Systems, methods and media for estimating compensatory reserve and predicting hemodynamic decompensation using physiological data
Inventors
Techentin, Robert W. • Curry, Timothy B. • Joyner, Michael J. • Haider, Clifton R. • Holmes, III, David R. • FELTON, Christopher L. • Gilbert, Barry K. • Van Dorn, Charlotte Sue • Carey, William A. • Convertino, Victor A.
Assignees
Mayo Foundation for Medical Education and Research • United States Department of the Army
Publication Number
US-12076120-B2
Publication Date
2024-09-03
Expiration Date
2040-07-21
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
In accordance with some embodiments, systems, methods, and media for estimating compensatory reserve and predicting hemodynamic decompensation using physiological data are provided. In some embodiments, a system for estimating compensatory reserve is provided, the system comprising: a processor programmed to: receive a blood pressure waveform of a subject; generate a first sample of the blood pressure waveform with a first duration; provide the sample as input to a trained CNN that was trained using samples of the first duration from blood pressure waveforms recorded from subjects while decreasing the subject's central blood volume, each sample being associated with a compensatory reserve metric; receive, from the trained CNN, a first compensatory reserve metric based on the first sample; and cause information indicative of remaining compensatory reserve to be presented.
Core Innovation
The invention provides systems, methods, and media for estimating compensatory reserve and predicting hemodynamic decompensation using physiological data. In particular, the system receives a blood pressure waveform from a subject, generates a sample of the waveform with a specific duration, and inputs the sample into a trained one-dimensional convolutional neural network (1D CNN). This CNN has been trained on similarly sized samples from blood pressure waveforms collected from multiple subjects during controlled decreases of central blood volume, each labeled with an associated compensatory reserve metric. The output from this trained CNN is a quantitative compensatory reserve metric indicating the remaining compensatory reserve of the subject, which can then be presented for clinical use.
The problem addressed arises from the difficulty of determining the amount of blood loss and the patient's proximity to hemodynamic collapse in trauma cases. Individuals respond differently to similar blood losses, and standard vital signs like systolic blood pressure remain stable due to physiological compensatory mechanisms until a threshold blood loss is exceeded. These mechanisms are not easily measurable, and after depletion, rapid hemodynamic collapse follows, often too late for timely intervention. Thus, there is a need for effective, quantitative assessment methods of compensatory reserve to predict and prevent hemodynamic decompensation.
The solution introduced uses a trained computational model, specifically a 1D CNN, to analyze time-series blood pressure waveform data and estimate the compensatory reserve metric (CRM) as a percentage ranging from full reserve to zero at decompensation. Training data is generated using lower body negative pressure (LBNP) on healthy subjects to simulate blood volume loss, providing labeled samples that reflect physiological compensatory states. This method enables automatic feature learning from raw physiological waveforms, overcoming the limitations of manual feature engineering and enabling predictions of impending hemodynamic decompensation even when conventional vital signs appear normal.
Claims Coverage
The claims encompass two independent claims—one system claim and one method claim—each focusing on estimating compensatory reserve using a trained one-dimensional convolutional neural network (1D CNN). The inventive features cover the input data, neural network design, training methodology, and output interpretation.
System for estimating compensatory reserve using trained 1D CNN
The system includes at least one hardware processor programmed to receive a blood pressure waveform from a subject, generate a timed first sample of specified duration, and input this sample into a trained 1D CNN. The CNN is trained as a regression model using samples from blood pressure waveforms recorded during controlled decreases of subjects’ central blood volume, with each sample associated with a compensatory reserve metric. The CNN’s output layer is a linear layer that outputs a single quantitative compensatory reserve metric value indicating the subject’s remaining compensatory reserve. The system also causes presentation of information indicative of this remaining compensatory reserve.
Method for estimating compensatory reserve using trained 1D CNN
The method involves receiving a subject's blood pressure waveform, generating a first sample comprising a time series of blood pressure values with a defined duration, providing this sample as input to a trained 1D CNN, and receiving from the CNN a compensatory reserve metric, which quantitatively indicates the subject’s remaining compensatory reserve. The 1D CNN is trained as a regression model with training samples labeled according to central blood volume decreases. The method concludes by causing presentation of information indicative of the subject’s compensatory reserve.
The independent claims define an innovative approach to estimating compensatory reserve using a trained one-dimensional CNN model that processes time-series blood pressure waveforms specifically labeled according to controlled central blood volume reductions. This includes the system architecture and method steps, detailing the neural network’s structure and the generation and use of a compensatory reserve metric indicative of hemodynamic status.
Stated Advantages
Enables timely and effective treatment of trauma victims by providing a metric indicative of health status prior to hemodynamic collapse.
Uses trained deep convolutional neural networks that automatically learn relevant features from physiological waveforms, eliminating the need for extensive manual feature engineering.
Requires relatively few labeled examples to train an effective model, overcoming the difficulty of obtaining ground truth labels in traumatic injury data.
Predicts individual subject tolerance to blood loss by automatically accounting for both high and low tolerance individuals through the training data.
Produces a quantitative compensatory reserve metric that changes immediately at the onset of blood loss, providing earlier warning signals than standard vital signs like systolic blood pressure.
Documented Applications
Estimating compensatory reserve and predicting hemodynamic decompensation for trauma victims using physiological data, such as blood pressure waveforms.
Training convolutional neural networks to analyze blood pressure waveforms obtained during experiments applying lower body negative pressure to simulate blood volume loss.
Using a portable or compact physiological recording device to capture waveforms like photoplethysmography and ECG signals at high sampling rates for compensatory reserve estimation.
Implementation in computing devices and servers that receive physiological data, process it using trained CNNs, and provide alerts or information to healthcare providers and emergency services regarding a subject’s compensatory reserve and risk of decompensation.
Interested in licensing this patent?