Synthesis of Annotated Data for Medical Image Segmentation

TY - CHAP

T1 - Synthesis of Annotated Data for Medical Image Segmentation

AU - Fernandez, Virginia

AU - Borges, Pedro

AU - Graham, Mark

AU - Lopez Pinaya, Walter Hugo

AU - Vercauteren, Tom

AU - Cardoso, Jorge

N1 - Publisher Copyright: © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

PY - 2025/1/1

Y1 - 2025/1/1

N2 - In the past decade, the advances in deep learning technologies have enabled their application to medical image segmentation, showing great potential. Nonetheless, the scarcity of available labelled data can result in a lack of model generalisability. This is especially true for supervised methods requiring annotated data. Data augmentation can be used to partially alleviate data scarcity when training deep learning models. In particular, the use of deep learning-based generative modelling, which allows for the sampling of synthetic data from the modelled data distribution, has shown its potential for data augmentation in the past years. In this work, we address the topic of generative modelling to generate images and annotations, going over brainSPADE, a 2D and 3D generative model of healthy and pathological segmentations and corresponding multi-modal images for brain MRI, and how the synthetic data it produces can be applied to a range of segmentation tasks to mitigate the effects of data scarcity or domain shift.

AB - In the past decade, the advances in deep learning technologies have enabled their application to medical image segmentation, showing great potential. Nonetheless, the scarcity of available labelled data can result in a lack of model generalisability. This is especially true for supervised methods requiring annotated data. Data augmentation can be used to partially alleviate data scarcity when training deep learning models. In particular, the use of deep learning-based generative modelling, which allows for the sampling of synthetic data from the modelled data distribution, has shown its potential for data augmentation in the past years. In this work, we address the topic of generative modelling to generate images and annotations, going over brainSPADE, a 2D and 3D generative model of healthy and pathological segmentations and corresponding multi-modal images for brain MRI, and how the synthetic data it produces can be applied to a range of segmentation tasks to mitigate the effects of data scarcity or domain shift.

UR - https://www.scopus.com/pages/publications/105005695218

U2 - 10.1007/9783031809651_1

DO - 10.1007/9783031809651_1

M3 - Chapter

AN - SCOPUS:105005695218

SN - 9783031809644

SP - 3

EP - 24

BT - Generative Machine Learning Models in Medical Image Computing

PB - Springer Nature

ER -