Method and system for assessing vessel obstruction based on machine learning
Methods and systems are described for assessing a vessel obstruction. The methods and systems obtain a volumetric image dataset of a myocardium and at least one coronary vessel, wherein the myocardium comprises muscular tissue of the heart. A three-dimensional (3D) image corresponding to a coronary vessel of interest is created from the volumetric image dataset. Feature data that represents features of both the myocardium and the coronary vessel of interest is generated. At least some of the feature data is determined by a first machine learning-based model based on the 3D image. A second machine learning-based model is used to determine at least one parameter based on the feature data, wherein the at least one parameter represents functionally significant coronary lesion severity of the coronary vessel of interest.
1. A method for assessing a vessel obstruction, comprising:
a) obtaining a volumetric image dataset of a myocardium and at least one coronary vessel, wherein the myocardium comprises muscular tissue of the heart;
b) creating a three-dimensional (3D) image corresponding to a coronary vessel of interest based on the volumetric image dataset of a) and axial trajectory of the coronary vessel of interest that is extracted from the volumetric image dataset of a);
c) generating feature data that represents features of both the myocardium and the coronary vessel of interest, wherein the feature data is determined by i) sampling image data of the 3D image of b) that lies along the axial trajectory of the coronary vessel of interest, ii) using the sampled image data as input to a first machine learning-based model configured to automatically generate at least first feature data representing features of the coronary vessel of interest, iii) generating second feature data from additional myocardium analysis of the volumetric image dataset of a), wherein the second feature data represents features of the myocardium, and iv) combining the first feature data and the second feature data to generate the feature data of c); and
d) employing a second machine learning-based model to determine at least one parameter based on the feature data of c), wherein the at least one parameter represents functionally significant coronary lesion severity of the coronary vessel of interest.
2. The method of claim 1 , wherein:
the at least one parameter that represents functionally significant coronary lesion severity is selected from the group consisting of: a fractional flow reserve (FFR) value, coronary flow reserve value, instantaneous wave-free ratio, hyperemic myocardium perfusion value, index of microcirculatory resistance, pressure drop along a coronary artery, FFR along a coronary artery, and combinations thereof.
3. The method of claim 1 , wherein:
the second feature data is determined using at least one of: a convolutional auto-encoder, Gaussian filters, transmural perfusion ratio, Haralick features, myocardium thickness, and heart shape.
4. The method of claim 1 , wherein:
the second feature data is determined using an unsupervised machine learning which employs a convolution auto-encoder.
5. The method of claim 1 , further comprising:
repeating the operations of c) and d) using image data corresponding to the axial trajectory of the coronary vessel of interest to determine values of the at least one parameter representing functionally significant coronary lesion severity along the length of the coronary vessel of interest.
6. The method of claim 5 , wherein:
the image data corresponding to the axial trajectory of the coronary vessel includes cubes or cuboids that are sampled from the 3D image and extend along the axial trajectory of the coronary vessel of interest.
7. The method of claim 1 , wherein:
the axial trajectory of the coronary vessel of interest corresponds to a centerline of the coronary vessel of interest.
8. The method of claim 1 , wherein:
the 3D image comprises a multi-planer reformatted (MPR) image that is constructed from samples of the volumetric image dataset along an axial trajectory of the coronary vessel of interest.
9. The method of claim 1 , wherein:
multiple 3D images are created based on the volumetric dataset, wherein the multiple 3D images correspond to a plurality of coronary vessels of interest and are based on extraction of axial trajectories for the plurality of coronary vessels of interest.
10. The method of claim 9 , wherein:
the axial trajectories for the plurality of coronary vessels of interest corresponds to centerlines for plurality of coronary vessels of interest.
11. The method of claim 9 , further comprising:
using sampled image data from the multiple 3D images that lies along the axial trajectories for the plurality of coronary vessels of interest as input to the first machine learning-based model to generate data representing features of the plurality of coronary vessels of interest, wherein such data is included in the feature data of c).
12. The method of claim 1 , wherein:
the first machine learning-based model comprises a recurrent convolutional neural network or at least one convolutional autoencoder.
13. The method of claim 1 , wherein:
the second machine learning-based model comprises a supervised machine learning classifier.
14. The method of claim 13 , wherein:
the second machine learning-based model comprises a neural network, at least one SoftMax classifier, or a support vector machine.
15. The method of claim 1 , wherein:
the second machine learning-based model is trained using reference values derived from invasive measurements involving pullback of a guidewire located in a target coronary vessel, measured coronary flow reserve, index of microcirculatory resistance, major adverse cardiac events or revascularization after acquisition of the volume image data set, results of cardiac stress test, and results of myocardial imaging by magnetic resonance imaging (MRI), SPECT, PET, CT perfusion, or ultrasound.
16. The method of claim 15 , wherein:
the training of the second machine learning-based model involves aligning the reference values to spatial coordinates of a 3D image corresponding to the target coronary vessel.
17. A system for assessing a vessel obstruction, comprising:
memory configured to store program instructions and to store a volumetric image dataset of a myocardium and at least one coronary vessel, wherein the myocardium comprises muscular tissue of the heart;
at least one processor that, when executing the program instructions stored in the memory, is configured to:
a) obtain a volumetric image dataset of a myocardium and at least one coronary vessel;
b) create a three-dimensional (3D) image corresponding to a coronary vessel of interest based on the volumetric image dataset of a) and axial trajectory of the coronary vessel of interest that is extracted from the volumetric image dataset of a);
c) generate feature data that represents features of both the myocardium and the coronary vessel of interest, wherein the feature data is determined by i) sampling image data of the 3D image of b) that lies along the axial trajectory of the coronary vessel of interest, ii) using the sampled image data as input to a first machine learning-based model configured to automatically generate at least first feature data representing features of the coronary vessel of interest, iii) generating second feature data from additional myocardium analysis of the volumetric image dataset of a), wherein the second feature data represents features of the myocardium, and iv) combining the first feature data and the second feature data to generate the feature data of c); and
d) employ a second machine learning-based model to determine at least one parameter based on the feature data of c), wherein the at least one parameter represents functionally significant coronary lesion severity of the coronary vessel of interest.
18. The system of claim 17 , wherein:
the at least one parameter that represents functionally significant coronary lesion severity is selected from the group consisting of: a fractional flow reserve (FFR) value, coronary flow reserve value, instantaneous wave-free ratio, hyperemic myocardium perfusion value, index of microcirculatory resistance, pressure drop along a coronary artery, FFR along a coronary artery, and combinations thereof.
19. The system of claim 17 , wherein:
the 3D image comprises a multi-planer reformatted (MPR) image that is constructed from samples of the volumetric image dataset along the axial trajectory of the coronary vessel of interest.
20. The system of claim 17 , wherein:
the axial trajectory of the coronary vessel of interest corresponds to a centerline of the coronary vessel of interest.
21. The system of claim 17 , wherein:
the at least one processor is configured to repeat the operations of c) and d) using image data corresponding to the axial trajectory of the coronary vessel of interest to determine values of the at least one parameter representing functionally significant coronary lesion severity along the length of the coronary vessel of interest.
22. The system of claim 17 , wherein:
the first machine-learning based model is configured to automatically extract image features associated with cubes or cuboids that are sampled from the 3D image and extend along the axial trajectory of the coronary vessel of interest.
23. The system of claim 17 , wherein:
the at least one processor is configured to create multiple 3D images based on the volumetric dataset, wherein the multiple 3D images correspond to a plurality of coronary vessels of interest and are based on extraction of axial trajectories for the plurality of coronary vessels of interest.
24. The system of claim 23 , wherein:
the axial trajectories for the plurality of coronary vessels of interest corresponds to centerlines for plurality of coronary vessels of interest.
25. The system of claim 23 , wherein:
the at least one processor is configured to use sampled image data from the multiple 3D images along the axial trajectories for the plurality of coronary vessels of interest as input to the first machine learning-based model to generate data representing features of the plurality of coronary vessels of interest, wherein such data is included in the feature data of c).
26. The system of claim 17 , wherein:
the first machine learning-based model comprises a recurrent convolutional neural network or at least one convolutional autoencoder.
27. The system of claim 17 , wherein:
the second machine learning-based model comprises a supervised machine learning classifier.
28. The system of claim 27 , wherein:
the second machine learning-based model comprises a neural network, at least one SoftMax classifier, or a support vector machine.
29. The system of claim 17 , wherein:
the second machine learning-based model is trained using reference values derived from invasive measurements involving pullback of a guidewire located in a target coronary vessel.
30. The system of claim 29 , wherein:
the second machine learning-based model is trained by aligning the reference values to spatial coordinates of a 3D image corresponding to the target coronary vessel.
31. The system of claim 17 , wherein:
the second feature data is determined using at least one of: a convolutional auto-encoder, Gaussian filters, transmural perfusion ratio, Haralick features, myocardium thickness, and heart shape.
32. The system of claim 17 , wherein:
the second feature data is determined using an unsupervised machine learning which employs a convolution auto-encoder.