TY - JOUR
T1 - Simplified Transfer Learning for Chest Radiography Models Using Less Data
AU - Sellergren, Andrew B.
AU - Chen, Christina
AU - Nabulsi, Zaid
AU - Li, Yuanzhen
AU - Maschinot, Aaron
AU - Sarna, Aaron
AU - Huang, Jenny
AU - Lau, Charles
AU - Kalidindi, Sreenivasa Raju
AU - Etemadi, Mozziyar
AU - Garcia-Vicente, Florencia
AU - Melnick, David
AU - Liu, Yun
AU - Eswaran, Krish
AU - Tse, Daniel
AU - Beladia, Neeral
AU - Krishnan, Dilip
AU - Shetty, Shravya
N1 - Funding Information:
Supported by Google. The authors thank the members of the Google Health Radiology and labeling software teams for software infrastructure support, logistical support, and assistance in data labeling. For the ChestX-ray14 data set, we thank the NIH Clinical Center for making it publicly available. Sincere appreciation also goes to the radiologists who enabled this work with their image interpretation and annotation efforts throughout the study, Jonny Wong, BA, for coordinating the imaging annotation work, and Akinori Mitani, MD, and Craig H. Mermel, MD, PhD, for providing feedback on the manuscript.
Publisher Copyright:
© RSNA, 2022.
PY - 2022/11
Y1 - 2022/11
N2 - Background: Developing deep learning models for radiology requires large data sets and substantial computational resources. Data set size limitations can be further exacerbated by distribution shifts, such as rapid changes in patient populations and standard of care during the COVID-19 pandemic. A common partial mitigation is transfer learning by pretraining a “generic network” on a large nonmedical data set and then fine-tuning on a task-specific radiology data set. Purpose: To reduce data set size requirements for chest radiography deep learning models by using an advanced machine learning approach (supervised contrastive [SupCon] learning) to generate chest radiography networks. Materials and Methods: SupCon helped generate chest radiography networks from 821 544 chest radiographs from India and the United States. The chest radiography networks were used as a starting point for further machine learning model development for 10 prediction tasks (eg, airspace opacity, fracture, tuberculosis, and COVID-19 outcomes) by using five data sets comprising 684 955 chest radiographs from India, the United States, and China. Three model development setups were tested (linear classifier, nonlinear classifier, and fine-tuning the full network) with different data set sizes from eight to 85. Results: Across a majority of tasks, compared with transfer learning from a nonmedical data set, SupCon reduced label requirements up to 688-fold and improved the area under the receiver operating characteristic curve (AUC) at matching data set sizes. At the extreme low-data regimen, training small nonlinear models by using only 45 chest radiographs yielded an AUC of 0.95 (noninferior to radiologist performance) in classifying microbiology-confirmed tuberculosis in external validation. At a more moderate data regimen, training small nonlinear models by using only 528 chest radiographs yielded an AUC of 0.75 in predicting severe COVID-19 outcomes. Conclusion: Supervised contrastive learning enabled performance comparable to state-of-the-art deep learning models in multiple clinical tasks by using as few as 45 images and is a promising method for predictive modeling with use of small data sets and for predicting outcomes in shifting patient populations.
AB - Background: Developing deep learning models for radiology requires large data sets and substantial computational resources. Data set size limitations can be further exacerbated by distribution shifts, such as rapid changes in patient populations and standard of care during the COVID-19 pandemic. A common partial mitigation is transfer learning by pretraining a “generic network” on a large nonmedical data set and then fine-tuning on a task-specific radiology data set. Purpose: To reduce data set size requirements for chest radiography deep learning models by using an advanced machine learning approach (supervised contrastive [SupCon] learning) to generate chest radiography networks. Materials and Methods: SupCon helped generate chest radiography networks from 821 544 chest radiographs from India and the United States. The chest radiography networks were used as a starting point for further machine learning model development for 10 prediction tasks (eg, airspace opacity, fracture, tuberculosis, and COVID-19 outcomes) by using five data sets comprising 684 955 chest radiographs from India, the United States, and China. Three model development setups were tested (linear classifier, nonlinear classifier, and fine-tuning the full network) with different data set sizes from eight to 85. Results: Across a majority of tasks, compared with transfer learning from a nonmedical data set, SupCon reduced label requirements up to 688-fold and improved the area under the receiver operating characteristic curve (AUC) at matching data set sizes. At the extreme low-data regimen, training small nonlinear models by using only 45 chest radiographs yielded an AUC of 0.95 (noninferior to radiologist performance) in classifying microbiology-confirmed tuberculosis in external validation. At a more moderate data regimen, training small nonlinear models by using only 528 chest radiographs yielded an AUC of 0.75 in predicting severe COVID-19 outcomes. Conclusion: Supervised contrastive learning enabled performance comparable to state-of-the-art deep learning models in multiple clinical tasks by using as few as 45 images and is a promising method for predictive modeling with use of small data sets and for predicting outcomes in shifting patient populations.
UR - http://www.scopus.com/inward/record.url?scp=85138217311&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85138217311&partnerID=8YFLogxK
U2 - 10.1148/radiol.212482
DO - 10.1148/radiol.212482
M3 - Article
C2 - 35852426
AN - SCOPUS:85138217311
SN - 0033-8419
VL - 305
SP - 454
EP - 465
JO - Radiology
JF - Radiology
IS - 2
ER -