Using AI to Improve Radiologist Performance in Detection of Abnormalities on Chest Radiographs

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Chest radiography is a common but challenging diagnostic tool.
• Interpreting chest radiographs requires expertise, and errors can occur.
• Artificial intelligence (AI) offers potential for improving diagnostic accuracy.

Purpose of the Study:

• To evaluate the impact of AI assistance on radiologist performance in detecting thoracic abnormalities.
• To assess AI's effect on sensitivity, specificity, and reading time across different radiologist expertise levels.
• To compare radiologist performance with and without AI assistance on chest radiographs.

Main Methods:

• Retrospective analysis of 500 chest radiographs from patients who also had CT scans.
• Radiographs were annotated for five key abnormalities (pneumothorax, pleural effusion, consolidation, mass, nodule) using CT as ground truth.
• 12 radiologists (residents, general, and thoracic) interpreted images with and without AI (ChestView).

Main Results:

• AI significantly increased sensitivity for all five abnormalities, with notable gains for pneumothorax (26%) and consolidation (14%).
• Specificity generally improved with AI, except for pneumothorax detection.
• Mean reading time decreased by 31% (from 81 to 56 seconds) with AI assistance.

Conclusions:

• AI-assisted interpretation enhances sensitivity in detecting thoracic abnormalities on chest radiographs for radiologists at all experience levels.
• AI tools can reduce interpretation time, improving workflow efficiency.
• While beneficial overall, AI's impact on specificity varies by abnormality, with a slight decrease noted for pneumothorax.