Assistive AI in Lung Cancer Screening: A Retrospective Multinational Study in the United States and Japan

Atilla P Kiraly ¹, Corbin A Cunningham ¹, Ryan Najafi ¹

⁰From Google Health Research, 1600 Amphitheatre Pkwy, Mountain View, CA 94043 (A.P.K., C.A.C., R.N., Z.N., C.L., J.R.L., D.A., S.M.M., R.P., Y.L., S.J., G.S.C., L.P., D.T., S.S., S.P., K.E.); Waymo, Mountain View, Calif (J.Y., N.B.), David Geffen School of Medicine at UCLA, Los Angeles, Calif (C.L.); Google, Mountain View, Calif (W.Y.); Department of Gastroenterology, Sendai Kousei Hospital, Sendai, Japan (H.S.); MNES Inc, Hiroshima, Japan (Y.S.); Department of Telemedicine, Northwestern University Feinberg School of Medicine, Chicago, Ill (M.E., D.M.); and Center for Biological Imaging, New York University-Langone Medical Center, New York, NY (D.P.N.).

Radiology. Artificial Intelligence +

|

March 13, 2024

View abstract on PubMed

Summary

An artificial intelligence (AI) assistant improved lung cancer screening specificity in multinational studies. This AI tool enhanced radiologists' ability to identify actionable findings, warranting further international investigation.

Area Of Science

Radiology
Artificial Intelligence
Oncology

Background

Lung cancer screening is crucial for early detection.
Multinational studies are needed to validate AI tools in diverse clinical settings.

Purpose Of The Study

To evaluate the impact of an AI assistant on lung cancer screening workflows.
To assess AI's effect on radiologists' performance in U.S. and Japan-based studies.

Main Methods

Two retrospective randomized multireader multicase studies involving 627 low-dose chest CT cases.
62 radiologists interpreted cases with and without AI assistance, totaling 7524 interpretations.
Evaluation metrics included AUC, sensitivity, and specificity for recall recommendations.

Main Results

AI assistance increased radiologists' AUC by 0.023 in both U.S. and Japan studies.
Specificity for actionable findings improved by 5.5% (U.S.) and 6.7% (Japan) with AI.
No significant difference in sensitivity was observed with AI assistance.

Conclusions

The AI interface enhanced lung cancer screening specificity in U.S. and Japan-based reader studies.
Further evaluation in diverse international screening environments is recommended.
AI shows promise in improving the accuracy of lung cancer detection.

Keywords:

Assistive Artificial Intelligence CT Lung Cancer Screening