Validation of a commercially available CAD-system for lung nodule detection and characterization using CT-scans

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Computer-Aided Detection (CAD) systems offer potential for automating lung nodule (LN) analysis on CT scans.
• External validation of commercial CAD systems is crucial for assessing their real-world performance.
• Current validation studies often focus solely on nodule detection, neglecting characterization aspects.

Purpose of the Study:

• To externally validate a commercial CAD system for the detection and characterization of solid, part-solid, and ground-glass lung nodules (LN).
• To assess the standalone performance of CAD in terms of sensitivity and false-positive rates.
• To evaluate CAD's characterization accuracy for nodule features like texture, calcification, spiculation, and location.

Main Methods:

• Retrospective analysis of 263 chest CT scans from a Dutch university hospital.
• Comparison of CAD performance against radiologist (R1) and adjudicating radiologist (R2) assessments.
• Calculation of CAD's detection sensitivity, false-positive rate, and characterization accuracy for various LN types and features.

Main Results:

• CAD detected 149 of 183 true nodules (81.4% sensitivity) with a false-positive rate of 0.405.
• Detection sensitivity varied by nodule type: solid (87.2%), part-solid (89.4%), and ground-glass (59.5%).
• Classification accuracy was high for calcification (91.9%) and location (94.6%) but lower for texture (77.2%) and part-solid nodules (38.1%).

Conclusions:

• The CAD system demonstrates potential to augment radiologist detection rates for lung nodules.
• While overall detection performance is slightly lower than a single radiologist, CAD's ability to identify missed nodules is valuable.
• Further improvement in texture classification is necessary for comprehensive characterization of lung nodules using this CAD system.