Automated intravascular plaque classification

• US 10,453,191 B2
• Filed: 04/19/2017
• Issued: 10/22/2019
• Est. Priority Date: 04/20/2016
• Status: Active Grant

• Current Assignee: Case Western Reserve University
• Original Assignee: Case Western Reserve University
• Inventors: Shalev, Ronny Y., Wilson, David L., Ray, Soumya
• Primary Examiner(s): Conner, Sean M
• Application Number:
  US15/491,544
• Publication Number:
  US 20170309018A1
• Time in Patent Office: 916 Days
• Field of Search: None
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06T 2207/10101 Optical tomography; Optical coherence tomography [OCT]
• CPC Class Codes: G06T 2207/30101 Blood vessel; Artery; Vein; Vascular
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06V 10/507 Summing image-intensity values; Histogram projection analysis
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images

Abstract

Methods and apparatus automatically classify intravascular plaque using features extracted from intravascular optical coherence tomography (IVOCT) imagery. One example apparatus includes an image acquisition circuit that accesses a set of IVOCT images, a pre-processing circuit that generates a blood vessel mask based on the IVOCT images, a feature extraction circuit that defines a three dimensional (3D) volume of interest centered on a location in a member of the set of IVOCT images, a classification circuit that generates a classification based on a probability that a voxel represents a type of plaque, and a visualization circuit that provides a visualization, substantially in real time, of a member of the set of IVOCT images and the classification, where the visualization includes a sector classification image, a labeled image, or a 3D visualization. A prognosis or treatment plan may be provided based on the visualization or the classification.

Claims

1. A non-transitory computer-readable storage device storing computer executable instructions that when executed by a computer control the computer to perform a method for intravascular plaque classification, the method comprising:

accessing an intravascular optical coherence tomography (IVOCT) image of a region of tissue, where the IVOCT image includes a plurality of voxels, where the region of tissue includes a blood vessel;

generating a pre-processed image by pre-processing the IVOCT image, where generating the pre-processed image includes generating a blood vessel mask;

defining a moving box (mBox) based on the pre-processed image and the blood vessel mask, where the mBox defines a three-dimensional (3D) volume of interest (VOI) centered on a location in a view of the pre-processed image, where the mBox includes at least one member of the plurality of voxels, where a scanning range of the mBox is based, at least in part, on the blood vessel mask;

extracting a set of local features from the at least one member of the plurality of voxels;

providing the set of local features to a machine learning classifier, where the machine learning classifier includes a plurality of one-versus-the-rest (OVR) support vector machine (SVM)s, where the plurality of OVR SVMs includes an OVR-lipid (OVR-L) classifier, an OVR-calcium (OVR-C) classifier, and an OVR-fibrous (OVR-F) classifier, where a member of the plurality of OVR SVMs classifies a voxel based on a probability threshold and the set of local features, where the probability threshold is computed using a minimum squared distance (MSD) approach, or a Youden index approach;

receiving, from the machine learning classifier, a classification of the at least one member of the plurality of voxels;

generating an annotated image of the region of tissue based on the classification; and

controlling a percutaneous coronary intervention (PCI) apparatus to display the annotated image.

2. The non-transitory computer-readable storage device of claim 1:

where the IVOCT image is acquired during a PCI pullback;

where the IVOCT image is acquired with a 1310 nm center wavelength, a 110 nm wavelength range, a 50 kHz sweep rate, a 20 mW output power, or a 12 mm coherence length.

3. The non-transitory computer-readable storage device of claim 2, where the IVOCT image is acquired with a pullback speed of 20 mm/s, a pullback length of 54 mm, where the pullback includes at least 270 image frames spaced at least 200 μ

m apart.

4. The non-transitory computer-readable storage device of claim 1, where pre-processing the IVOCT image includes:

generating a noise reduced image based on the IVOCT image;

subtracting a baseline intensity from the noise reduced image;

correcting the noise reduced image using optical image system correction;

generating a segmented image by segmenting a lumen and a back border from the noise reduced image;

generating a lumen border by correcting for catheter eccentricity with respect to a vessel wall, based on the segmented image;

defining a guide wire shadow in the segmented image;

generating the blood vessel mask based on the guide wire shadow, the lumen border, and the back border; and

linearizing an exponential attenuation model of an IVOCT signal.

5. The non-transitory computer-readable storage device of claim 1, where the set of local features includes an optical feature set, an intensity variation feature set, or a spatial texture feature set.

6. The non-transitory computer-readable storage device of claim 5, where the optical feature set includes an attenuation coefficient feature, a reflected intensity feature, a distance to lumen feature, a beam penetration depth feature, a beam incident angle feature, or a mean intensity feature.

7. The non-transitory computer-readable storage device of claim 5, where the intensity variation feature set includes a homogeneity feature, a relative smoothness of intensity feature, or an entropy feature.

8. The non-transitory computer-readable storage device of claim 5, where the spatial texture feature set includes a contrast feature, a correlation feature, an energy feature, or a co-occurrence homogeneity feature.

9. The non-transitory computer-readable storage device of claim 1, where displaying the annotated image includes displaying, substantially in real time:

a sector classification image, where the sector classification image includes a masked frame of the region of tissue, where the masked frame includes a plurality of sectors, where a sector includes a representation of a type of plaque detected in the sector displayed at the perimeter of the sector;

a labeled image of the region of tissue;

or

a three dimensional (3D) volumetric visualization of the region of tissue, where the 3D volumetric visualization includes an overlay that indicates a plaque type within the region of tissue.

10. The non-transitory computer-readable storage device of claim 1, the method further comprising training the OVR-L classifier, the OVR-C classifier, or the OVR-F classifier.

11. The non-transitory computer-readable storage device of claim 10, where training the OVR-L classifier, the OVR-C classifier, or the OVR-F classifier includes:

accessing a clinical training data set, where the clinical training data set includes a set of IVOCT images acquired during a PCI pullback;

training the OVR-L classifier, the OVR-C classifier, or the OVR-F classifier using the clinical training data set;

accessing an independent validation set, where the independent validation set includes a set of voxels labeled using a 3D cryo-imaging approach;

testing the OVR-L classifier, the OVR-C classifier, or the OVR-F classifier using the independent validation set; and

optimizing a set of classifier parameters based on a 5-fold stratified cross validation of the clinical training data set.

12. The non-transitory computer-readable storage device of claim 1, the method further comprising generating a post-processed image by applying a sized-based filter to the annotated image.

13. An apparatus, comprising:

a processor;

a memory;

an input/output interface;

a set of circuits, where the set of circuits includes an image acquisition circuit, a pre-processing circuit, a feature extraction circuit, a classification circuit, and a visualization circuit; and

an interface to connect the processor, the memory, the input/output interface and the set of circuits;

where the image acquisition circuit accesses a set of intravascular optical coherence tomography (IVOCT) images of a region of tissue, where a member of the set of IVOCT images includes a plurality of voxels, where a member of the plurality of voxels includes a set of discriminative features, where the region of tissue includes a blood vessel;

where the pre-processing circuit generates a blood vessel mask based on the set of IVOCT images, where the pre-processing circuit provides the blood vessel mask to the feature extraction circuit;

where the feature extraction circuit defines a moving box (mBox), where the mBox defines a three dimensional (3D) volume of interest centered on a location in the member of the set of IVOCT images, where the mBox includes at least one member of the plurality of voxels, where the mBox is defined by a set of dimensions and the blood vessel mask, where the feature extraction circuit extracts the set of discriminative features from the at least one member of the plurality of voxels, and where the feature extraction circuit provides the set of discriminative features to the classification circuit;

where the classification circuit generates a classification of the voxel as a type of plaque based on a probability that the voxel represents the type of plaque, where the probability is based, at least in part, on the set of discriminative features, where the classification circuit includes a plurality of one-versus-the-rest (OVR) support vector machines (SVM)s, where a first member of the plurality of OVR SVMs is an OVR-lipid (OVR-L) classifier, where a second, different member of the plurality of OVR SVMs is an OVR-calcium (OVR-C) classifier, and where a third, different member of the plurality of OVR SVMs is an OVR-fibrous (OVR-F) classifier, where a member of the plurality of OVR SVMs classifies a voxel based on a probability threshold and the set of discriminative features, where the classification circuit computes the probability threshold using a minimum squared distance (MSD) approach, or a Youden index approach, and where the OVR-L classifier, the OVR-C classifier, or the OVR-F classifier employs a regularization parameter and a radial basis function (RDF) Gaussian kernel; and

where the visualization circuit generates a visualization of a member of the set of IVOCT images and the classification, where the visualization includes a sector classification image, a labeled image, or a 3D visualization, and where the visualization circuit provides the visualization to a display, substantially in real time.

14. The apparatus of claim 13, where the image acquisition circuit acquires the set of IVOCT images during a percutaneous coronary intervention (PCI) pullback, where the PCI pullback has a pullback speed of 20 mm/s, a pullback length of 54 mm, and where the pullback includes at least 270 image frames spaced at least 200 μ

m apart, and

where a member of the set of IVOCT images is acquired using a 1310 nm center wavelength, a 110 nm wavelength range, a 50 kHz sweep rate, a 20 mW output power, or a 12 mm coherence length.

15. The apparatus of claim 13, where the set of discriminative features includes an optical feature set, an intensity variation feature set, or a spatial texture feature set,

where the optical feature set includes an attenuation coefficient feature, a reflected intensity feature, a distance to lumen feature, a beam penetration depth feature, a beam incident angle feature, or a mean intensity feature,

where the intensity variation feature set includes a homogeneity feature, a relative smoothness of intensity feature, or an entropy feature, and

where the spatial texture feature set includes a contrast feature, a correlation feature, an energy feature, or a co-occurrence homogeneity feature.

16. The apparatus of claim 13, where the set of circuits further comprises a training circuit that:

trains the classification circuit using a training set of images, where the training circuit accesses the training set or a set of independent testing images, where a member of the training set represents a region of tissue demonstrating lipid plaque, calcified plaque, or fibrous plaque;

tests the classification circuit using the set of independent testing images, where a member of the set of independent testing images includes a volume of interest labeled as lipid plaque, calcified plaque, or fibrous plaque using 3D cryo-imaging; and

optimizes the regularization parameter and the RDF Gaussian kernel'"'"'s bandwidth.

17. A non-transitory computer-readable storage device storing computer executable instructions that when executed by a computer control the computer to perform a method, the method comprising:

accessing a set of intravascular optical coherence tomography (IVOCT) images of a region of tissue demonstrating intravascular plaque, where the set of IVOCT images is acquired from a patient during a percutaneous coronary intervention (PCI), where a member of the set of IVOCT images includes at least one voxel, where the region of tissue includes a blood vessel;

generating a blood vessel mask by preprocessing the set of IVOCT images;

defining a volume of interest (VOI) based on the set of IVOCT images and the blood vessel mask, where the VOI includes a member of the at least one voxels;

extracting a set of discriminatory features from the member of the at least one voxels, where the set of discriminatory features includes an optical feature, an intensity variation feature, or a spatial texture feature;

providing the set of discriminatory features to a plurality of support vector machines (SVMs), where the plurality of support vector machines includes a one-versus-rest lipid (OVR-L) SVM, an OVR-fibrous (OVR-F) SVM, and an OVR-calcium (OVR-C) SVM;

receiving, from the plurality of SVMs, a classification of the voxel;

generating a visualization of the blood vessel based on the IVOCT and the classification, where the visualization includes a sector classification, a labeled image, or a 3D visualization;

displaying the visualization; and

generating a personalized treatment plan for the patient, based, at least in part, on the visualization and the classification.