AI Network for Real-Time Pressure-Volume Loop Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging techniques for estimating left ventricular pressure-volume loops are invasive, costly, and computationally demanding, making them unsuitable for real-time applications such as intra-operative guidance due to the need for numerical solutions that take minutes or hours to determine.
Innovation Solution
A machine-learned network is applied for quicker estimation of physiological parameters using synthetic data generated by a generative adversarial network, allowing for real-time determination of pressure-volume loops from medical scan data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If model-based approaches with complex mathematical models are used for non-invasive estimation of PV loops, then measurement precision is improved, but productivity deteriorates due to computational demands taking minutes or hours
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning networks offline using complex physiological models and大量 training data. The trained network parameters are stored and reused during clinical operations, eliminating the need to solve complex differential equations in real-time while maintaining accurate PV loop estimation.
Solution Approach 2:
The patent substitutes the mechanical/computational system of solving complex mathematical models with a machine learning-based system. The trained neural network replaces the complex physiological models and numerical solvers, providing fast inference that runs in seconds rather than minutes or hours, thus achieving real-time performance.
2Measurement precision
If invasive measurement approaches are used to obtain PV loops, then measurement precision is improved, but object-affected harmful factors worsen due to patient risks and cost
Solution Approach 1:
The patent introduces an intermediary approach by using machine learning models trained on combined data from invasive measurements and non-invasive imaging. The trained network then performs non-invasive estimation, acting as a mediator that captures the accuracy of invasive methods while avoiding their direct risks and costs during clinical application.
Solution Approach 2:
The patent uses copying by training the machine learning model on data from invasive measurements (which provide ground truth PV loops) and then using this trained model to generate copies of the invasive measurement capability through non-invasive imaging alone, eliminating the need for actual invasive procedures during operation.
3Measurement precision
If complex image segmentation and physiological models are used, then measurement precision is improved, but device complexity worsens making real-time adoption difficult
Solution Approach 1:
The patent extracts the complexity from the real-time operation phase by moving it to the offline training phase. The complex physiological models and segmentation algorithms are used during training to build accurate predictions, but only simple data input and network inference remain during clinical use, dramatically reducing operational complexity.
Solution Approach 2:
The patent applies preliminary action by performing all complex model personalization, parameter estimation, and model training in advance during an offline training phase. During clinical operations, the system simply applies the pre-trained model to new patient data, eliminating the need for complex real-time computations while maintaining high precision.
Data Source
AI summary
For quicker estimation of physiological parameters than using a numerical solution, a machine-learned network is applied. The PV loop may be estimated for a specific patient in real-time without invasive pressure measurements. Synthetic data instead of or in addition to actual patient examples may be used to machine train the network, providing a broader and/or controlled range of examples for more accurate estimation even in rarely occurring pathologies. The synthetic data may be generated by a generative adversarial network.