Data Augmentation Using Learned Transformations for One-Shot Medical Image Segmentation

Abstract

Image segmentation is an important task in many medical applications. Methods based on convolutional neural networks attain state-of-the-art accuracy; however, they typically rely on supervised training with large labeled datasets. Labeling medical images requires significant expertise and time, and typical hand-tuned approaches for data augmentation fail to capture the complex variations in such images.We present an automated data augmentation method for synthesizing labeled medical images. We demonstrate our method on the task of segmenting magnetic resonance imaging (MRI) brain scans. Our method requires only a single segmented scan, and leverages other unlabeled scans in a semi-supervised approach. We learn a model of transformations from the images, and use the model along with the labeled example to synthesize additional labeled examples. Each transformation is comprised of a spatial deformation field and an intensity change, enabling the synthesis of complex effects such as variations in anatomy and image acquisition procedures. We show that training a supervised segmenter with these new examples provides significant improvements over state-of-the-art methods for oneshot biomedical image segmentation.

Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation

“…The authors demonstrate that the lung segmentation model trained using images with synthetic pleural nodules is superior to the model trained using only real images wherein the pleural nodules are under-represented. Zhao et al (2019) propose a data synthesis method to generate pairs of brain MR images and the segmentation masks from only one labeled MR image. For the task of brain structures segmentation, the suggested data augmentation method, which is further discussed in Section 4.4.3, enables four points increase in Dice over a model trained using Conditional GAN is used to synthesize X-ray images with desired abnormalities tradition data augmentation and 3 points increase in Dice over atlas-based data augmentation.…”

“…Unsupervised task Bai et al (2017) Embedding consistency Zhang et al (2017b) Image classification Sedai et al (2017) Image reconstruction Baur et al (2017) Manifold learning Chartsias et al (2018) Image reconstruction Huo et al (2018a) Image synthesis Zhao et al (2019) Image registration Li et al (2019) Transformation consistency the same-class pixels as close as possible while pushing apart the feature embedding of the pixels from different classes. To identify same-class pixels between labeled and unlabeled images, the authors assume the availability of a noisy label prior for unlabeled images.…”

“…The authors demonstrate that, given small quantities of training data, this semi-supervised learning framework is effective for skin lesion segmentation from dermoscopy images, optic disc segmentation from fundus images, and liver segmentation from volumetric CT scans. Zhao et al (2019) propose a semi-supervised learning framework for 1-shot medical image segmentation where only one example in the training set has the segmentation mask and the rest of images are unlabeled. For this purpose, the authors suggest a hybrid spatial-intensity transformation model.…”

Multi-domain Adaptation in Brain MRI Through Paired Consistency and Adversarial Learning

“…In the context of medical imaging, the latter includes physics-based image augmentation, synthetic bias field addition or registration-based approaches [14]. These methods lean on domain-specific knowledge to generate plausible transformations.…”

Neural Style Transfer Improves 3D Cardiovascular MR Image Segmentation on Inconsistent Data

“…This dataset shift phenomena [7] significantly hampered the generalization of deep neural network models. Zhao et al [12] proposed using learned transforms to generate samples used in data augmentation aiming at one-shot segmentation. In our preliminary experiments, we found that augmenting the training set with images generated from the lowquality domain contributed little to the overall performance.…”