AI Improves Bronchoscopic Navigation to Difficult-to-Reach Lung Lesions

Lung cancer is a leading cause of cancer mortality, and early detection improves survival. Reaching small, peripheral lung tumors for biopsy remains difficult because bronchoscopic navigation depends on complete airway maps. Many peripheral branches are extremely thin on CT, leaving navigation systems vulnerable to gaps. To help address this challenge, researchers have developed an AI framework that identifies overlooked airways and generates more complete maps for bronchoscopy.

ASTRA-Net (Anatomical Segmentation with Tree-aware Refinement Attention) was developed at Pusan National University to improve airway segmentation on CT. The framework targets the clinical need for reliable guidance to peripheral lesions by explicitly searching for branches likely missed during manual annotation. Its design aims to produce anatomically plausible extensions that can enhance three-dimensional bronchoscopic roadmaps.

The system uses a multistage deep-learning architecture in which one component learns global lung structure and another refines regions with uncertain boundaries. An attention mechanism directs the model toward areas where small airways are difficult to distinguish or absent from labels. By incorporating contextual anatomical cues from surrounding lung tissue and parallel blood vessels, the model can infer the continuity of previously unrecognized airway paths.

Validation included multiple datasets and clinical CT scans from Pusan National University Yangsan Hospital. The model demonstrated strong performance in identifying fine peripheral airways and remained robust despite variations in CT image quality and slice thickness. Expert review determined that some predictions first counted as false positives were genuine branches omitted from the original annotations.

The work, involving the School of Biomedical Convergence Engineering at Pusan National University and the Division of Pulmonary and Critical Care Medicine at the Research Institute for Convergence of Biomedical Science and Technology, Pusan National University Yangsan Hospital, was published in IEEE Transactions on Medical Imaging in 2026. The researchers indicate that more complete airway maps could support improved navigation to difficult-to-reach lesions and future AI-assisted or robotic bronchoscopy platforms.

“ASTRANET is not simply another airway segmentation model. It was specifically designed to identify peripheral airways that may have been overlooked during manual annotation, thereby helping to create a more complete roadmap for bronchoscopy,” said MinWoo Kim of the School of Biomedical Convergence Engineering, Pusan National University.

“By providing physicians with a more complete airway roadmap, we hope this technology will improve navigation to difficult-to-reach lung lesions and support the development of future AI-assisted and robotic bronchoscopy systems,” added Kim.

Related Links
Pusan National University