UltraAugment: Fan-shape and Artifact-based Data Augmentation for 2D Ultrasound Images

Deep learning systems for medical image analysis have shown remarkable performance. However, performance is heavily dependent on the size and diversity of the training data as small datasets might lead to overfitting. Unfortunately, labeled data is often hard to acquire because of the high cost and required medical expertise. Data augmentation is an effective strategy to combat this and has proven to significantly improve model generalisability as it increases the size and diversity of the dataset. However for ultrasound images classic data transformations may not always be appropriate. In this paper we focus on developing data augmentations specifically designed for fan-shaped ultrasound images by simulating artifacts, altering speckle patterns, and adapting conventional techniques to make them fanshape preserving. We apply the suggested augmentations to two segmentation tasks and demonstrate that the proposed augmentation techniques can improve performance and can remedy the harm caused by there conventional alternatives.