Association of Artificial Intelligence-Aided Chest Radiograph Interpretation With Reader Performance and Efficiency

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate interpretation of radiologic images is crucial for patient care.
• Deep learning-based artificial intelligence (AI) offers potential to augment radiologist performance.

Purpose of the Study:

• To evaluate if an AI engine can improve radiologist performance and efficiency in interpreting chest radiograph abnormalities.
• To assess the impact of AI on sensitivity, specificity, and interpretation time.

Main Methods:

• A multicenter cohort study involving 6 radiologists (attending, fellow, resident) interpreting 497 chest radiographs.
• Radiologists performed two sessions: one with AI assistance (heatmap and probability) and one without, in a randomized crossover design.
• Ground truths were established by consensus reading; sensitivity, specificity, and AUROC were calculated.

Main Results:

• AI-aided interpretation significantly improved reader sensitivity for all four target findings (nodule, pneumonia, pleural effusion, pneumothorax).
• No negative impact on specificity was observed with AI assistance.
• Overall AUROCs improved, particularly for pneumothorax and nodule detection.
• Interpretation time decreased by 10% with AI assistance.

Conclusions:

• AI-aided interpretation enhances radiologist performance and efficiency in identifying major thoracic findings on chest radiographs.
• AI tools show promise in improving diagnostic accuracy and workflow in medical imaging.