Education&ScienceEducational Article

Prediction of Lymph Node Metastasis in Lung Cancer Using Deep Learning of Endobronchial Ultrasound Images with Size on CT and PET-CT Findings

June 2025

Authors: Ji Eun Oh, Hyun Sung Chung, Hye Ran Gwon, Eun Young Park, Hyae Young Kim, Geon Kook Lee, Tae-Sung Kim, Bin Hwangbo

Comment by Mark Lavercombe: Endobronchial ultrasound (EBUS) with transbronchial needle aspiration (TBNA) is the current standard for minimally invasive mediastinal lymph node staging for patients with lung cancer, however various factors might influence which lymph nodes are targeted by the bronchoscopist. In this paper, the authors report a Deep Learning model developed using findings from computed tomography (CT), positron emission tomography (PET) and EBUS images, as well as EBUS-TBNA and surgical pathologic diagnoses. Their model utilising CT size, PET avidity and EBUS region of interest labelling achieved an area under the receiver operating curve of 0.914 with accuracy of 82.3%, significantly outperforming other models. However, the positive and negative predictive values depend on the prevalence of metastasis, and the authors acknowledge that further research will be required before reaching clinical utility.