Automated lung segmentation for thoracic CT impact on computer-aided diagnosis

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Automated lung segmentation in thoracic computed tomography (CT) scans is crucial for developing computer-aided diagnostic (CAD) systems.
• A single, universally applicable segmentation method may not suffice, as specific clinical tasks often necessitate tailored approaches.

Purpose of the Study:

• To evaluate an automated lung segmentation method for two distinct CAD applications: automated lung nodule detection and computer-assisted measurement of pleural mesothelioma tumor thickness.
• To determine if modifications to a core lung segmentation method improve performance for these specific tasks.

Main Methods:

• A core automated lung segmentation method using gray-level thresholding was applied.
• Modifications included separating lungs, removing the trachea/bronchi, suppressing the diaphragm, and employing a rolling ball algorithm and morphologic erosion for nodule detection.
• The core method and modified versions were tested on nodule detection and mesothelioma thickness measurement tasks.

Main Results:

• For nodule detection, modifying the segmentation excluded 4.9% of nodules compared to 17.1% without modifications.
• For mesothelioma measurement, the core method alone achieved a correlation of 0.990 with manual measurements.
• Applying nodule detection modifications to mesothelioma measurement reduced the correlation to 0.977.

Conclusions:

• The requirements for automated lung segmentation vary significantly depending on the specific CAD application.
• Adapting the lung segmentation approach to the particular CAD task is essential for optimal performance in medical image analysis.