Artificial intelligence system for identification of false-negative interpretations in chest radiographs

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Chest radiographs are crucial for diagnosing various conditions.
• Radiologists interpret these images, but false negatives can occur.
• Artificial intelligence (AI) offers potential for improving diagnostic accuracy.

Purpose of the Study:

• To evaluate an AI system's efficacy in detecting false negatives in normal chest radiographs.
• To assess the impact of AI as an advisor to radiologists in re-interpreting these images.

Main Methods:

• Retrospective analysis of 4208 chest radiographs initially interpreted as normal.
• Application of a commercial AI system to identify potential abnormalities.
• Re-interpretation by radiologists with AI feedback in an "AI as advisor" scenario.
• Evaluation using detection yield and false referral rate (FRR), with decision curve analysis.

Main Results:

• The AI system detected initially overlooked abnormalities with a 2.4% detection yield and 14.0% FRR.
• In the "AI as advisor" scenario, radiologists improved detection with a 1.2% detection yield and 0.97% FRR.
• Decision curve analysis showed a positive net benefit for the AI advisor scenario.

Conclusions:

• AI systems can identify relevant abnormalities missed in initial radiologist interpretations of chest radiographs.
• AI feedback can enable radiologists to correct false-negative findings, improving diagnostic outcomes.
• The AI system shows promise in reducing diagnostic errors in chest radiography.