Machine-learnt prediction of uncertainty or sensitivity for hemodynamic quantification in medical imaging
The uncertainty, sensitivity, and/or standard deviation for a patient-specific hemodynamic quantification is determined. The contribution of different information, such as the fit of the geometry at different locations, to the uncertainty or sensitivity is determined. Alternatively or additionally, the amount of contribution of information at one location (e.g., geometric fit at the one location) to uncertainty or sensitivity at other locations is determined. Rather than relying on time consuming statistical analysis for each patient, a machine-learnt classifier is trained to determine the uncertainty, sensitivity, and/or standard deviation for the patient.
1. A method for hemodynamic quantification in a medical imaging system, the method comprising:
scanning a patient with the medical imaging system, the scanning providing cardiac data representing part of a cardiac system of the patient;
determining a patient-specific cardiac geometry from the cardiac data;
extracting values for features of a first input vector of a machine-learnt predictor of the hemodynamic quantification from the patient-specific cardiac geometry;
predicting, by the machine-learnt predictor, a value of the hemodynamic quantification in response to the values of the features of the first input vector;
extracting values for features of a second input vector of a machine-learnt classifier of uncertainty and/or sensitivity of the hemodynamic quantification;
classifying, by the machine-learnt classifier, a contribution to a value or values of the uncertainty and/or sensitivity of the hemodynamic quantification from locations in the patient-specific cardiac geometry in response to the values for the features of the second input vector; and
generating an output based on the value of the hemodynamic quantification and the value or values of the uncertainty and/or sensitivity.
2. The method of claim 1 wherein determining the patient-specific geometry comprises fitting a model to the cardiac data.
3. The method of claim 1 wherein extracting the values for the features of the first input vector and/or the second input vector comprises extracting radii along a vessel represented by the patient-specific cardiac geometry.
4. The method of claim 1 wherein predicting comprises predicting with the machine-learnt predictor trained, at least in part, on synthetic samples.
5. The method of claim 1 wherein predicting comprises predicting a fractional flow reserve as the hemodynamic quantification.
6. The method of claim 1 wherein classifying comprises classifying with the machine-learnt classifier trained based on an uncertainty level based on a scan configuration, reconstruction of the cardiac data, and/or patient characteristics.
7. The method of claim 1 wherein classifying comprises classifying with the machine-learnt classifier trained based on a distribution of the hemodynamic quantification given a sampled distribution of noise.
8. The method of claim 1 wherein classifying comprises classifying with the machine-learnt classifier trained based on sensitivities from a standard deviation of the distribution of the hemodynamic quantification and a correlation of an uncertain variable with a distribution of the hemodynamic quantification.
9. The method of claim 1 wherein predicting comprises predicting the value of the hemodynamic quantification for a first location,
wherein classifying comprises classifying for each of a plurality of second locations, and
wherein generating the output comprises generating a map of the second locations showing contribution to uncertainty and/or sensitivity to the value of the hemodynamic quantification at the first location.
10. The method of claim 1 wherein predicting comprises predicting the value and other values of the hemodynamic quantification for a plurality of first locations, and
wherein classifying comprise classifying the sensitivity of the value and other values to the uncertainty at a second location.
11. The method of claim 1 further comprising separating the patient-specific cardiac geometry into vessel branches and performing the predicting and classifying separately for each of the vessel branches.
12. The method of claim 1 wherein generating comprises generating a map of the sensitivity and/or uncertainty of the hemodynamic quantification as a function of location of the patient-specific cardiac geometry.
13. The method of claim 1 wherein generating comprises generating an output as alphanumeric text of the value of the hemodynamic quantification and the value or values of the uncertainty and/or sensitivity.
14. The method of claim 1 further comprising receiving user selection of a location of the patient-specific cardiac geometry and wherein generating comprises generating the output as the value of the hemodynamic quantification for the location and at least the value or values of the uncertainty and/or sensitivity.
15. The method of claim 1 further comprising receiving a modification of the patient-specific cardiac geometry and repeating the predicting and classifying based on the patient-specific cardiac geometry with the modification.
16. The method of claim 1 further comprising comparing the value of the uncertainty to a threshold, and outputting a recommendation based on a result of the comparison.
17. A system for hemodynamic quantification, the system comprising:
a medical scanner for scanning a patient, the medical scanner configured to output coronary data for the patient;
an image processor configured to extract a patient-specific coronary geometry from the coronary data, compute a hemodynamic quantity for a first location on the patient-specific coronary geometry, determine a confidence statistic of the hemodynamic quantity, the confidence statistic indicating a contribution to the hemodynamic quantity from a second location on the patient-specific coronary geometry different than the first location; and
a display configured to display the hemodynamic quantity and the confidence statistic.
18. The system of claim 17 wherein the image processor is configured to determine the confidence statistic at each of the second location, the first location, and a plurality of third locations of the patient-specific cardiac geometry, and
wherein the display is configured to display a map of the determined confidence statistics.
19. The system of claim 17 wherein the image processor is configured to determine the confidence statistic as uncertainty, sensitivity, and/or standard deviation of the hemodynamic quantity by application of a machine-learnt classifier.
20. The system of claim 17 wherein the image processor is configured to compute the hemodynamic quantity for each of the first location and the second location, and
wherein the image processor is configured to determine the confidence statistics of the hemodynamic quantities based on variation of the second location.