Ziegler, Jonathan D. (2023) The Digital Patient - How synthetic data can advance medical research. Medium.

Abstract

Continuous patient data collection through research and clinical practice in healthcare and drug development has the potential to greatly improve our understanding of disease and treatment. However, patient data are subject to many constraints that can limit how this potential may be realized:

• Consent and data privacy laws limit the purpose for which data may be used or even just moved or processed.
• Data collection is not implemented in the same way across data sources: For example, images may have different resolution or quality across measurements, or patients may have different visit schedules.
• Availability of different data modalities, such as particular measurements or medical images, may vary across patients and datasets.

In this article, we’ll take a quick look at the foundations of Generative Adversarial Networks and then examine some advances we made in creating multi-modal synthesizers which can help us address some of the challenges outlined above. In particular, we will look at creating three-dimensional, high-resolution images and clinical data in one pass, ensuring cross-modal correlation. A big aspect of this project so far was enabling the capability of conditioned synthesis. We’ll look at some example results and talk about potential applications.

Item Type:	Article
Keywords:	Synthetic Data, Image Synthesis, Deep Learning, Generative Adversarial Networks, Conditional Synthesis, Multimodal Data
Date Deposited:	18 Apr 2023 00:46
Last Modified:	18 Apr 2023 00:46
URI:	https://oak.novartis.com/id/eprint/49581