Noninvasive LVEDP Detection Using ECG and Pressure Waveforms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring left ventricular end diastolic pressure (LVEDP) in heart failure patients are invasive, costly, and difficult to use in outpatient settings, leading to suboptimal management and high rehospitalization rates.
Innovation Solution
A non-invasive machine learning model using cardiac data from wearable devices, such as smartwatches and smartphones, analyzes arterial pressure waveforms and electrocardiograms to estimate LVEDP through intrinsic frequency and cardiac triangle mapping methodologies, providing accurate and continuous monitoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If implantable pulmonary artery pressure sensor is used, then LVEDP monitoring accuracy is improved, but device invasiveness increases
Solution Approach 1:
The patent uses arterial pressure waveform as an intermediary measurement that can be obtained non-invasively from peripheral arteries. This waveform serves as a mediator to indirectly estimate LVEDP without requiring direct intracardiac sensing, thus resolving the contradiction between measurement accuracy and invasiveness by finding an intermediate measurement approach.
Solution Approach 2:
The patent replaces the mechanical/invasive pulmonary artery pressure sensor with a non-invasive arterial pressure measurement system combined with machine learning algorithms. This substitution transitions from direct mechanical sensing to indirect measurement with computational analysis, eliminating the need for implantable devices while maintaining LVEDP estimation capability.
2Measurement precision
If implantable pulmonary artery pressure sensor is used, then LVEDP monitoring accuracy is improved, but device cost increases
Solution Approach 1:
The patent uses readily available arterial pressure waveform data (which can be obtained from standard medical devices or wearable sensors) as a copy or proxy for the more expensive implantable sensor measurements. This copying approach allows the system to leverage existing, lower-cost measurement infrastructure while achieving comparable LVEDP monitoring through computational methods.
Solution Approach 2:
The patent employs inexpensive, non-implantable arterial pressure measurement devices (such as standard blood pressure monitors or wearable sensors) instead of costly implantable sensors. These disposable or reusable external devices provide sufficient data for LVEDP estimation through machine learning, significantly reducing device cost while maintaining monitoring effectiveness.
3Measurement precision
If conventional LVEDP evaluation techniques are used, then measurement accuracy is improved, but ease of operation decreases
Solution Approach 1:
The patent implements an automated machine learning model that performs LVEDP estimation independently without requiring manual interpretation by clinicians. The system self-processes arterial pressure waveform data and automatically generates LVEDP estimates, eliminating the need for complex manual analysis and making the tool as easy to use as obtaining a standard blood pressure reading.
Solution Approach 2:
The patent transforms the complex task of LVEDP evaluation into a simplified parameter estimation process. By changing the approach from direct pressure measurement to computational estimation based on arterial waveform parameters, the system maintains measurement accuracy while dramatically improving ease of operation through automated algorithmic analysis.
Data Source
AI summary
Some embodiments relate to non-invasive techniques for determining whether a patient has experienced heart failure. In some embodiments, a machine learning model may be trained to determine whether the patient has experienced heart failure using blood pressure waveforms and ECGs taken concurrently. Using the intrinsic frequency methodology and the cardiac triangle mapping methodology, features about the patient's cardiac cycle can be extracted and provided to the trained model as input.