Artificial intelligence for physiological quantification in medical imaging

1. A method for physiological quantification in a medical system, the method comprising:

acquiring medical scan data representing anatomy of a patient;

applying, by a machine, the medical scan data or features from the medical scan data to a machine-trained estimation network, the machine-trained estimation network having been trained, at least in part, from synthetic data generated by a generator of a machine-trained generative adversarial network, the synthetic data not specific to existing patient cases;

determining, by the application to the machine-trained estimation network, a pressure-volume loop for the patient, the pressure-volume loop determined within a second of the acquisition of the medical scan data; and

outputting the pressure-volume loop.

2. The method of claim 1 wherein acquiring comprises acquiring computed tomography data, magnetic resonance data, or molecular imaging data, and further comprising acquiring other patient information for the patient, the other patient information being other measures, a state, test results, biochemical information, demographic information, and/or clinical history, wherein the other patient information or features from the other patient information are applied to the machine-trained estimation network with the medical scan data or features from the medical scan data.

3. The method of claim 1 wherein the machine-trained estimation network comprises a deep learned network, and wherein determining comprises determining by the deep learned network.

4. The method of claim 1 wherein determining comprises determining the pressure-volume loop and a value of another quantification of a cardiac system of the patient.

5. The method of claim 1 wherein applying comprises applying to the machine-trained estimation network having been trained from the synthetic data generated by the generator of the machine-trained generative adversarial network constrained by a discrete variable characterizing a condition.

6. The method of claim 5 wherein the machine-trained generative adversarial network (GAN) constrained by the discrete variable comprises an infoGAN.

7. The method of claim 5 wherein the discrete variable characterizes cardiac pathology.

8. The method of claim 1 wherein applying comprises applying to the machine-trained estimation network having been trained from the synthetic data generated by the generator of the machine-trained generative adversarial network, the machine-trained generative adversarial network comprising the generator and a discriminator, the discriminator receiving the synthetic data and ground truth from the generator and a pressure-volume loop data from a physiological model.

9. The method of claim 1 wherein determining comprises determining a prediction of an evolution of the pressure-volume loop for the patient.

10. The method of claim 9 wherein applying comprises applying data from a patient wearable sensor to the machine-trained estimation network.

11. The method of claim 9 wherein determining the prediction and the pressure-volume loop comprises determining with the machine-learned estimation network being a multi-task neural network.

12. The method of claim 1 further comprising inputting the pressure-volume loop or a feature from the pressure-volume loop to a machine-learned decision support network, the machine-learned decision support network outputting a clinical decision in response to the inputting.

13. The method of claim 1 wherein outputting comprises displaying the pressure-volume loop and an image of the anatomy of the patient.

14. A method for machine training for synthetic data generation in a medical system for quantification, the method comprising:

training, by a machine, a generative adversarial network to generate synthetic data by a generator representing various instances of anatomy based on a plurality of samples of anatomy of patients and feedback from a discriminator; and

training, by the machine, a quantification network to infer a physiological quantity, the quantification network trained based on the synthetic data from the generative adversarial network.

15. The method of claim 14 wherein training the generative adversarial network comprises training to generate the synthetic data as medical images, and wherein training the quantification network comprises training the quantification network to infer the physiological quantity based on the medical images.

16. The method of claim 14 wherein training the generative adversarial network comprises training the generative adversarial network with one or more discrete random variables parameterizing characteristics of the synthetic data, and further comprising controlling the discrete random variables as a control on a distribution of conditions represented by the synthetic data.

17. The method of claim 14 wherein training the generative adversarial network comprises training the generative adversarial network to generate the synthetic data representing the various instances of the anatomy and to generate ground truth for each of the samples of the synthetic data, wherein training the quantification network comprises training based on the synthetic data and the ground truth.

18. A system for physiological quantification, the system comprising:

a medical imaging scanner configured to scan a patient;

an image processor configured to apply information from the scan to a machine-learned network, the machine-learned network configured by training to output a pressure volume loop in response to the application of the information, the output being in real time with the scan, wherein the machine-learned network having been trained with data generated by perturbing a physiological model; and

a display configured to display the pressure volume loop.

19. The system of claim 18 wherein the output being in real time comprises within a second of completion of the scan.

20. The system of claim 18 wherein the data represents anatomy and is generated by perturbing the physiological model of the anatomy.