End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography

Diego Ardila, Atilla P. Kiraly, Sujeeth Bharadwaj, Bokyung Choi, Joshua J. Reicher, Lily Peng, Daniel Tse*, Mozziyar Etemadi, Wenxing Ye, Greg Corrado, David P. Naidich, Shravya Shetty

*Corresponding author for this work

Research output: Contribution to journalLetter

51 Citations (Scopus)

Abstract

With an estimated 160,000 deaths in 2018, lung cancer is the most common cause of cancer death in the United States1. Lung cancer screening using low-dose computed tomography has been shown to reduce mortality by 20–43% and is now included in US screening guidelines1–6. Existing challenges include inter-grader variability and high false-positive and false-negative rates7–10. We propose a deep learning algorithm that uses a patient’s current and prior computed tomography volumes to predict the risk of lung cancer. Our model achieves a state-of-the-art performance (94.4% area under the curve) on 6,716 National Lung Cancer Screening Trial cases, and performs similarly on an independent clinical validation set of 1,139 cases. We conducted two reader studies. When prior computed tomography imaging was not available, our model outperformed all six radiologists with absolute reductions of 11% in false positives and 5% in false negatives. Where prior computed tomography imaging was available, the model performance was on-par with the same radiologists. This creates an opportunity to optimize the screening process via computer assistance and automation. While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.

Original languageEnglish (US)
Pages (from-to)954-961
Number of pages8
JournalNature Medicine
Volume25
Issue number6
DOIs
StatePublished - Jun 1 2019

Fingerprint

Early Detection of Cancer
Tomography
Lung Neoplasms
Screening
Thorax
Learning
Imaging techniques
Cone-Beam Computed Tomography
Automation
Learning algorithms
Area Under Curve
Cause of Death
Deep learning
Mortality
Neoplasms
Radiologists

ASJC Scopus subject areas

  • Biochemistry, Genetics and Molecular Biology(all)

Cite this

Ardila, D., Kiraly, A. P., Bharadwaj, S., Choi, B., Reicher, J. J., Peng, L., ... Shetty, S. (2019). End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nature Medicine, 25(6), 954-961. https://doi.org/10.1038/s41591-019-0447-x
Ardila, Diego ; Kiraly, Atilla P. ; Bharadwaj, Sujeeth ; Choi, Bokyung ; Reicher, Joshua J. ; Peng, Lily ; Tse, Daniel ; Etemadi, Mozziyar ; Ye, Wenxing ; Corrado, Greg ; Naidich, David P. ; Shetty, Shravya. / End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. In: Nature Medicine. 2019 ; Vol. 25, No. 6. pp. 954-961.
@article{07b6ac6b20f949ba9dbe6b10efbb253a,
title = "End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography",
abstract = "With an estimated 160,000 deaths in 2018, lung cancer is the most common cause of cancer death in the United States1. Lung cancer screening using low-dose computed tomography has been shown to reduce mortality by 20–43{\%} and is now included in US screening guidelines1–6. Existing challenges include inter-grader variability and high false-positive and false-negative rates7–10. We propose a deep learning algorithm that uses a patient’s current and prior computed tomography volumes to predict the risk of lung cancer. Our model achieves a state-of-the-art performance (94.4{\%} area under the curve) on 6,716 National Lung Cancer Screening Trial cases, and performs similarly on an independent clinical validation set of 1,139 cases. We conducted two reader studies. When prior computed tomography imaging was not available, our model outperformed all six radiologists with absolute reductions of 11{\%} in false positives and 5{\%} in false negatives. Where prior computed tomography imaging was available, the model performance was on-par with the same radiologists. This creates an opportunity to optimize the screening process via computer assistance and automation. While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.",
author = "Diego Ardila and Kiraly, {Atilla P.} and Sujeeth Bharadwaj and Bokyung Choi and Reicher, {Joshua J.} and Lily Peng and Daniel Tse and Mozziyar Etemadi and Wenxing Ye and Greg Corrado and Naidich, {David P.} and Shravya Shetty",
year = "2019",
month = "6",
day = "1",
doi = "10.1038/s41591-019-0447-x",
language = "English (US)",
volume = "25",
pages = "954--961",
journal = "Nature Medicine",
issn = "1078-8956",
publisher = "Nature Publishing Group",
number = "6",

}

Ardila, D, Kiraly, AP, Bharadwaj, S, Choi, B, Reicher, JJ, Peng, L, Tse, D, Etemadi, M, Ye, W, Corrado, G, Naidich, DP & Shetty, S 2019, 'End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography', Nature Medicine, vol. 25, no. 6, pp. 954-961. https://doi.org/10.1038/s41591-019-0447-x

End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. / Ardila, Diego; Kiraly, Atilla P.; Bharadwaj, Sujeeth; Choi, Bokyung; Reicher, Joshua J.; Peng, Lily; Tse, Daniel; Etemadi, Mozziyar; Ye, Wenxing; Corrado, Greg; Naidich, David P.; Shetty, Shravya.

In: Nature Medicine, Vol. 25, No. 6, 01.06.2019, p. 954-961.

Research output: Contribution to journalLetter

TY - JOUR

T1 - End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography

AU - Ardila, Diego

AU - Kiraly, Atilla P.

AU - Bharadwaj, Sujeeth

AU - Choi, Bokyung

AU - Reicher, Joshua J.

AU - Peng, Lily

AU - Tse, Daniel

AU - Etemadi, Mozziyar

AU - Ye, Wenxing

AU - Corrado, Greg

AU - Naidich, David P.

AU - Shetty, Shravya

PY - 2019/6/1

Y1 - 2019/6/1

N2 - With an estimated 160,000 deaths in 2018, lung cancer is the most common cause of cancer death in the United States1. Lung cancer screening using low-dose computed tomography has been shown to reduce mortality by 20–43% and is now included in US screening guidelines1–6. Existing challenges include inter-grader variability and high false-positive and false-negative rates7–10. We propose a deep learning algorithm that uses a patient’s current and prior computed tomography volumes to predict the risk of lung cancer. Our model achieves a state-of-the-art performance (94.4% area under the curve) on 6,716 National Lung Cancer Screening Trial cases, and performs similarly on an independent clinical validation set of 1,139 cases. We conducted two reader studies. When prior computed tomography imaging was not available, our model outperformed all six radiologists with absolute reductions of 11% in false positives and 5% in false negatives. Where prior computed tomography imaging was available, the model performance was on-par with the same radiologists. This creates an opportunity to optimize the screening process via computer assistance and automation. While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.

AB - With an estimated 160,000 deaths in 2018, lung cancer is the most common cause of cancer death in the United States1. Lung cancer screening using low-dose computed tomography has been shown to reduce mortality by 20–43% and is now included in US screening guidelines1–6. Existing challenges include inter-grader variability and high false-positive and false-negative rates7–10. We propose a deep learning algorithm that uses a patient’s current and prior computed tomography volumes to predict the risk of lung cancer. Our model achieves a state-of-the-art performance (94.4% area under the curve) on 6,716 National Lung Cancer Screening Trial cases, and performs similarly on an independent clinical validation set of 1,139 cases. We conducted two reader studies. When prior computed tomography imaging was not available, our model outperformed all six radiologists with absolute reductions of 11% in false positives and 5% in false negatives. Where prior computed tomography imaging was available, the model performance was on-par with the same radiologists. This creates an opportunity to optimize the screening process via computer assistance and automation. While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.

UR - http://www.scopus.com/inward/record.url?scp=85066066623&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85066066623&partnerID=8YFLogxK

U2 - 10.1038/s41591-019-0447-x

DO - 10.1038/s41591-019-0447-x

M3 - Letter

C2 - 31110349

AN - SCOPUS:85066066623

VL - 25

SP - 954

EP - 961

JO - Nature Medicine

JF - Nature Medicine

SN - 1078-8956

IS - 6

ER -