Systems and methods for automated coronary plaque characterization and risk assessment using intravascular optical coherence tomography
Exemplary embodiments of the present disclosure include apparatus and methods to classify the plaque tissue present in the coronary artery using intravascular optical coherence tomography (IVOCT) images.
1. A system comprising:
an imaging device comprising an optical coherence tomography light source, wherein the imaging device is configured to obtain an image of intravascular tissue comprising plaque; and
a non-transitory computer readable medium configured to:
analyze a pixel of the image with a first neural network configured to classify the plaque as a first tissue type of a plurality of tissue types;
analyze the pixel of the image with a second neural network configured to classify the plaque as a second tissue type of the plurality of tissue types; and
analyze the pixel of the image with a third neural network configured to classify the plaque as a third tissue type of the plurality of tissue types, wherein the non-transitory computer readable medium is configured to optimize the first, second and third neural networks by evaluating a plurality of features of the image with nodes of the first, second and third neural networks to calculate sensitivity and specificity of the plurality of features using a receiver operating characteristic (ROC) curve.
2. The system of claim 1 wherein histological data from the plurality of tissue types is analyzed to characterize tissue types of pixels selected to train the first, second and third neural networks.
3. The system of claim 1 wherein the first tissue type is lipid plaque, the second tissue type is a calcific plaque, and the third tissue type is a fibrous plaque.
4. The system of claim 1 wherein the plurality of features comprise one or more of the following Gray Level Co-Occurrence Matrix (GLCM) features: contrast, energy, correlation, homogeneity, entropy, and maximum probability.
5. The system of claim 1 wherein the plurality of features comprise one or more of the following two-dimensional image statistics: mean value, variance, skewness, kurtosis, and energy.
6. The system of claim 1 wherein the optical coherence tomography light source is configured as a swept source optical coherence tomography light source.
7. The system of claim 1 wherein the optical coherence tomography light source is configured as a broadband optical coherence tomography light source.
8. The system of claim 1 wherein the imaging device further comprises a short pulsed excitation light source.
9. The system of claim 8 wherein the short pulsed excitation light source is a two photon luminescence light source.
10. The system of claim 8 wherein the imaging device further comprises a photonic crystal fiber configured to simultaneously:
enable single-mode propagation of a first wavelength from the optical coherence tomography light source to a sample site;
enable single-mode propagation of a second wavelength from the short-pulsed light source to the sample site;
transmit an optical coherence tomography signal from the sample site, wherein the optical coherence tomography signal is generated from the first wavelength; and
transmit an emission signal from the sample site, wherein the emission signal is induced by the second wavelength from the short-pulsed light source.
11. The system of claim 10 further comprising a first dichroic element, wherein the first dichroic element is configured to direct the first and second wavelengths to the sample path.
12. The system of claim 10 further comprising a second dichroic element.
13. The system of claim 12 wherein the second dichroic element is configured to direct two photon luminescence toward a photon counting detector.
14. The system of claim 8 further comprising a balanced detector, wherein the balanced detector is configured to minimize a non-interfering OCT component.
15. The apparatus of claim 1 further comprising a photon counting detector.
16. The apparatus of claim 15 wherein the photon counting detector is a photomultiplier tube, or an avalanche photo diode.
17. The apparatus of claim 15 wherein the photon counting detector is configured to detect two-photon luminescence.
18. A method of characterizing coronary plaque, the method comprising:
obtaining an image of a sample site using an optical coherence tomography light source emitting light from an optical fiber, wherein the image comprises intravascular tissue comprising plaque;
analyzing quantitative data of a pixel of the image with a first neural network configured to classify the plaque as a first tissue type of a plurality of tissue types, wherein the first neural network comprises a first plurality of nodes and reads a first plurality of features;
analyzing quantitative data of the pixel of the image with a second neural network configured to classify the plaque as a second tissue type of the plurality of tissue types, wherein the second neural network comprises a second plurality of nodes and reads a second plurality of features; and
analyzing quantitative data of the pixel of the image with a third neural network configured to classify the plaque as a third tissue type of the plurality of tissue types, wherein the third neural network comprises a third plurality of nodes and reads a third plurality of features, wherein the quantitative data includes classifying features comprising one or more of the following: contrast, energy, correlation, homogeneity, entropy, and maximum probability.
19. The method of claim 18 wherein histological data from the plurality of tissue types is analyzed to characterize tissue types of pixels selected to train the first, second and third neural networks.
20. The method of claim 18 wherein the first tissue type is lipid plaque, the second tissue type is a calcific plaque, and the third tissue type is a fibrous plaque.
21. The method of claim 18 wherein the plurality of features comprise one or more of the following two-dimensional image statistics: mean value, variance, skewness, kurtosis, and energy.
22. The method of claim 18 further comprising:
optimizing the first, second and third neural networks by calculating a receiver operating characteristic (ROC) curve which plots a true positive versus a false positive rate for a plurality of classifying features of the image; and
calculating an area under each receiver operating characteristic (ROC) curve for each of the plurality of classifying features.
23. The method of claim 22 further comprising:
ranking the plurality of classifying features by the area under each receiver operating characteristic (ROC) curve for each of the plurality of classifying features; and
calculating a sensitivity and a specificity of the classifying features for the first, second and third neural networks.
24. The method of claim 23 wherein the sensitivity is a proportion of known plaque type data points that are correctly classified by each of the first, second and third neural networks.
25. The method of claim 23 wherein the specificity is a ratio of correct classifications to total classifications for a certain category of plaque tissue types for each of the first, second and third neural networks.
26. The method of claim 23 wherein each of the first, second and third neural networks is optimized by selecting a combination of nodes and classifying features for each of the first, second and third neural networks that result in the highest value of a sum of the specificity and sensitivity.