Dr. Kaplan’s research centers around developing data-driven methods for problems in several application areas. He currently leads efforts developing advanced machine learning (ML) methods for healthcare and precision medicine applications. The focus of this work is on probabilistic approaches for noisy, heterogeneous, and irregularly sampled data streams, which present difficulties for many existing ML methods. One area is in traumatic brain injury (TBI), which has a large and diverse outcome space, in addition to arbitrary missingness patterns in clinical data points. The developed probabilistic methods allow for the analysis of patient subgroups and likelihood estimation over a large number of outcomes. Research in imaging involves the development of representation learning methods for brain MRI using a novel artificial neural network architecture. This approach compresses volumetric images and partitions the brain into regions that correspond across subjects in a purely data-driven approach.
Unsupervised Learning Approaches Using Messy Data for Precision Medicine
Unsupervised probabilistic models can be used to capture the underlying structure of data despite messy characteristics such as missingness. We use the models for traumatic brain injury data which consist of detailed records of clinical, blood biomarker, imaging, and comprehensive multidimensional outcome data.
The value of data in healthcare is undeniable and realized when raw information is successfully converted into knowledge that changes clinical practice. To drive value improvements and ensure that the right patient receives the right care requires the right data in combination with the right data analytics. This session will cover various aspects and challenges of data science in hospitals and health systems that drive healthcare with better outcomes.