TY - GEN
T1 - Contrastive Self-Supervised Learning for Spatio-Temporal Analysis of Lung Ultrasound Videos
AU - Chen, Li
AU - Rubin, Jonathan
AU - Ouyang, Jiahong
AU - Balaraju, Naveen
AU - Patil, Shubham
AU - Mehanian, Courosh
AU - Kulhare, Sourabh
AU - Millin, Rachel
AU - Gregory, Kenton W.
AU - Gregory, Cynthia R.
AU - Zhu, Meihua
AU - Kessler, David O.
AU - Malia, Laurie
AU - Dessie, Almaz
AU - Rabiner, Joni
AU - Coneybeare, Di
AU - Shopsin, Bo
AU - Hersh, Andrew
AU - Madar, Cristian
AU - Shupp, Jeffrey
AU - Johnson, Laura S.
AU - Avila, Jacob
AU - Dwyer, Kristin
AU - Weimersheimer, Peter
AU - Raju, Balasundar
AU - Kruecker, Jochen
AU - Chen, Alvin
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Self-supervised learning (SSL) methods have shown promise for medical imaging applications by learning meaningful visual representations, even when the amount of labeled data is limited. Here, we extend state-of-the-art contrastive learning SSL methods to 2D+time medical ultrasound video data by introducing a modified encoder and augmentation method capable of learning meaningful spatio-temporal representations, without requiring constraints on the input data. We evaluate our method on the challenging clinical task of identifying lung consolidations (an important pathological feature) in ultrasound videos. Using a multi-center dataset of over 27k lung ultrasound videos acquired from over 500 patients, we show that our method can significantly improve performance on downstream localization and classification of lung consolidation. Comparisons against baseline models trained without SSL show that the proposed methods are particularly advantageous when the size of labeled training data is limited (e.g., as little as 5% of the training set).
AB - Self-supervised learning (SSL) methods have shown promise for medical imaging applications by learning meaningful visual representations, even when the amount of labeled data is limited. Here, we extend state-of-the-art contrastive learning SSL methods to 2D+time medical ultrasound video data by introducing a modified encoder and augmentation method capable of learning meaningful spatio-temporal representations, without requiring constraints on the input data. We evaluate our method on the challenging clinical task of identifying lung consolidations (an important pathological feature) in ultrasound videos. Using a multi-center dataset of over 27k lung ultrasound videos acquired from over 500 patients, we show that our method can significantly improve performance on downstream localization and classification of lung consolidation. Comparisons against baseline models trained without SSL show that the proposed methods are particularly advantageous when the size of labeled training data is limited (e.g., as little as 5% of the training set).
KW - Self-supervised learning
KW - contrastive learning
KW - lung ultrasound
KW - spatio-temporal augmentation
UR - http://www.scopus.com/inward/record.url?scp=85172119790&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85172119790&partnerID=8YFLogxK
U2 - 10.1109/ISBI53787.2023.10230816
DO - 10.1109/ISBI53787.2023.10230816
M3 - Conference contribution
AN - SCOPUS:85172119790
T3 - Proceedings - International Symposium on Biomedical Imaging
BT - 2023 IEEE International Symposium on Biomedical Imaging, ISBI 2023
PB - IEEE Computer Society
T2 - 20th IEEE International Symposium on Biomedical Imaging, ISBI 2023
Y2 - 18 April 2023 through 21 April 2023
ER -