Computer-aided detection of endobronchial valves using volumetric CT

Area of Science:

• Medical Imaging
• Pulmonology
• Biomedical Engineering

Background:

• Implanted devices require monitoring for treatment efficacy.
• Automated detection can improve patient care and evaluation.
• One-way endobronchial valves (EBVs) are used for less invasive lung volume reduction.

Purpose of the Study:

• Develop an automated system for detecting one-way endobronchial valves (EBVs).
• Evaluate the system's performance in identifying implanted devices for lung volume reduction.

Main Methods:

• Utilized volumetric thin-section computed tomographic (CT) data from 194 subjects.
• Developed a detection system involving preprocessing, pattern recognition, and classifier training (AdaBoost).
• Employed local region features and a template model for valve detection.

Main Results:

• Achieved 96.5% sensitivity with 0.18 false positives per subject.
• Incorporating device count improved performance to 96.9% sensitivity and 0.061 false positives per subject.
• The test dataset had 12 false negatives and 6 false positives across 99 subjects.

Conclusions:

• Successfully developed software for automated endobronchial valve detection on CT scans.
• The proposed techniques may be applicable to other implanted devices.
• The system can aid in evaluating treatment response for lung volume reduction.