Cardiac digital twins combine AI approaches in physics/physiologically-constrained frameworks
To develop trustworthiness in digital twins we will test how to perform global sensitivity analysis on a virtual cohort
Establish checks and reporting of cohort global sensitivity results
Accelerate sensitivity analysis across cohorts
Reporting of sensitivity analysis within a regulatory framework
Cardiac digital twins are increasingly combined with machine learning and AI approaches to provide a physics and physiologically-constrained framework to interpret multi-modal imaging data. These twins are then used to characterize tissue properties from medical imaging and diagnostics and provide decision-support tools to guide the delivery, development, and evaluation of new medical devices and therapies. The development of regulatory standards and the credibility of cardiac digital twins will support wider adoption of computational modelling and simulation tools as “software as medical device” for guiding patient therapy. However, best practice for regulatory submissions has yet to be established. The US Food and Drug Administration (FDA) has recently published proposed steps needed to establish the credibility of patient-specific models or digital twins. As part of this process sensitivity analysis is a crucial step in the quantitative evaluation of digital twin trustworthiness and credibility. A conventional sensitivity analysis may be performed over a single model, however, for patient digital twins there may be thousands of models for different patients. Each model may be subtly different due to inter-patient variability, different data used to constrain the model or different model creation steps. How these variations impact sensitivity analysis and if this means a sensitivity analysis is required on each patient’s digital twin is not known. As medical imaging companies adopt digital twins for intelligently interpreting and integrating medical imaging data there is a growing need to establish best practices for performing sensitivity analysis in digital twins across cohorts of patients. This project will work with Siemens Healthineers and regulatory bodies, to create an exemplary global sensitivity analysis for a virtual patient digital twin cohort follow that will act as a regulatory science tool for informing regulatory submissions.