Synthetic and augmented data
Citation, DOI, disclosures and article data
At the time the article was created Candace Makeda Moore had no recorded disclosures.View Candace Makeda Moore's current disclosures
At the time the article was last revised Dimitrios Toumpanakis had no recorded disclosures.View Dimitrios Toumpanakis's current disclosures
In the context of radiological images, synthetic and augmented data are data that are not completely generated by direct measurement from patients.
Machine learning models improve with increased data. However, there is a relative lack of open, free available radiology data sets. Issues of patient privacy and legal restrictions on data use make machine learning without synthetic and augmented data challenging. Additionally, some diseases are so rare that even large data sets do not contain sufficient samples to generate robust machine learning algorithms.
In addition to contributing to making algorithms for image identification and classification, synthetic data can also be used in algorithms for artifact correction.
- synthetic data: partly or completely artificial. Synthetic data are often produced by generative adversarial networks
- augmented data: derived from real images with some sort of minor and realistic transformation (such as translation, flipping, rotation, or the addition of noise) in order to increase the diversity of the training set
Criticisms of the use of augmented, and especially synthetic, data in medical AI include the potential amplification of statistical biases and a general lack of research on their consequences. Nonetheless, data augmentation is used routinely in the development of AI applications for radiology.
- 1. Kanghyun Ryu, Yoonho Nam, Sung‐Min Gho, Jinhee Jang, Ho‐Joon Lee, Jihoon Cha, Hye Jin Baek, Jiyong Park, Dong‐Hyun Kim. Data‐driven synthetic MRI FLAIR artifact correction via deep neural network. (2019) Journal of Magnetic Resonance Imaging. doi:10.1002/jmri.26712 - Pubmed
- 2. Goodfellow I, Pouget-Abadie J, Mirza M, et al (2014) Generative Adversarial Nets. In: 1124Ghahramani Z, Welling M, Cortes C, et al (eds) Advances in Neural Information Processing 1125Systems 27. Curran Associates, Inc., pp 2672–2680.