Deep Learning for Chest X-ray Diagnosis: Competition Between Radiologists with or Without Artificial Intelligence...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Chest radiography is a common diagnostic tool.
• Accurate interpretation of chest X-rays is crucial for patient care.
• Radiologists face challenges in efficiently and accurately diagnosing a wide range of abnormalities.

Purpose of the Study:

• To evaluate the impact of a deep learning algorithm on radiologist performance in chest radiograph diagnosis.
• To assess improvements in diagnostic accuracy and efficiency with AI assistance.
• To compare radiologist performance with and without AI support.

Main Methods:

• A deep learning algorithm was developed to detect normal findings and 13 abnormalities in chest X-rays.
• 111 radiologists (junior, intermediate, senior) interpreted 100 radiographs in two groups: control (no AI) and test (with AI).
• Performance metrics included accuracy, false-positive/negative rates, analysis time, and Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• Radiologists with AI assistance achieved significantly higher average scores (619 vs. 597, P < 0.001) and spent less time (1926s vs. 3279s, P < 0.001).
• AI assistance improved performance (higher AUCs) in recognizing normal findings, pulmonary fibrosis, heart enlargement, mass, pleural effusion, and consolidation.
• Radiologists without AI showed better performance in identifying aortic calcification, calcification, cavity, nodule, pleural thickening, and rib fracture.

Conclusions:

• Deep learning methods positively impact radiologist performance in interpreting chest X-rays.
• AI assistance enhances both the efficacy and efficiency of radiologists in diagnostic tasks.
• The study validates the beneficial role of AI in improving chest radiograph interpretation.