Applying Artificial Intelligence in the Bronchoscopy Suite

A team of investigators studied an artificial intelligence model’s accuracy in classification of images obtained from rapid on-site examination of EBUS-TBNA smears in the bronchoscopy suite.

A proof-of-concept study suggests that artificial intelligence (AI) may classify images captured during rapid onsite examination of endobronchial ultrasound guided transbronchial need aspiration (EBUS-TBNA) with high accuracy. The results of this study were published in the European Respiratory Journal.

The use of AI in medicine has become more common in areas such as cervical cancer screening, which has led experts to question its potential in other fields of medicine. No data have been published on the application of AI during rapid on-site examination of EBUS-TBNA. A team of investigators “evaluated the performance of an AI model, consisting of an open-sounded convolutional neural network using transfer learning, for its ability to accurately classify images of [rapid onsite examination] of EBUS-TBNA smears in the bronchoscopy suite.”

The researchers retrospectively collected 441 cytology images of rapid onsite examination of EBUS-TBNA smears from patients who underwent bronchoscopy from June 2019 to June 2020. These images were classified into 1 of 4 categories including granulomas (n=47), adequate lymphocytes (more than 40 lymphocytes per high power field; n=212), malignant cells (n=80), and inadequate (meeting none of the criteria; n=102).

Images were separated to either form a test set (66 images) or the training set that was used for the AI model and its evaluation (375 images). The model used was a 19-layered VGG19 model by the Visual Geometry Group. Following 25 epochs of training, it was found that epoch 13 had the highest accuracy (93.4%) and the lowest log-loss (0.18), and it was subsequently used for further evaluation.

To determine the validity of the model to differentiate between adequate and inadequate smears, the investigators combined the 51 smears in the test set that made up the granuloma, adequate lymphocyte, and malignant cell categories and compared them with 15 images of inadequate smears. The model was found to have a 96.1% sensitivity and a 93.3% specificity to differentiate adequate vs inadequate smears.

“This [Digital-Rapid On-Site Examination in Endobronchial Ultrasound-Guided Transbronchial Needle Aspiration] study lays the groundwork to develop similar AI models and opens up the possibility of automating the process which may further result in a substantial reduction in time and expense associated with the procedure in the bronchoscopy suite,” the authors concluded. 


Asfahan S, Elhence P, Dutt N, Jalandra RN, Chauhan NK. Digital-Rapid On-site Examination in Endobronchial Ultrasound-Guided Transbronchial Needle Aspiration (DEBUT) – a proof of concept study for the application of artificial intelligence in the bronchoscopy suite. Eur Respir J. Published online June 17, 2021. doi:10.1183/13993003.00915-2021