Revisiting Data Augmentation for Ultrasound Images

TL;DR

This paper evaluates the effectiveness of various data augmentation techniques on ultrasound image analysis tasks, introducing a new benchmark and demonstrating that many natural image augmentations are also beneficial for ultrasound imaging.

Contribution

It provides a comprehensive benchmark for ultrasound image augmentation techniques and shows that common natural image augmentations are effective for ultrasound tasks.

Findings

Many natural image augmentations improve ultrasound model performance.

Diverse augmentation with TrivialAugment is effective for ultrasound images.

Proposed methodology offers a structured approach for assessing augmentations.

Abstract

Data augmentation is a widely used and effective technique to improve the generalization performance of deep neural networks. Yet, despite often facing limited data availability when working with medical images, it is frequently underutilized. This appears to come from a gap in our collective understanding of the efficacy of different augmentation techniques across different tasks and modalities. One modality where this is especially true is ultrasound imaging. This work addresses this gap by analyzing the effectiveness of different augmentation techniques at improving model performance across a wide range of ultrasound image analysis tasks. To achieve this, we introduce a new standardized benchmark of 14 ultrasound image classification and semantic segmentation tasks from 10 different sources and covering 11 body regions. Our results demonstrate that many of the augmentations commonly used for tasks on natural images are also effective on ultrasound images, even more so than augmentations developed specifically for ultrasound images in some cases. We also show that diverse augmentation using TrivialAugment, which is widely used for natural images, is also effective for ultrasound images. Moreover, our proposed methodology represents a structured approach for assessing various data augmentations that can be applied to other contexts and modalities.