Higher agreement between readers with deep learning CAD software for reporting pulmonary nodules on CT

Area of Science:

• Radiology
• Medical Imaging
• Pulmonology

Background:

• Incidental pulmonary nodules detected on CT scans require careful management.
• Radiologist interpretation variability can affect nodule management recommendations.
• Computer-aided detection (CAD) systems offer potential solutions for improving diagnostic consistency.

Purpose of the Study:

• To evaluate the impact of CAD software on radiologist recommendations for pulmonary nodule management.
• To assess changes in reading times and interobserver agreement with CAD assistance.

Main Methods:

• Retrospective analysis of 50 chest CT cases with incidentally detected pulmonary nodules.
• Two radiologists independently assessed cases twice: unaided and aided by CAD software.
• Management recommendations graded using BTS guidelines; reading times recorded.
• Interobserver agreement assessed using linear weighted kappa.

Main Results:

• Significant reduction in reading times for both radiologists when using CAD (33.4% and 42.6%).
• Interobserver agreement improved substantially with CAD assistance (kappa from 0.61 to 0.84).
• Mean reading times per case decreased notably with CAD implementation.

Conclusions:

• CAD software can enhance the uniformity of pulmonary nodule management recommendations.
• The use of CAD systems may lead to more consistent clinical practice in nodule reporting.
• CAD aids in efficient and reliable assessment of incidentally detected pulmonary nodules.