Method of and system for calcium scoring of coronary arteries
A method of automatically determining a calcium score for at least one coronary artery is disclosed. The method comprises receiving cardiac non-contrast CT data indicative of a cardiac non-contrast CT scan carried out on a patient, analysing the cardiac non-contrast CT data in a calcified components identifier to detect candidate coronary artery calcified components, and analysing cardiac non-contrast CT data associated with the candidate coronary artery calcified components using a radiomics analyser to determine radiomic characteristics of the candidate coronary artery calcified components. The method also comprises applying machine learning to the determined radiomic characteristics associated with each candidate coronary artery calcified component to identify any calcifications that are located on a coronary artery, analysing the cardiac non-contrast CT data to identify at least one body component in the cardiac non-contrast CT data not associated with a coronary artery of the patient, and using the identified at least one body component in the cardiac non-contrast CT data to remove or avoid misclassification of calcifications on a coronary artery that are located on the at least one identified body component.
1. A method of automatically determining a calcium score for at least one coronary component, the method comprising:
receiving cardiac non-contrast CT data indicative of a cardiac non-contrast CT scan carried out on a patient;
analysing the cardiac non-contrast CT data in a calcified components identifier to detect at least one candidate coronary calcified component associated with at least one target coronary anatomical structure;
analysing cardiac non-contrast CT data associated with the at least one candidate coronary calcified component using a radiomics analyser to determine radiomic characteristics of the at least one candidate coronary calcified component;
applying machine learning to the determined radiomic characteristics associated with each candidate coronary calcified component to identify any calcifications that are located on the at least one target coronary anatomical structure;
analysing the cardiac non-contrast CT data to identify at least one body component in the cardiac non-contrast CT data not associated with the at least one target coronary anatomical structure; and
after identifying any calcifications that are located on a coronary artery, using the identified at least one body component in the cardiac non-contrast CT data to remove misclassification of calcifications on the at least one coronary target anatomical structure that are located on the at least one identified body component.
2. The method as claimed in claim 1 , wherein the at least one target coronary anatomical structure comprises at least one coronary artery.
3. The method as claimed in claim 2 , comprising using machine learning to analyse the cardiac non-contrast CT data to identify at least one body component in the cardiac non-contrast CT data not associated with a coronary artery of the patient.
4. The method as claimed in claim 2 , comprising:
analysing the cardiac non-contrast CT data to identify aortic components in the cardiac non-contrast CT data associated with an aorta of the patient; and
using the identified aortic components of the cardiac non-contrast CT data to remove misclassification of calcifications on a coronary artery that are located on the aortic components.
5. The method as claimed in claim 4 , wherein the step of analysing the cardiac non-contrast CT data to identify aortic components of the cardiac non-contrast CT data associated with an aorta of the patient comprises analysing the identified aortic components using size, shape and position of the identified aortic components.
6. The method as claimed in claim 4 , wherein the step of analysing the cardiac non-contrast CT data to identify aortic components in the cardiac non-contrast CT data associated with an aorta of the patient comprises:
progressively processing single slices of the cardiac non-contrast CT data, and assembling the results of a plurality of individual slices into a volumetric segmentation; and/or
processing volumetric inputs or cross-hair type orthogonal inputs.
7. The method as claimed in claim 2 comprising:
analysing the cardiac non-contrast CT data to identify a cardiac region of interest (ROI) around a heart in the cardiac non-contrast CT data; and
using the identified cardiac ROI to remove misclassification of calcifications on a coronary artery that are located outside the cardiac ROI.
8. The method as claimed in claim 2 , wherein the determined radiomic characteristics include position, shape, size and/or density.
9. The method as claimed in claim 2 , comprising using a first classifier to classify each candidate coronary artery calcified component as located on a coronary artery or not located on a coronary artery, and a second classifier to identify each coronary artery.
10. The method as claimed in claim 2 , comprising analysing the cardiac non-contrast CT data indicative of the candidate coronary artery calcified components to determine image patch data associated with a region of the cardiac non-contrast CT data around each candidate coronary artery calcified component, and applying machine learning to the determined image patch data to identify any calcifications that are located on a coronary artery.
11. The method as claimed in claim 10 , comprising using a hybrid neural network to combine the output of the step of applying machine learning to the determined image patch data to identify any calcifications that are located on a coronary artery using a convolutional neural network, and the determined radiomic characteristics associated with each candidate coronary artery calcified component.
12. The method as claimed in claim 2 , comprising directly applying machine learning to the cardiac non-contrast CT data indicative of the candidate coronary artery calcified components to identify any calcifications that are located on a coronary artery.
13. The method as claimed in claim 12 , comprising using outputs of the directly applied machine learning and outputs of the step of applying machine learning to the determined radiomic characteristics to identify any calcifications that are located on a coronary artery.
14. The method as claimed in claim 2 , comprising:
analysing the cardiac non-contrast CT data to identify a mitral valve in the cardiac non-contrast CT data, and using the identified mitral valve to remove misclassification of calcifications; and/or
analysing the cardiac non-contrast CT data to identify a heart in the cardiac non-contrast CT data, and using the identified heart to remove misclassification of calcifications on a coronary artery that are located outside the heart.
15. The method as claimed in claim 2 , comprising adding calibration markers manually to the cardiac non-contrast CT data and using the added markers to provide the machine learning with positional information.
16. A system for automatically determining a calcium score for at least one coronary component, the system comprising:
a calcified components identifier for analysing received cardiac non-contrast CT data indicative of a cardiac non-contrast CT scan carried out on a patient to detect at least one candidate coronary calcified component associated with at least one target coronary anatomical structure;
a radiomics analyser for analysing cardiac non-contrast CT data associated with the at least one candidate coronary calcified component to determine radiomic characteristics of the at least one candidate coronary calcified component;
a radiomic machine learning component arranged to apply machine learning to the determined radiomic characteristics associated with each candidate coronary calcified component to identify any calcifications that are located on the at least one target coronary anatomical structure;
a body component identifier arranged to analyse the cardiac non-contrast CT data after any calcifications have been located on a coronary artery to identify at least one body component in the cardiac non-contrast CT data not associated with the at least one target coronary anatomical structure; and
a misclassification remover that uses the identified at least one body component in the cardiac non-contrast CT data to remove misclassification of calcifications on the at least one coronary target anatomical structure that are located on the at least one identified body component.
17. The system as claimed in claim 16 , wherein the at least one target coronary anatomical structure comprises at least one coronary artery.
18. The system as claimed in claim 17 , wherein the body component analyser uses machine learning to analyse the cardiac non-contrast CT data to identify at least one body component in the cardiac non-contrast CT data not associated with a coronary artery of the patient.
19. The system as claimed in claim 17 , comprising:
an aorta identifier for analysing the cardiac non-contrast CT data to identify aortic components in the cardiac non-contrast CT data associated with an aorta of the patient;
the system using the identified aortic components of the cardiac non-contrast CT data to remove misclassification of calcifications on a coronary artery that are located on the aortic components.
20. The system as claimed in claim 17 , comprising:
a cardiac region of interest (ROI) identifier for analysing the cardiac non-contrast CT data to identify a cardiac region of interest (ROI) around a heart in the cardiac non-contrast CT data; and
the system using the identified cardiac ROI used to remove misclassification of calcifications on a coronary artery that are located outside the cardiac ROI.
21. The system as claimed in claim 17 , wherein the determined radiomic characteristics include position, shape, size and/or density.
22. The system as claimed in claim 17 , wherein the calcified components identifier comprises an image patch analyser arranged to analyse the cardiac non-contrast CT data indicative of the candidate coronary artery calcified components to determine image patch data associated with a region of the cardiac non-contrast CT data around each candidate coronary artery calcified component, and apply machine learning to the determined image patch data to identify any calcifications that are located on a coronary artery.
23. The system as claimed in claim 22 , comprising a hybrid neural network for combining an output of the image patch convolutional neural network with an output of the radiomic machine learning component.
24. A method of automatically determining a calcium score for at least one coronary artery, the method comprising:
receiving cardiac non-contrast CT data indicative of a cardiac non-contrast CT scan carried out on a patient;
analysing the cardiac non-contrast CT data in a calcified components identifier using machine learning to identify candidate calcifications located on the at least one coronary artery;
analysing the cardiac non-contrast CT data to identify aortic components in the cardiac non-contrast CT data associated with an aorta of the patient; and
using the identified aortic components in the cardiac non-contrast CT data to remove misclassification of calcifications on the at least one coronary artery that are located on the aortic components.
25. The method as claimed in claim 24 , wherein the method comprises analysing the cardiac non-contrast CT data to identify ascending and descending portions of the aorta.
26. The method as claimed in claim 25 , comprising using machine learning to predict whether each voxel of the cardiac non-contrast CT data is part of the ascending or descending aorta and produce candidate aorta voxels.
27. The method as claimed in claim 26 , comprising applying a connected component analysis to the candidate aorta voxels to identify neighbouring voxels that belong to the same aortic component.