Cardiovascular Device Innovation & Regulatory Science: Virtual Chimaeras and In-Silico Trials with novel Hybrid Machine Learning
Lead Research Organisation:
University of Manchester
Department Name: School of Health Sciences
Abstract
INSILICO will establish the first integrated framework combining data- and knowledge-driven machine learning, realising in-silico trials (ISTs) in medical devices (MDs). Novel in-silico insights on MD safety and efficacy will impact regulatory science and innovation significantly by reducing R&D costs and speeding up regulatory clearance.
I propose a new way to conceive ISTs as multi-model ensemble spaces of virtual experiments, equivalent to enrolling a cohort of synthetic, verifiably realistic virtual patients (VPs). Each VP will harbour a virtually implanted MD operating within physiological envelopes, modelling the patient's short-/long-term response. MD's performance and design will be predicted under diverse physiological regimes, highlighting uncertainties only encountered in late-phase clinical testing.
INSILICO will overcome 3 high-risk high-impact technical barriers by 1) creating virtual patient cohorts reflecting various anatomy, physiology, and pathology ingesting real-world data from real patient populations, 2) accurately predicting interventional outcomes in virtual populations, 3) ensuring the reliability and scalability of computational predictions while accounting for aleatoric/epistemic uncertainties.
The proposed unified physics-informed graph learning scheme will facilitate both the generation of VPs and physically consistent simulations. This project will 1) introduce the concept of virtual chimaeras, 2) extend physics-informed learning over graph networks to construct new reliable, accurate and fast multiphysics simulators, and 3) re-enact a unique industry-provided trial dataset to grow trust in ISTs by industry, trialists and regulators.
INSILICO underpins next-generation ISTs, a paradigm shift beyond current conventional clinical trials as the primary source of scientific evidence on MD safety and efficacy. INSILICO will fundamentally transform MD regulatory science and innovation.
I propose a new way to conceive ISTs as multi-model ensemble spaces of virtual experiments, equivalent to enrolling a cohort of synthetic, verifiably realistic virtual patients (VPs). Each VP will harbour a virtually implanted MD operating within physiological envelopes, modelling the patient's short-/long-term response. MD's performance and design will be predicted under diverse physiological regimes, highlighting uncertainties only encountered in late-phase clinical testing.
INSILICO will overcome 3 high-risk high-impact technical barriers by 1) creating virtual patient cohorts reflecting various anatomy, physiology, and pathology ingesting real-world data from real patient populations, 2) accurately predicting interventional outcomes in virtual populations, 3) ensuring the reliability and scalability of computational predictions while accounting for aleatoric/epistemic uncertainties.
The proposed unified physics-informed graph learning scheme will facilitate both the generation of VPs and physically consistent simulations. This project will 1) introduce the concept of virtual chimaeras, 2) extend physics-informed learning over graph networks to construct new reliable, accurate and fast multiphysics simulators, and 3) re-enact a unique industry-provided trial dataset to grow trust in ISTs by industry, trialists and regulators.
INSILICO underpins next-generation ISTs, a paradigm shift beyond current conventional clinical trials as the primary source of scientific evidence on MD safety and efficacy. INSILICO will fundamentally transform MD regulatory science and innovation.