Left ventricular volume and cardiac output estimation using machine learning model
Methods and systems are disclosed for creating and using a neural network model to estimate a cardiac parameter of a patient, and using the estimated parameter in providing blood pump support to improve patient cardiac performance and heart health. Particular adaptations include adjusting blood pump parameters and determining whether and how to increase or decrease support, or wean the patient from the blood pump altogether. The model is created based on neural network processing of data from a first patient set and includes measured hemodynamic and pump parameters compared to a cardiac parameter measured in situ, for example the left ventricular volume measured by millar (in animals) or inca (in human) catheter. After development of a model based on the first set of patients, the model is applied to a patient in a second set to estimate the cardiac parameter without use of an additional catheter or direct measurement.
1. A method of estimating a cardiac parameter for a patient, the method comprising:
obtaining patient data for a first set of patients, wherein the patient data comprises measurements of at least one blood pump parameter for each patient in the first set of patients and measurements of at least one hemodynamic parameter for each patient in the first set of patients;
building, with a neural network, a model of a cardiac parameter based on the patient data, wherein the neural network communicates over time in hidden states, and wherein the neural network iteratively develops the model based on a plurality of activation functions;
operating a blood pump in a second patient in a second set of patients; and
applying the model to the second patient by:
measuring the at least one blood pump parameter in the second patient to acquire a blood pump parameter measurement;
measuring the at least one hemodynamic parameter in the second patient to acquire a hemodynamic parameter measurement; and
estimating a cardiac parameter for the second patient, wherein the cardiac parameter for the second patient is output by the model based on the blood pump parameter measurement for the second patient and the hemodynamic parameter measurement for the second patient.
2. The method of claim 1 , wherein each one of the measurements of the at least one hemodynamic parameter for the first set of patients is obtained with a respective sensing catheter separate from a respective blood pump.
3. The method of claim 1 , further comprising associating the model with patient information describing the first patient set, wherein the patient information comprises a diagnosis or a demographic for each patient in the first set of patients.
4. The method of claim 3 , further comprising determining whether the model applies to the second patient based on the patient information associated with the model.
5. The method of claim 1 , further comprising computing a suggested change in a pump speed based on the estimated cardiac parameter for the second patient.
6. The method of claim 5 , further comprising implementing the suggested change in the pump speed.
7. The method of claim 1 , wherein the neural network comprises a plurality of cells, and wherein a first cell of the plurality of cells accepts as inputs the measurements of the at least one hemodynamic parameter and the measurements of the at least one blood pump parameter for the first set of patients at a first time point.
8. The method of claim 7 , wherein the first cell transforms the measurements of the at least one hemodynamic parameter and the measurements of the at least one blood pump parameter based on one or more model fits, before transmitting the transformed measurements of the at least one hemodynamic parameter and the transformed measurements of the at least one blood pump parameter to a second cell of the plurality of cells.
9. The method of claim 1 , wherein the at least one hemodynamic parameter is an aortic pressure, and wherein the at least one blood pump parameter is a pump flow.
10. The method of claim 9 , wherein the cardiac parameter is one of a cardiac output, a cardiac power output, a stroke volume, or a compliance.
11. The method of claim 1 , wherein the neural network is a stacked bi-directional recurrent neural network.
12. The method of claim 1 , wherein the neural network has three or more levels of cells stacked between an input row and an output row.
13. The method of claim 1 , wherein the neural network is a fully recurrent neural network, an Elman network, a Hopfield network, an Echo state network, or a hierarchical neural network.
14. The method of claim 1 , wherein the plurality of activation functions comprises a sigmoidal function.
15. The method of claim 1 , wherein the plurality of activation functions comprises a hyperbolic tangent function.
16. The method of claim 1 , wherein the plurality of activation functions comprises a sigmoidal function and a hyperbolic tangent function.
17. A system for estimating a cardiac parameter of a patient based on a pre-determined model, the system comprising:
a blood pump comprising:
a rotor configured to be driven at one or more pump speeds; and
a sensor configured to measure a hemodynamic parameter; and
a controller comprising:
a driver configured to drive the rotor;
a display configured to display one or more parameters; and
a non-transitory computer readable storage medium storing:
a pre-determined model of a cardiac parameter, wherein the pre-determined model is formed by a neural network based on patient data for a first set of patients, wherein the patient data comprises measurements of at least one blood pump parameter for each patient in the first set of patients and measurements of at least one hemodynamic parameter for each patient in the first set of patients, wherein the neural network communicates over time in hidden states, and wherein the neural network iteratively develops the pre-determined model based on a plurality of activation functions; and
instructions that, when executed by one or more processors, cause the one or more processors to:
receive a hemodynamic parameter measurement for a second patient in a second set of patients from the sensor and record the hemodynamic parameter measurement;
receive a pump speed of the driven blood pump rotor from the driver and record the pump speed;
determine from the pre-determined model, based on the hemodynamic parameter measurement and the pump speed, an associated cardiac parameter; and
transmit the determined cardiac parameter to the display.
18. The system of claim 17 , wherein the instructions further cause the one or more processors to:
determine a recommended change to the pump speed based on the determined cardiac parameter; and
transmit the recommended change to the pump speed to the display.
19. The system of claim 17 , wherein the sensor is configured to measure at least one of aortic pressure, left ventricular end diastolic pressure, and capillary wedge pressure.
20. The system of claim 17 , wherein the plurality of activation functions comprises a sigmoidal function and a hyperbolic tangent function.