TY - JOUR
T1 - Reducing Contextual Bias in Cardiac Magnetic Resonance Imaging Deep Learning Using Contrastive Self-Supervision
AU - Nakashima, Makiya
AU - Salem, Donna
AU - Wilson Tang, H. W.
AU - Nguyen, Christopher
AU - Hwang, Tae Hyun
AU - Zhao, Ding
AU - Kim, Byung Hak
AU - Kwon, Deborah
AU - Chen, David
N1 - Publisher Copyright:
© 2023 M. Nakashima, D. Salem, H.W. Tang, C. Nguyen, T.H. Hwang, D. Zhao, B.-H. Kim, D. Kwon & D. Chen.
PY - 2023
Y1 - 2023
N2 - Applying deep learning to medical imaging tasks is not straightforward due to the variable quality and relatively low volume of healthcare data. There is often considerable risk that deep learning models may use contextual cues instead of physiologically relevant features to achieve the clinical task. Although these cues can provide shortcuts to high performance within a carefully crafted training set, they often lead to poor performance in real-world applications. Contrastive self-supervision (CSS) has recently been shown to boost performance of deep learning on downstream applications in several medical imaging tasks. However, it is unclear how much of these pre-trained representations are impacted by contextual cues, both known and unknown. In this work, we evaluate how CSS pre-training can produce not only more accurate but also more trustworthy and generalizable models for clinical imaging applications. Specifically, we evaluate the saliency and accuracy of deep learning models using CSS in contrast to end-to-end supervised training and conventional transfer learning from natural image datasets using an institutional specific and public cardiomyopathy cohorts. We find that CSS pre-training models not only improve downstream diagnostic performance in each cohort, but more importantly, also produced models with higher saliency in cardiac anatomy. Our code is available at https://github.com/makiya11/ssl_spur_cmr.
AB - Applying deep learning to medical imaging tasks is not straightforward due to the variable quality and relatively low volume of healthcare data. There is often considerable risk that deep learning models may use contextual cues instead of physiologically relevant features to achieve the clinical task. Although these cues can provide shortcuts to high performance within a carefully crafted training set, they often lead to poor performance in real-world applications. Contrastive self-supervision (CSS) has recently been shown to boost performance of deep learning on downstream applications in several medical imaging tasks. However, it is unclear how much of these pre-trained representations are impacted by contextual cues, both known and unknown. In this work, we evaluate how CSS pre-training can produce not only more accurate but also more trustworthy and generalizable models for clinical imaging applications. Specifically, we evaluate the saliency and accuracy of deep learning models using CSS in contrast to end-to-end supervised training and conventional transfer learning from natural image datasets using an institutional specific and public cardiomyopathy cohorts. We find that CSS pre-training models not only improve downstream diagnostic performance in each cohort, but more importantly, also produced models with higher saliency in cardiac anatomy. Our code is available at https://github.com/makiya11/ssl_spur_cmr.
UR - http://www.scopus.com/inward/record.url?scp=85184280097&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85184280097&partnerID=8YFLogxK
M3 - Conference article
AN - SCOPUS:85184280097
SN - 2640-3498
VL - 219
SP - 473
EP - 488
JO - Proceedings of Machine Learning Research
JF - Proceedings of Machine Learning Research
T2 - 8th Machine Learning for Healthcare Conference, MLHC 2023
Y2 - 11 August 2023 through 12 August 2023
ER -