Funded Studies
Study Rationale: In today’s modern age of computational smart health, actionable machine learning (AML) methods automatically and progressively learn how to correctly answer queries about the state of a person’s health based on the individual’s digital clinical, imaging and biological measurement data. Our proposed study designs, develops and applies such AML methods to identify essential, sensitive, patient-specific biomarkers that reveal an individual’s state and point toward trigger events that signal changes in that person’s health. AML computations automatically train themselves and demonstrably improve their machine-learned ability for diagnosis and prognosis of Parkinson’s disease (PD).
Hypothesis: We hypothesize that the accumulation of misfolded proteins in intracellular spaces can be diagnosed early when localized and cross-correlated with multimodal data (from clinical, imaging, biological, genetic and motor exams) combined with newer biomarkers using modern AML methods.
Study Design: Our modern computational bio-discovery framework harnesses the supercomputers at the University of Texas, which are trained to explore and exploit and combine multimodal data through gestalt reinforcement learning. Task-specific and data-modality specific AML multi-agents are trained individually, and in orchestrated control of a master agent, to rapidly perform accurate localization, identification and ranking of bio-markers that signal the accumulation of misfolded proteins and neuronal cell loss.
Impact on Diagnosis/Treatment of Parkinson’s disease: The inclusion of advanced, multi-parametric MRI to the data analysis and training mix of our AML methods offers the opportunity for rapid, patient-specific diagnosis, treatment and prognosis.
Next Steps for Development: The multi-AML trained agents we design and develop through this study will be able to perform clinical, biological, and radiological interpretations. Next, our AML agents must be field tested in realistic clinical environments to enable health practitioners and diagnostic radiologists to make more accurate, computer-assisted data interpretation, diagnoses, and prognoses.