Synthetic data-driven hemodynamic determination in medical imaging

Patent metadata

Field	Value
Publication number	US-10993687-B2
Application number	US-201816146045-A
Country	US
Kind code	B2
Filing date	Sep 28, 2018
Priority date	Nov 24, 2014
Publication date	May 4, 2021
Grant date	May 4, 2021

Abstract

In hemodynamic determination in medical imaging, the classifier is trained from synthetic data rather than relying on training data from other patients. A computer model (in silico) may be perturbed in many different ways to generate many different examples. The flow is calculated for each resulting example. A bench model (in vitro) may similarly be altered in many different ways. The flow is measured for each resulting example. The machine-learnt classifier uses features from medical scan data for a particular patient to estimate the blood flow based on mapping of features to flow learned from the synthetic data. Perturbations or alterations may account for therapy so that the machine-trained classifier may estimate the results of therapeutically altering a patient-specific input feature. Uncertainty may be handled by training the classifier to predict a distribution of possibilities given uncertain input distribution. Combinations of one or more of uncertainty, use of synthetic training data, and therapy prediction may be provided.

First claim

We claim:

1. A method for hemodynamic determination in medical imaging, the method comprising:
acquiring medical scan data representing an anatomical structure of a patient;
extracting a set of features from the medical scan data;
specifying an uncertainty for one or more of the features;
inputting, by a processor, the features to a machine-trained classifier, the machine trained classifier trained to output a confidence; and
outputting, by the processor with application of the machine-trained classifier to the features, a value of a hemodynamic metric and the confidence for the value, the confidence part of a representation of the hemodynamic metric given the uncertainty.

2. The method of claim 1 wherein acquiring comprises acquiring angiography data.

3. The method of claim 1 wherein acquiring comprises acquiring with the medical scan data comprising a two or three-dimensional representation of the anatomical structure.

4. The method of claim 1 wherein extracting the set of the features comprises:
extracting geometrical features of the anatomical structure; and
extracting the features of one or more abnormalities of the anatomical structure.

5. The method of claim 1 wherein extracting the set of the features comprises extracting functional features representing operation of the anatomical structure, wherein the machine-trained classifier is trained from virtual representations of the operation of the anatomical structure.

6. The method of claim 1 wherein extracting the set of the features comprises extracting an ischemic weight and an ischemic contribution score, the ischemic contribution score being a function of the ischemic weight.

7. The method of claim 6 wherein extracting the ischemic weight comprises computing branch ischemic weights as a function of a global ischemic weight.

8. The method of claim 6 wherein extracting the ischemic contribution score comprises computing the ischemic contribution score as a function of the ischemic weight and a radius.

9. The method of claim 1 wherein extracting the set of the features comprises extracting branch interaction features.

10. The method of claim 1 wherein inputting comprises inputting to the machine-trained classifier trained from synthetic data, the synthetic data comprising an in vitro model with a ground truth of the hemodynamic metric measured from the in vitro model.

11. The method of claim 1 wherein inputting comprises inputting to the machine-trained classifier trained from synthetic data, the synthetic data comprising an in silico model with a ground truth of the hemodynamic metric computed with computation fluid dynamics.

12. The method of claim 1 wherein inputting comprises inputting a sub-set of the set of features, the sub-set for a sub-part of the anatomical structure, and wherein outputting comprises outputting the hemodynamic metric for the sub-part of the anatomical structure; and
further comprising subsequently repeating the inputting and outputting for remaining features of the set for another part of the anatomical structure.

13. The method of claim 1 wherein outputting comprises outputting the value of the hemodynamic metric on a display with an image of the anatomical structure generated from the medical scan data.

14. The method of claim 1 further comprising predicting another value of the hemodynamic metric with another machine-trained classifier using at least one of the features and patient characteristics as input features.

15. The method of claim 1 wherein inputting comprises inputting to the machine-trained classifier trained from synthetic data, the synthetic data comprising examples generated by regular variation of an in vitro model, in silico model, or both in vitro and in silico models.

16. The method of claim 1 wherein outputting the confidence comprises outputting confidence intervals.

17. The method of claim 1 wherein specifying comprises specifying the uncertainty by a user input or selection of a distribution.

18. A method for hemodynamic determination in medical imaging, the method comprising:
generating a plurality of examples of anatomical arrangements;
assigning a first uncertainty to a feature of the anatomical arrangements;
storing a value for a flow characteristic for each of the examples of the anatomical arrangements;
determining a second uncertainty for the flow characteristic based on the first uncertainty; and
training, with machine learning, using the second uncertainty and the stored value for the flow characteristic for each of the examples of the anatomical arrangements, a classifier for predicting a distribution of the flow characteristics for different anatomical arrangements, the classifier trained to output the distribution as confidence intervals.

19. The method of claim 18 wherein generating comprises generating with synthetic data not representing any particular patient with perturbing computer modeling, physical modeling, or both in a systematic pattern based on the first uncertainty.

Assignees

• Siemens Healthcare Gmbh

Inventors

• Itu Lucian Mihai
• Passerini Tiziano
• Rapaka Saikiran
• Sharma Puneet
• Schwemmer Chris
• Schoebinger Max
• Redel Thomas
• Comaniciu Dorin

Classifications

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06F18/2413

  based on distances to training or reference patterns · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06V10/42

  Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation · CPC title