Photoplethysmography, Foundation Models, Hypertension and Diabetes

George Searle

Keywords: type 2 diabetes; hypertension; photoplethysmography; tabpfn

TL;DR: TabPFN improves the detection of hypertension and diabetes using a single heartbeat's worth of PPG signal.

Abstract:

Cardiovascular disease remains the leading cause
of death globally, with hypertension and diabetes
as two key risk factors. These conditions are fre-
quently underdiagnosed because current diagnos-
tic methods often require in-clinic or invasive
procedures, which delay detection until symp-
toms arise - often too late for optimal intervention.
In this work, we focus on photoplethysmogra-
phy (PPG), a non-invasive signal that can be pas-
sively collected using widely available consumer
devices such as smartwatches and smartphones.
This makes PPG particularly well-suited for re-
mote, continuous health monitoring. We leverage
foundation models (PaPaGeI and TabPFN) to ex-
tract features from single-heartbeat PPG signals
to detect hypertension and diabetes. Using data
from 215,000 subjects in the UK Biobank, we
demonstrate that these models significantly out-
perform current state-of-the-art approaches for
PPG-based disease detection.

Submission Number: 102