While domain-specific data augmentation can be useful in training neural networks for medical imaging tasks, such techniques have not been widely used to date. Our objective was to test whether domain-specific data augmentation is useful for medical imaging using a well-benchmarked task: view classification on fetal ultrasound FETAL-125 and OB-125 datasets. We found that using a context-preserving cut-paste strategy, we could create valid training data as measured by performance of the resulting trained model on the benchmark test dataset. When used in an online fashion, models trained on this hybrid data performed similarly to those trained using traditional data augmentation (FETAL-125 F-score 85.33 ± 0.24 vs 86.89 ± 0.60, p-value 0.014; OB-125 F-score 74.60 ± 0.11 vs 72.43 ± 0.62, p-value 0.004). Furthermore, the ability to perform augmentations during training time, as well as the ability to apply chosen augmentations equally across data classes, are important considerations in designing a bespoke data augmentation. Finally, we provide open-source code to facilitate running bespoke data augmentations in an online fashion. Taken together, this work expands the ability to design and apply domain-guided data augmentations for medical imaging tasks.
While prenatal congenital heart disease (CHD) screening has improved, accuracy remains as low as 30 percent. Standard fetal biometrics—cardiac axis (CA), cardiothoracic ratio (CTR), RV fractional area change (FAC), LV FAC, RA:LA area ratio, RV:LV area ratio—are available from screening imaging and can each aid in CHD screening, but can be cumbersome to measure. Combinations of biometrics may offer further utility but are challenging to integrate at the point of care. We tested whether using these biometrics in combination has utility in CHD screening (normal vs. abnormal). Further, we tested whether automatically predicted biometrics could function similarly to manually-labeled biometrics for this purpose. We included 105 fetal echocardiograms (20 normal, 85 abnormal across 12 different CHD lesions). We manually calculated the six biometrics above, performed dimensionality reduction using principal component analysis, and then clustered the resulting data by K-means. A previously developed deep learning model (Arnaout et al Nature 2021) was also used to automatically predict biometrics for normal, tetralogy of Fallot, and hypoplastic left heart syndrome hearts and plotted on the above cluster map. Optimal number of clusters was four, with RV:LV ratio and CTR as the most important features distinguishing clusters. Cluster 1 was predominantly normal hearts with cluster 2-4 largely abnormal hearts (Figure 1). The sensitivity and specificity for predicting abnormal hearts (e.g. CHD) was 86% and 75%, respectively. Model-predicted biometrics landed in the same clusters as the manually labeled lesions (Figure 1). To our knowledge, this is the first use of clustering to provide visualization of multiple fetal cardiac biometrics at once and reveal diagnostic utility. Once tested in screening ultrasounds on a larger scale, clustering of automated biometrics may be clinically useful at the screening point of care to augment scalable population-based screening.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.