Automatic Segmentation of Mediastinal Lymph Nodes and Blood Vessels in Endobronchial Ultrasound (EBUS) Images Using Deep Learning

• DOI: 10.3390/jimaging10080190
• Publication Date: 2024-08-06T19:24:16Z
• Authors: Øyvind Ervik; Ingrid Tveten; Erlend Fagertun H...; Thomas Langø; Håkon Olav Leira; Tore Amundsen; Hanne Sorger
• Keywords: Endobronchial ultrasound; Mediastinal lymph node; Sørensen–Dice coefficient

Endobronchial ultrasound (EBUS) is used in the minimally invasive sampling of thoracic lymph nodes. In lung cancer staging, accurate assessment mediastinal structures essential but challenged by variations anatomy, image quality, and operator-dependent interpretation. This study aimed to automatically detect segment nodes blood vessels employing a novel U-Net architecture-based approach EBUS images. A total 1161 images from 40 patients were annotated. For training validation, 882 30 145 5 utilized. separate set 134 was reserved for testing. node vessel segmentation, mean ± standard deviation (SD) values Dice similarity coefficient 0.71 0.35 0.76 0.38, those precision 0.69 0.36 0.82 0.22, sensitivity 0.38 0.80 0.25, specificity 0.98 0.02 0.99 0.01, F1 score 0.85 0.16 0.81 0.21, respectively. The average processing segmentation run-time per 55 1 ms (mean SD). new (EBUS-AI) could method performed well feasible fast, enabling real-time automatic labeling.