Back to search

NAERINGSPH-Nærings-phd

Synthetic data for clinical validation of Artificial Intelligence powered tools in healthcare

Alternative title: Syntetiske data for klinisk validering av kunstig intelligens-drevne verktøy i helsesektoren

Awarded: NOK 1.9 mill.

Project Number:

333913

Application Type:

Project Period:

2022 - 2026

Funding received from:

Organisation:

AI powered tools are entering the healthcare domain to help us make better use of the data produced. Data driven decision support requires large datasets that are diverse and representative for training algorithms and for verifying that the tools perform as expected. As real health data is difficult and time consuming to capture and gain access to, the healthcare community is looking to synthetically generated data as a potential way forward. Synthetic data can be created through simulations or from real data. Augmented data is when new synthetic data is generated based on existing data, to create a larger dataset or complete a dataset with missing parameters. Synthetic data has proved to be useful for training and testing algorithms and for verifying performance in a laboratory setting. Many see it as a practical way to protect patient privacy, yet some are questioning whether privacy is sufficiently protected. Could synthetically augmented datasets also provide value in informing regulatory decision-making for authorisations and monitoring of products, as an alternative or addition to clinical validation? In this proposed PhD research project, the main objective is to examine how synthetic data could be used for validation of AI powered tools prior to clinical implementation and regulatory decision-making. The project will investigate the potential use of synthetic data for validation of AI tools, focusing on a series of questions: • What are the main risks and benefits that must be considered for assurance of AI tools for healthcare? • What are the legal implications of using augmented synthetic data? • How could synthetic data be used for clinical validation without increasing risk of patient harm? Three projects have been defined to investigate answers to these questions: “Assurance of AI adoption in healthcare”, “Legal implications of using augmented synthetic data” and “The role of synthetic data for trustworthy AI tools”.

The vision of the national lighthouse project BigMed (2017-2021) was to address barriers to the clinical implementation of Precision medicine and to lay the foundation of Big Data analytics in the clinic. The project identified access to health data as the major barrier to clinical implementation. Among the needs identified were ways to ensures trust in new technology like AI powered tools while increasing the speed of implementation of new technology. In this PhD research project, the main objective is to examine how synthetic data could be used for validation of AI powered tools prior to clinical implementation and regulatory decision-making. The project will investigate the potential use of synthetic data for validation of AI tools for clinical implementation, focusing on a series of questions: - What are the main risks and benefits that must be considered for assurance of AI tools for healthcare? - What are the legal implications of using augmented synthetic data? - How could synthetic data be used for clinical validation without increasing risk of patient harm? The PhD is divided into three parts. The first part will identify the gaps between existing regulations and risk ares in assurance of AI adoption in healthcare. The second will investigate legal implications of synthetic data and whether it can be used freely or has privacy risks. The third will investigate properties of synthetic datasets to see in what cases a synthetic dataset could outperform real world data and create a framework for evaluating whether a synthetically augmented dataset can be fit for validation.

Funding scheme:

NAERINGSPH-Nærings-phd