New AI Tool Boosts Detection of Airway Nodules

TOPLINE:

A deep learning–based artificial intelligence (AI) system detected 75.1% of airway nodules, with 0.25 false positives per scan in patients without airway nodules. The system achieved an increased sensitivity of 79% in detecting tumorous nodules, including subtle cases often missed by radiologists.

METHODOLOGY:

• Researchers retrospectively analysed a dataset of 320 chest and chest-abdomen CT scans, including patients with airway nodules (n = 160; median age, 64 years; 58 women) and those without airway nodules (n = 160; median age, 60 years; 80 women).
• An airway nodule was defined as a focal opacity primarily confined within the lumen of the airway (bronchus or trachea).
• Primary cancers were verified through bronchoscopy with biopsy or cytologic testing. The malignancy status of other nodules was confirmed using bronchoscopy alone or follow-up CT scans.
• An AI system ꟷ a deep learning–based computer-aided detection (DL-CAD) system ꟷ was trained and evaluated using 10-fold cross-validation, and its performance was assessed through free-response receiver operating characteristic curves.

TAKEAWAY:

• The AI system demonstrated a sensitivity of 75.1% (95% CI, 67.6-81.6%) for detecting all airway nodules, with an average of 0.25 false positives per scan in patients without airway nodules and 0.56 false positives per scan in those with airway nodules.
• For tumorous nodules, the system achieved a sensitivity of 79.0% (95% CI, 70.4-86.6%), with a similar sensitivity of 78.1% (95% CI, 68.2-86.4%) for malignant tumours.
• The sensitivity for detecting non-tumorous nodules was 71.4% (95% CI, 60.3-82.3%), with the same false positive rates as those for detecting all airway nodules.
• A subgroup analysis showed that the system could detect most of the subtle tumours.

IN PRACTICE:

“In conclusion, we found that a DL-CAD system can detect most benign and malignant airway nodules in routine clinical chest CT scans with an acceptable false positive rate. Despite the rarity of these nodules, our findings demonstrate the technical feasibility of developing a DL-CAD system that could be useful for radiologists in real-world clinical settings,” the authors wrote.

SOURCE:

The study was led by Ward Hendrix, Radboud University Medical Center, Nijmegen, the Netherlands. It was published online on March 5, 2025, in European Radiology.

LIMITATIONS:

The study was limited by the lack of external validation data for the AI system. The limited size of the training samples may have affected the system’s performance. Additionally, the authors acknowledged that no observer study was conducted to compare the performance of the AI system with that of radiologists.

DISCLOSURES:

The study received support from the Junior Researcher grant provided by the Radboud Institute for Health Sciences, Radboud University Medical Center, and Jeroen Bosch Hospital. Several authors declared having various ties with various sources.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.