Blood Pressure Estimation for Hemodynamic Monitoring Using Reduced Photoplethysmography Features in Intensive Care Units: Prospective Machine Learning Study

Background: Continuous, noninvasive blood pressure (BP) monitoring is clinically important for managing hemodynamic instability in the intensive care unit (ICU). However, current ICU practice primarily relies on invasive arterial catheterization. While this is the clinical gold standard for continuous measurement, it carries risks such as infection and thrombosis and is typically limited to select critical care settings. Objective: To reduce reliance on invasive arterial lines, we developed an interpretable estimation framework for systolic blood pressure (SBP), mean arterial pressure (MAP), and diastolic blood pressure (DBP) from single-sensor photoplethysmography (PPG). Methods: Synchronized PPG and arterial line 10-minute recordings from 100 patients in the ICU were analyzed. Using a hybrid feature selection strategy (RReliefF and intrinsic model-specific rankings), we reduced 201 initial features to an optimized subset of 30 core features. Model development and evaluation were designed to minimize information leakage and overfitting, incorporating nested cross-validation and patient-wise testing to assess generalizability across heterogeneous ICU populations. We developed and validated predictive models using machine learning techniques, specifically XGBoost (XGB), random forest, and support vector regression, with XGB serving as our primary model. Results: Among the evaluated models, the Hybrid Top 30 XGB configuration provided the best balance between estimation performance, model compactness, and computational efficiency. The model achieved mean absolute errors of 8.61 mm Hg (SBP), 3.50 mm Hg (MAP), and 2.62 mm Hg (DBP). For MAP and DBP, mean difference (SD) values of -0.09 (4.54) mm Hg and -1.07 (3.51) mm Hg, respectively, satisfied the Association for the Advancement of Medical Instrumentation numerical criteria, while performance corresponded to Grade A according to the British Hypertension Society grading criteria. Although SBP exhibited greater variability, estimation performance was largely preserved using a physiologically informed subset of 30 features (approximately 15% of the original feature set). Feature importance analysis revealed distinct physiological feature patterns across SBP, MAP, and DBP estimation. Conclusions: Interpretability analysis identified key predictors associated with wave reflection, systolic timing, low-frequency vasomotor activity, and autonomic regulation. By explicitly linking predictive features to underlying cardiovascular physiology, the proposed framework addresses a key limitation of existing black-box cuffless BP models. Ultimately, this study demonstrates that an optimized XGB framework using physiologically informed PPG features can provide accurate and computationally efficient BP estimation in ICU settings and may support the assessment of clinically relevant hemodynamic states using a single-sensor PPG platform. These findings provide a foundation for future development of more robust and generalizable interpretable BP estimation models and cuffless monitoring technologies.

Authors

Institutions

Publication Details

Journal
JMIR AI
Published
2026-11-17
DOI
https://doi.org/10.2196/105057
Primary Topic
Non-Invasive Vital Sign Monitoring
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Blood Pressure Estimation for Hemodynamic Monitoring Using Reduced Photoplethysmography Features in Intensive Care Units: Prospective Machine Learning Study

Shivaram P. Arunachalam, Scott A. Helgeson, Suganti Shivaram, JIEUN LEE et al.
JMIR AI
Non-Invasive Vital Sign Monitoring
article

Blood Pressure Estimation for Hemodynamic Monitoring Using Reduced Photoplethysmography Features in Intensive Care Units: Prospective Machine Learning Study

Shivaram P. Arunachalam, Scott A. Helgeson, Suganti Shivaram, JIEUN LEE, Saurabh Ashok Sujanyal, Megha Sasidhar Handral, Mohammed Naveed Shariff, Surbhi Dadwal, Heba M. Alazab, Elcin Omercikoglu, Muhammed Asil Nechooli, Saravanabhavan Chandrasekar, Vidhya C M
article en

Abstract

Background: Continuous, noninvasive blood pressure (BP) monitoring is clinically important for managing hemodynamic instability in the intensive care unit (ICU). However, current ICU practice primarily relies on invasive arterial catheterization. While this is the clinical gold standard for continuous measurement, it carries risks such as infection and thrombosis and is typically limited to select critical care settings. Objective: To reduce reliance on invasive arterial lines, we developed an interpretable estimation framework for systolic blood pressure (SBP), mean arterial pressure (MAP), and diastolic blood pressure (DBP) from single-sensor photoplethysmography (PPG). Methods: Synchronized PPG and arterial line 10-minute recordings from 100 patients in the ICU were analyzed. Using a hybrid feature selection strategy (RReliefF and intrinsic model-specific rankings), we reduced 201 initial features to an optimized subset of 30 core features. Model development and evaluation were designed to minimize information leakage and overfitting, incorporating nested cross-validation and patient-wise testing to assess generalizability across heterogeneous ICU populations. We developed and validated predictive models using machine learning techniques, specifically XGBoost (XGB), random forest, and support vector regression, with XGB serving as our primary model. Results: Among the evaluated models, the Hybrid Top 30 XGB configuration provided the best balance between estimation performance, model compactness, and computational efficiency. The model achieved mean absolute errors of 8.61 mm Hg (SBP), 3.50 mm Hg (MAP), and 2.62 mm Hg (DBP). For MAP and DBP, mean difference (SD) values of -0.09 (4.54) mm Hg and -1.07 (3.51) mm Hg, respectively, satisfied the Association for the Advancement of Medical Instrumentation numerical criteria, while performance corresponded to Grade A according to the British Hypertension Society grading criteria. Although SBP exhibited greater variability, estimation performance was largely preserved using a physiologically informed subset of 30 features (approximately 15% of the original feature set). Feature importance analysis revealed distinct physiological feature patterns across SBP, MAP, and DBP estimation. Conclusions: Interpretability analysis identified key predictors associated with wave reflection, systolic timing, low-frequency vasomotor activity, and autonomic regulation. By explicitly linking predictive features to underlying cardiovascular physiology, the proposed framework addresses a key limitation of existing black-box cuffless BP models. Ultimately, this study demonstrates that an optimized XGB framework using physiologically informed PPG features can provide accurate and computationally efficient BP estimation in ICU settings and may support the assessment of clinically relevant hemodynamic states using a single-sensor PPG platform. These findings provide a foundation for future development of more robust and generalizable interpretable BP estimation models and cuffless monitoring technologies.

JMIR AIVol. 5
Jacksonville College (US), Mayo Clinic in Florida (US), Artificial Intelligence in Medicine (Canada) (CA)
Good health and well-being
Non-Invasive Vital Sign Monitoring
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon