TY - GEN
T1 - Surrogate supervision for medical image analysis
T2 - 16th IEEE International Symposium on Biomedical Imaging, ISBI 2019
AU - Tajbakhsh, Nima
AU - Hu, Yufei
AU - Cao, Junli
AU - Yan, Xingjian
AU - Xiao, Yi
AU - Lu, Yong
AU - Liang, Jianming
AU - Terzopoulos, Demetri
AU - Ding, Xiaowei
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/4
Y1 - 2019/4
N2 - We investigate the effectiveness of a simple solution to the common problem of deep learning in medical image analysis with limited quantities of labeled training data. The underlying idea is to assign artificial labels to abundantly available unlabeled medical images and, through a process known as surrogate supervision, pre-train a deep neural network model for the target medical image analysis task lacking sufficient labeled training data. In particular, we employ 3 surrogate supervision schemes, namely rotation, reconstruction, and colorization, in 4 different medical imaging applications representing classification and segmentation for both 2D and 3D medical images. 3 key findings emerge from our research: 1) pre-training with surrogate supervision is effective for small training sets; 2) deep models trained from initial weights pre-trained through surrogate supervision outperform the same models when trained from scratch, suggesting that pre-training with surrogate supervision should be considered prior to training any deep 3D models; 3) pre-training models in the medical domain with surrogate supervision is more effective than transfer learning from an unrelated domain (e.g., natural images), indicating the practical value of abundant unlabeled medical image data.
AB - We investigate the effectiveness of a simple solution to the common problem of deep learning in medical image analysis with limited quantities of labeled training data. The underlying idea is to assign artificial labels to abundantly available unlabeled medical images and, through a process known as surrogate supervision, pre-train a deep neural network model for the target medical image analysis task lacking sufficient labeled training data. In particular, we employ 3 surrogate supervision schemes, namely rotation, reconstruction, and colorization, in 4 different medical imaging applications representing classification and segmentation for both 2D and 3D medical images. 3 key findings emerge from our research: 1) pre-training with surrogate supervision is effective for small training sets; 2) deep models trained from initial weights pre-trained through surrogate supervision outperform the same models when trained from scratch, suggesting that pre-training with surrogate supervision should be considered prior to training any deep 3D models; 3) pre-training models in the medical domain with surrogate supervision is more effective than transfer learning from an unrelated domain (e.g., natural images), indicating the practical value of abundant unlabeled medical image data.
KW - Limited training data
KW - Medical imaging
KW - Surrogate supervision
KW - Unlabeled data
UR - http://www.scopus.com/inward/record.url?scp=85073887988&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85073887988&partnerID=8YFLogxK
U2 - 10.1109/ISBI.2019.8759553
DO - 10.1109/ISBI.2019.8759553
M3 - Conference contribution
AN - SCOPUS:85073887988
T3 - Proceedings - International Symposium on Biomedical Imaging
SP - 1251
EP - 1255
BT - ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging
PB - IEEE Computer Society
Y2 - 8 April 2019 through 11 April 2019
ER -