loading
loading
NIH grant 1R01NR022545-01: SCH: Integrated AI and Sensing Systems for Mitigating Heart Failure Readmissions in High-Risk Groups with Obesity
Institute: {'code': 'NR', 'abbreviation': 'NINR', 'name': 'National Institute of Nursing Research', 'admin_org_id': '12500052', 'admin_funding_url': ''}. PI: Wei Gao. Award: $1200000.
Start: 2026-09-25T00:00:00.
Heart failure (HF) is a common, costly and potentially fatal cardiac condition, and the leading cause of
hospitalization in senior adults. HF patients’ readmissions after hospital discharge are frequent and nearly
always associated with worse clinical outcomes, and reducing such readmissions is the key to improve
the HF survival rate. Especially in the group of obese patients, this readmission rate could be higher by
another 30%, due to their higher chances of residual congestion after hospital discharge. A possible
solution is to continuously monitor the patients’ cardiac vital signs and predict any HF readmission risk
from these vital signs, but current wearables are highly inaccurate when measuring cardiac vital signs in
the high-risk group with obesity. The goal of this research is to fundamentally address the above
incapability and reduce HF readmissions in high-risk group with obesity, by developing integrated AI and
sensing systems. Custom wearable sensors will monitor HF patients’ cardiac vital signs, and new
machine learning (ML) models will predict the HF readmission risk based on cross-modal correlations
among multimodal sensory data. The proposed research aims to ensure accuracy, adaptability and
reliability of such prediction. First, we will design biomagnetism-based sensing systems for precise
monitoring of cardiac vital signs with the presence of extra subcutaneous fat. Second, we will extract
cross-modal features from multimodal sensory data, and apply these features to Large Language Models
(LLMs) for predicting HF readmission risk in the natural language domain. Third, to ensure the prediction
model’s generality, we will continuously fine-tune our cardiac LLMs in a distributed manner out of clinic, by
developing new federated learning (FL) techniques. Fourth, when some input data modalities are
unavailable, we will develop incomplete multimodal learning techniques based on knowledge distillation,
to distill useful information related to the missing modalities. The proposed technologies will be evaluated
via lab-controlled testing, experimentation over healthy volunteers and efficacy verification over subjects
with HF conditions. The successful completion of this project will develop cutting-edge embedded sensing
and AI techniques to ensure accurate monitoring of cardiac vital signs on HF patients with obesity, hence
enabling real-time tracking and prediction of HF readmission risk out of clinic and significantly enhancing
the detection and prevention of this fatal cardiac condition.document fetched 2026-10-04 02:43