AI Tool Maps Overlooked Lung Airways to Help Reach Hard-to-Reach Tumors

About the Research

A university in the Republic of Korea is leveraging artificial intelligence to assist doctors in detecting small lung tumors in hard-to-reach areas of the lungs. Researchers at Pusan National University developed ASTRA-Net (Anatomical Segmentation with Tree-aware Refinement Attention), an AI framework that uses CT scans to identify overlooked peripheral airways, creating a more complete map to guide doctors during bronchoscopy.

Locating small tumors deep within the lungs is challenging, and CT scans can miss small or complex airways. ASTRA-Net addresses this by providing a detailed view of the lung's airway network, improving navigation to difficult-to-access areas.

How ASTRA-Net Works

ASTRA-Net employs multi-stage deep learning networks to learn the overall structure of the lungs and identify smaller airway branches. The tool is designed to recover peripheral airways that may have been missed in manual annotations or CT scans, offering a more accurate roadmap for bronchoscopy.

Researchers found that some areas initially labeled as false positives were actually genuine airway branches that had been excluded from original annotations. The model demonstrated strong performance in identifying these peripheral airways.

Research Findings and Impact

Dr. MinWoo Kim, Dr. Hee Yun Seol, and other researchers published their findings on ASTRA-Net this year, emphasizing its potential to enhance lung cancer detection and treatment. Lung cancer is the leading cause of cancer-related deaths worldwide, with chest X-rays missing about one-quarter of cases and radiologists sometimes overlooking abnormalities due to observer error.

Most lung navigation systems rely on reconstructed CT scans, which may miss smaller airways. ASTRA-Net aims to fill these gaps, potentially improving navigation to difficult-to-reach lung lesions and supporting the development of AI-assisted and robotic bronchoscopy systems.

“ASTRA-Net 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 Dr. Kim. “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.”