Scholarlink

Background <p>Invasive pressure–volume loop (PVL) analysis is the gold standard method to quantify myocardial physiology. To enhance broader clinical applicability, non-invasive approaches to derive left ventricular (LV) PVLs have gained prominence, typically estimating LV end-systolic pressure (ESP) from brachial blood pressure (BP) alone. However, this approach does not account for factors that may modulate the pressure waveform as it propagates from the LV to the brachial artery, notably arterial stiffness. We trained an artificial-intelligence (AI) model to predict invasive LVESP from a set of non-invasive hemodynamic parameters and compared its performance with commonly used equations for estimating LVESP using solely brachial BPs.</p> Methods <p>An in silico cohort of healthy adults (<i>N</i> = 3,837, age: 48.6 ± 17.0y), previously shown to have good agreement for aortic and brachial BP with in vivo data, was split into training and validation cohorts using an internal hold-out validation approach. Clinically relevant parameters were selected as inputs to a random forest machine learning model.</p> Results <p>The best performing model included systolic and diastolic brachial BP, heart rate, age, and carotid-femoral pulse wave velocity (PWV), and reliably predicted LVESP (root mean square error: 0.99&#xa0;mmHg). AI-derived LVESP correlated more strongly with simulated invasive LVESP (r = 0.994) than commonly used equations (r = 0.909–0.908). Bland–Altman analysis demonstrated significant bias in brachial BP-only equations (mean bias: -4.01, limits of agreement [LoA] [± 1.96 standard deviations]: -11.1 to 3.06&#xa0;mmHg), with the direction of bias varying across different ages. AI-derived LVESP displayed smaller bias (mean bias: 0.002, LoA: -1.94 to 1.94&#xa0;mmHg) and had negligible age-related bias across the lifespan.</p> Conclusions <p>In this hypothesis generating study, a novel non-invasive, AI-enabled approach to LVESP quantification incorporating age, PWV and other key hemodynamic parameters alongside brachial BPs yielded smaller average error across the lifespan than commonly used brachial BP-only equations. These results warrant validation in vivo to ascertain real-world predictive performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial intelligence-enabled non-invasive quantification of left ventricular end-systolic pressure: an in silico analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Edouard Long,
  • Roberta Dunn,
  • Vitaliy Androshchuk

摘要

Background

Invasive pressure–volume loop (PVL) analysis is the gold standard method to quantify myocardial physiology. To enhance broader clinical applicability, non-invasive approaches to derive left ventricular (LV) PVLs have gained prominence, typically estimating LV end-systolic pressure (ESP) from brachial blood pressure (BP) alone. However, this approach does not account for factors that may modulate the pressure waveform as it propagates from the LV to the brachial artery, notably arterial stiffness. We trained an artificial-intelligence (AI) model to predict invasive LVESP from a set of non-invasive hemodynamic parameters and compared its performance with commonly used equations for estimating LVESP using solely brachial BPs.

Methods

An in silico cohort of healthy adults (N = 3,837, age: 48.6 ± 17.0y), previously shown to have good agreement for aortic and brachial BP with in vivo data, was split into training and validation cohorts using an internal hold-out validation approach. Clinically relevant parameters were selected as inputs to a random forest machine learning model.

Results

The best performing model included systolic and diastolic brachial BP, heart rate, age, and carotid-femoral pulse wave velocity (PWV), and reliably predicted LVESP (root mean square error: 0.99 mmHg). AI-derived LVESP correlated more strongly with simulated invasive LVESP (r = 0.994) than commonly used equations (r = 0.909–0.908). Bland–Altman analysis demonstrated significant bias in brachial BP-only equations (mean bias: -4.01, limits of agreement [LoA] [± 1.96 standard deviations]: -11.1 to 3.06 mmHg), with the direction of bias varying across different ages. AI-derived LVESP displayed smaller bias (mean bias: 0.002, LoA: -1.94 to 1.94 mmHg) and had negligible age-related bias across the lifespan.

Conclusions

In this hypothesis generating study, a novel non-invasive, AI-enabled approach to LVESP quantification incorporating age, PWV and other key hemodynamic parameters alongside brachial BPs yielded smaller average error across the lifespan than commonly used brachial BP-only equations. These results warrant validation in vivo to ascertain real-world predictive performance.