The recent advances in AI have accelerated interest in the use of digital twins and virtual models in medicine. The presentation will provide an overview of biomedical digital twins and how AI is impacting the development of precision medicine digital twins. The presentation will provide a glimpse into the future where diagnostics, virtual models, and AI algorithms converge to provide new avenues for optimal treatment in precision medicine.

Once an AI model is developed, it is only a fraction of the way to impacting outcomes. The model must seamlessly integrate into an extremely complex and nuanced health care system, where heterogeneous clinical workflows can challenge the success of any given AI model. Innovators, clinicians, and health systems must learn how to safely deploy solutions. In order to accomplish this, however, a novel approach to solution classification, clinical risk assessment, solution evaluation, quality management, monitoring, and bias assessment must be developed and applied in practice. Then, AI systems may be more effectively deployed within workflows and create real, sustainable value and impact health outcomes.

We apply reinforcement learning (RL) to optimize sequential lab test panel selection for cost-effective and accurate diagnosis in cases where clinical data is imbalanced. We introduce a reward shaping approach to find Pareto-optimal policies for budget-constrained score maximization. Our model achieves state-of-the-art diagnosis accuracy with substantial cost savings (up to 85% reduction) across various clinical tasks.

Our group is developing an AI-powered clinician decision health care platform comprised of real-time ingestion of EHR data features aggregated in a federated learning architecture. Federated learning models dynamically couple to an AI-enabled wearable sensor that indicates early-stage infection (12 hours prior to standard-of-care). The dynamically coupled model will notify the clinician of critical change in patient infectious status allowing for expedited lifesaving therapeutics.

Can we leverage Artificial Intelligence (AI) to address clinical tasks that involve interpreting multimodal patient data and clinicians’ cognitive capabilities—or even beat the clinicians’ performance? We describe novel AI tools developed at Houston Methodist Hospital that address clinical challenges in triage, diagnosis, and treatment of stroke in emergency rooms and beyond.

Novel technologies including AI promise to revolutionize healthcare, from earlier diagnosis to better treatments and improved clinical care—to name just a few. However, some of the perpetual concerns are the scarcity of data and the uncertainty in the regulatory frameworks. In addition, many of the stakeholders in the fast-moving space are working in silos. This presentation will highlight a few concrete examples of how public-private partnerships such as MDIC are bringing together stakeholders across the ecosystem and delivering unique solutions to accelerate and adopt innovation in healthcare.