the validation of an artificial intelligence (AI)-based model (ROSE-AI model) to substitute manual ROSE during EUS-FNA.

Methods: A total of 467 digitized images from Diff-Quik (D&F)-stained EUS-FNA slides were divided into training (3642 tiles from 367 images) and internal validation (916 tiles from 100 images) datasets. The ROSE-AI model was trained and validated using training and internal validation datasets, respectively. The specificity was emphasized while developing the model. Then, we evaluated the AI model on a 693-image external dataset. We assessed the performance of the AI model to detect cancer cells (CCs) regarding the accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).

Results: The ROSE-AI model achieved an accuracy of 83.4% in the internal validation dataset and 88.7% in the external test dataset. The sensitivity and PPV were 79.1% and 71.7% in internal validation dataset and 78.0% and 60.7% in external test dataset, respectively.

Conclusion: We provided a proof of concept that AI can be used to replace manual ROSE during EUS-FNA. The ROSE-AI model can address the shortage of cytopathologists and make ROSE available in more institutes.

Keywords: AI; EUS-FNA; ROSE; accuracy; cytopathologists.