Method and system for assessing vessel obstruction based on machine learning
Methods and systems are provided for assessing the presence of functionally significant stenosis in one or more coronary arteries, further known as a severity of vessel obstruction. The methods and systems can implement a prediction phase that comprises segmenting at least a portion of a contrast enhanced volume image data set into data segments corresponding to wall regions of the target organ, and analyzing the data segments to extract features that are indicative of an amount of perfusion experiences by wall regions of the target organ. The methods and systems can obtain a feature-perfusion classification (FPC) model derived from a training set of perfused organs, classify the data segments based on the features extracted and based on the FPC model, and provide, as an output, a prediction indicative of a severity of vessel obstruction based on the classification of the features.
1. A method for assessing a severity of vessel obstruction, comprising:
a) obtaining a contrast enhanced volume image dataset for a heart;
b) segmenting a myocardium of the heart from the contrast enhanced volume data set of a);
c) determining at least one geometrical feature of an inner vessel wall of at least one blood vessel surrounding the myocardium of the heart;
d) determining a value of at least one hemodynamic parameter based on the at least one geometrical feature of c);
e) defining data segments characterizing the myocardium of the heart of b);
f) generating feature vectors related to the data segments of e), wherein the feature vectors include the at least one hemodynamic parameter of d);
g) classifying the data segments of e) based on the feature vectors of f) and a classification model; and
h) providing, as an output, a prediction indicative of a severity of vessel obstruction based on the classifying of g);
wherein the classifying of g) uses a machine learning system to classify the feature vectors of f), wherein the machine learning system is trained from a database of contrast enhanced volume image data sets and associated training feature vectors extracted from the contrast enhanced volume image data sets.
2. The method of claim 1 , wherein:
the feature vectors of f) are generated by inputting patches of myocardial voxels to an additional machine learning system that generates data encodings that characterize the myocardium of the heart, and applying an automated clustering method to the data encodings.
3. The method of claim 2 , wherein:
the additional machine learning system uses at least one of: a convolutional auto-encoder, Gaussian filters, transmural perfusion ratio, Haralick features, myocardium thickness, myocardium volume, ventricular volume and organ shape.
4. The method of claim 2 , wherein:
the automated clustering method is based on voxels representing coronary tree anatomy.
5. The method of claim 1 , wherein:
the machine learning system represents a relationship between the training feature vectors and at least one reference parameter, wherein the at least one reference parameter is selected from the group consisting of i) an invasive fractional flow reserve measurement, ii) an index of microcirculatory resistance, iii) a coronary flow reserve measurement, iv) occurrence of major adverse cardiac events (MACE) within a predefined amount of time after acquisition of the contrast enhanced volume image dataset, v) occurrence of revascularization within a predefined amount of time after acquisition of the contrast enhanced volume image dataset, vi) the results of a cardiac stress test, and vii) the results of myocardial magnetic resonance imaging (MRI) perfusion, SPECT, PET, CT perfusion, or ultrasound.
6. The method of claim 1 , wherein:
the machine learning system is based on a supervised machine learning algorithm.
7. The method of claim 6 , wherein:
the supervised machine learning algorithm is selected from the group consisting of a support vector machine, a neural network, a Bayesian classifier, and a Tree Ensemble.
8. The method of claim 1 , wherein:
the feature vectors off) include data indicative of an amount of perfusion experienced by the myocardium of the heart.
9. The method of claim 1 , wherein:
the feature vectors of f) include additional information based on a patient ECG signal.
10. The method of claim 1 , wherein:
the at least one geometrical feature of c) represents a diameter or cross-section area of the inner vessel wall.
11. The method of claim 1 , wherein:
the at least one hemodynamic parameter of d) represents a pressure-drop or pressure gradient of a coronary vessel surrounding the myocardium.
12. The method of claim 1 , wherein:
the at least one hemodynamic parameter of d) represents at least one hemodynamic index of a coronary vessel surrounding the myocardium.
13. The method of claim 12 , wherein:
the at least one hemodynamic index relates to fractional flow reserve, coronary flow reserve, instantaneous wave-free ratio, hyperemic myocardium perfusion, index of microcirculatory resistance or other hemodynamic parameter along a coronary vessel surrounding the myocardium.
14. The method of claim 1 , wherein:
the feature vectors of f) represent variation in a characteristic of interest over voxels within clusters of myocardial voxels; and/or
the feature vectors of f) are derived by calculating variation factors for the data encodings; and/or
the variation factor for a given data encoding represents a deviation of a characteristic of interest over all clusters for the given data encoding; and/or
the feature vectors of f) are derived from factors representing a characteristic of interest over multiple segments of the myocardium of the heart; and/or
the characteristic of interest represents a mean intensity computed by a multidimensional gaussian operator; and/or
the feature vectors of f) include additional information selected from the group consisting of: i) information that characterizes global features of the entire myocardium of the heart, ii) patient demographic information, iii) information that characterizes the presence or amount or type of coronary artery calcification or plaque, iv) information characterizing tissue composition or tissue type or contrast agent, v) information characterizing myocardium layering or myocardium layer, vi) information characterizing an ECG signal parameter, vii) information pertaining to a cardiac biomarker in blood such as cardiac troponin or creatine kinase, viii) information that characterizes fat surrounding the heart or inside the heart, ix) information that characterizes shape of the myocardium, x) information that characterizes myocardial strain, xi) information that characterizes blood volume or blood pressure, xii) information that characterizes ejection fraction, xiii) information that characterizes cardiac output, and xiv) information that characterizes parts of the heart, ascending aortic or coronary tree.
15. The method of claim 1 , wherein:
the operations of a) to h) are performed by at least one processor.
16. A system for assessing vessel obstruction, comprising:
memory configured to store a contrast enhanced volume image dataset for a heart; and
one or more processors that, when executing program instructions stored in the memory, are configured to perform the method of claim 1 .