Influence of background lung characteristics on nodule detection with computed tomography

Area of Science:

• Radiology and Medical Imaging
• Pulmonary Medicine
• Computational Pathology

Background:

• Pulmonary nodule detection on chest computed tomography (CT) is crucial for early lung cancer diagnosis.
• Lung complexity can influence the accuracy of nodule detection, leading to missed diagnoses.
• Understanding factors contributing to detection variability is essential for improving radiologist performance.

Purpose of the Study:

• To characterize local lung complexity in chest CT images.
• To evaluate the impact of various image features on pulmonary nodule detectability.
• To identify radiologist-specific factors influencing nodule detection accuracy.

Main Methods:

• Creation of 40 volumetric chest CT scans with embedded simulated pulmonary nodules.
• Evaluation of 157 nodules by 13 radiologists, yielding 2041 detection opportunities.
• Measurement of 14 image features around nodule locations and statistical modeling (generalized linear mixed-effects model) to assess feature impact on detectability.

Main Results:

• Five categories of image features (local structural distractors, local intensity, global context, local vascularity, contiguity with structural distractors) were significant factors influencing nodule detection.
• Reader-specific models revealed significant differences in distraction types, with local intensity and local structure distraction being mutually exclusive.
• Radiologists distracted by local intensity detected significantly fewer nodules (46.1%) compared to those distracted by local structure (65.3%).

Conclusions:

• Local lung complexity, encompassing various image features, significantly impacts pulmonary nodule detectability on chest CT.
• The type of image-based distraction (local intensity vs. local structure) differentially affects radiologist performance in nodule detection.
• Targeted training or image analysis tools addressing specific distraction types may improve radiologist accuracy in pulmonary nodule detection.