Population

500 chest X-rays with CT scanner performed within 72 hours from Hôpital Cochin (AP-HP)

Design

• Gold standard: CT-based annotation by chest radiologist

• Readers: 4 chest radiologists, 4 general radiologists, 4 radiology residents

• MRMC study design: reading with and without AI

Highlights

• AI-assisted chest radiography interpretation resulted in an increased sensitivity of 5.9 to 26.2 % (P<.001) for all readers including thoracic radiologists, general radiologists, and radiology residents.

• General radiologists and radiology residents assisted by AI achieved the performance of chest radiologists without AI

• Mean reading time was 81s without AI vs 56s with AI (-31%, P<.001), with a 17% reduction for radiographs with abnormalities vs 38% for no abnormalities.

Conclusion

AI assistance can improve the detection accuracy of thoracic abnormalities on chest radiographs across radiologists of varying expertise, leading to marked improvements in sensitivity and a reduction in interpretation time.