Deep learning improves physician accuracy in the comprehensive detection of abnormalities on chest X-rays

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Chest X-rays are the most frequent medical imaging procedure.
• Physician misinterpretation of chest X-rays is a common issue.
• AI offers potential solutions for improving diagnostic accuracy.

Purpose of the Study:

• To present an FDA-cleared AI system for chest X-ray abnormality detection.
• To evaluate the AI system's performance and its impact on physician accuracy.
• To assess the AI system's generalizability and efficiency.

Main Methods:

• Developed and validated an AI system using a deep learning algorithm on a large dataset.
• Tested AI generalizability on publicly available chest X-ray data.
• Compared physician accuracy (aided vs. unaided) and evaluation speed.

Main Results:

• The AI system demonstrated high accuracy in detecting abnormalities (AUC: 0.976) and generalized well (AUC: 0.975).
• Physicians showed significant improvement in detection accuracy when aided by AI (AUC difference: 0.101, p < 0.001).
• Non-radiologists achieved radiologist-level accuracy with AI assistance and evaluated X-rays faster.

Conclusions:

• The AI system is accurate and effective in reducing physician errors in chest X-ray interpretation.
• AI has the potential to enhance access to rapid, high-quality radiograph interpretation.
• AI-assisted interpretation can improve diagnostic outcomes and efficiency in medical imaging.