Performance of Lung Nodule Computer-Aided Detection Systems on Standard-Dose and Low-Dose Pediatric CT Scans: An...

Area of Science:

• Radiology and Imaging
• Artificial Intelligence in Medicine
• Pediatric Pulmonology

Background:

• Computer-aided detection (CAD) systems for lung nodules may experience reduced performance on low-dose pediatric CT scans due to increased image noise.
• Adult-trained CAD systems are increasingly applied to pediatric imaging, raising questions about their efficacy across different dose levels.

Purpose of the Study:

• To compare the performance of two CAD systems trained on adult data for lung nodule detection between low-dose and standard-dose pediatric chest CT scans within the same patients.
• To evaluate the impact of image noise on CAD system sensitivity in pediatric populations.

Main Methods:

• Retrospective analysis of 73 pediatric patients who underwent both standard-dose and low-dose chest CT scans.
• Two publicly available CAD systems (FlyerScan and MONAI), trained on adult data, were used to analyze nodules (3-30 mm).
• Sensitivity was calculated for each system at a fixed false-positive rate, comparing performance between dose levels.

Main Results:

• Both CAD systems demonstrated decreased sensitivity on low-dose scans compared to standard-dose scans.
• FlyerScan sensitivity dropped from 76.9% (standard-dose) to 66.8% (low-dose); MONAI sensitivity decreased from 67.6% to 62.3%.
• The reduction in detection was more pronounced for nodules smaller than 5 mm.

Conclusions:

• Pediatric lung nodule detection using adult-trained CAD systems is less sensitive on low-dose CT scans.
• The findings highlight the need for caution when applying these CAD systems in low-dose pediatric CT protocols, especially for small nodules.
• Further research may be needed to optimize CAD algorithms for pediatric low-dose imaging.