Adaptive patient benchmarking
Most blood tests are benchmarked against ‘reference intervals’ derived from healthy populations. These
population-level
definitions of normality ignore genetic heritability and clinical context. We develop computational
approaches to
create adaptive benchmarks which are both patient- and context-specific, helping to realize a
longstanding goal of personalized medicine.
AI for clinical and anatomic pathology
Pathologists spend so much time on work that isn’t patient-facing. We build tools that can help reduce
clinical workload,
and lead to more efficient and reproducible workflows. This includes developing computer vision tools
for cell analyses,
predicting demand for blood product usage, and using large-language models to augment sign-outs.
Modelling pulmonary and hematologic biophysics
Blood is a living fluid, whose dynamic properties depend on properties of the cells within it. We use
mathematical modelling
to simulate the behavior of blood both as a fluid, and as a series of interacting cell populations.
Models are used to improve
mechanistic understanding of oxygenation, to capture and predict recovery responses to acute bleeding
events, and to ultimately
inform treatment decisions in acute care.
Understanding immune dynamics following acute events
Trauma, ischemia, and infection all set off a cascade of inflammatory signaling processes as the body
recovers. We use computational
tools to better understand how blood cells respond to and influence the inflammatory environment, and
how this defines recovery.
We collaborative extensively with clinicians and experimentalists to probe the immune system using
mathematical models,
machine learning, clinical specimen analysis, and mouse models.
