Chartsias et al, MICCAI 2021.
Analysis of cardiac ultrasound images is commonly performed in routine clinical practice for quantification of cardiac function. Its increasing automation frequently employs deep learning networks that are trained to predict disease or detect image features. However, such models are extremely data-hungry and training requires labelling of many thousands of images by experienced clinicians.
Ultromics proposes the use of contrastive learning to mitigate the labelling bottleneck, and trained view classification models for imbalanced cardiac ultrasound datasets and show improved performance for views/classes for which minimal labelled data is available.
Compared to a naive baseline model, Ultromics achieved an improvement in F1 score of up to 30% in those views while maintaining state-of-the-art performance for the views with sufficiently many labelled training observations.
See full publication: https://arxiv.org/pdf/2108.03124.pdf