< Program

Special Session: Artificial Intelligence and Hearing Healthcare

Digital Patient Clones for Individualized Clinical Inference
Dennis Barbour, MD, PhD
Professor of Biomedical Engineering, Washington University, St. Louis, MO

Traditional frameworks in evidence-based and precision medicine are deeply rooted in etiological diagnosis, aiming for a mechanistic understanding of diseases. While effective for conditions with universal, identifiable, and predictive mechanisms, this approach struggles to cope with the complexity and heterogeneity inherent in disorders of brain and behavior. I will present an alternative clinical inference methodology that retains the predictive power of conventional diagnostic methods but is engineered to adapt to variable population traits. Central to this new approach are "digital clones"—generative models that recapitulate a patient's clinical history and current findings. These models offer the flexibility to contribute to robust outcome forecasts in both homogeneous and diverse populations. I will demonstrate the application of this framework in auditory and cognitive assessments, discuss its potential for broader medical applications, and highlight how this approach is designed to be intrinsically equitable, thereby better serving outliers, minorities, and patients with rare diseases.

Dennis Barbour, MD, PhD, is a Professor of Biomedical Engineering, Computer Science, Psychological and Brain Sciences, Neuroscience and Otolaryngology at Washington University in St. Louis. His research interests include auditory processing, cognitive neuroscience, machine learning and medical informatics. His development of machine learning perceptual testing has contributed a key founding principle to the emerging field of computational audiology. More recently, he has generalized these concepts toward latent variable models of cognition. The resulting estimation algorithms enable complex models to be trained effectively with limited amounts of data. Currently he is working to incorporate disparate data streams throughout electronic health records into unified models of patient outcomes for optimizing diagnostic and treatment decisions.