Dual energy x-ray coronary calcium grading

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 predicting coronary heart disease, the method comprising:

acquiring a set of energy dependent images of a region of tissue, where the set of images includes a first energy x-ray image and a second, different energy x-ray image;

generating a set of rib suppressed images based on the set of energy dependent images, where the set of rib suppressed images includes a first energy rib suppressed image and a second energy rib suppressed image;

generating a calcium-enhanced image of the region of tissue, where generating the calcium-enhanced image includes:

generating a set of pre-decomposition images by pre-decomposition processing the set of energy dependent images, where generating the set of pre-decomposition images includes registering the set of energy dependent images or the set of rib suppressed images using anatomy aware registration;

generating a set of selectively filtered images by selectively filtering the set of pre-decomposition images; and

generating a set of dual energy subtraction images by dual energy decomposing the set of selectively filtered images;

generating a post-decomposition image based, at least in part, on the calcium-enhanced image;

computing a coronary calcium score based, at least in part, on the post-decomposition image; and

computing a coronary heart disease probability based, at least in part, on the coronary calcium score;

where registering the set of energy dependent images or rib suppressed images using anatomy aware registration includes:

accessing a set of training chest x-ray images of a region of tissue, where the set of training chest x-ray images includes a plurality of chest x-ray images;

defining a set of key points based on the set of training chest x-ray images, where the set of key points is defined by annotating a set of anatomical landmarks represented in a member of the set of training chest x-ray images, where the set of key points identifies a set of anatomical landmarks common to at least two members of the plurality of chest x-ray images;

computing a model based on the set of key points, where the model describes a shape of a heart represented in the set of training chest x-ray images;

generating a set of corresponding locations by detecting, in the first energy x-ray image and the second, different energy x-ray image, or in the first energy rib suppressed image and the second energy rib suppressed image, a location corresponding to an anatomical landmark represented by the set of key points;

generating a set of deformation vectors by computing, for a member of the set of corresponding locations, a deformation vector;

interpolating a deformation field based, at least in part, on the set of deformation vectors, where the deformation field is defined using a B-spline interpolation; and

registering the first energy x-ray image and the second, different energy x-ray image, or the first energy rib suppressed image and the second energy rib suppressed image, using the deformation field.

2. The non-transitory computer-readable storage device of claim 1 , where the first energy x-ray image is a 60 kV exposure and the second, different energy x-ray image is a 120 kV exposure, where the first energy x-ray image and the second, different energy x-ray image are acquired within a 150 ms interval.

3. The non-transitory computer-readable storage device of claim 1 , where acquiring the set of energy dependent images includes creating a high x-ray spectrum and a low x-ray spectrum by switching kV between a first, high kV value, and a second, lower kV value, where the first, high kV value is 120 kV, and the second, lower kV value is 60 kV.

4. The non-transitory computer-readable storage device of claim 1 , where acquiring the set of energy dependent images includes creating a high x-ray spectrum and a low x-ray spectrum by switching kV and mA, by switching pre-filtration x-ray filters, or by switching pre-filtration x-ray filters and kV.

5. The non-transitory computer-readable storage device of claim 1 , where the set of energy dependent images is acquired using a photon counting, energy sensitive detector, or by using a sandwich detector comprising a first detector and a second different detector, where the first energy x-ray image is acquired using the first detector, and the second energy x-ray image is acquired using the second detector.

6. The non-transitory computer-readable storage device of claim 1 , where generating the set of pre-decomposition images includes normalizing the set of energy dependent images, where normalizing the set of energy dependent images includes normalizing the first energy x-ray image, or correcting the second energy x-ray image.

7. The non-transitory computer-readable storage device of claim 1 , where generating the set of rib suppressed images includes:

generating a bone image of the region of tissue based on the set of energy dependent images, where the bone image includes a plurality of pixels;

computing a rib edge map based on the bone image, where the rib edge map is computed using an edge filter, or where the rib edge map is computed using a convolutional neural network trained on a set of rib-annotated images using a supervised machine learning process, where the rib edge map displays a probability of rib edge occurrence at a pixel within the bone image, and where the rib edge map has an intensity;

generating a binary segmentation rib mask based on the rib edge map by integrating the rib edge map intensity and a prior rib shape model, where the prior rib shape model includes a single rib model and an inter-rib model, and where the binary segmentation rib mask defines a rib area and a non-rib area in the bone image;

estimating, using the binary segmentation rib mask and independent component analysis (ICA), a first rib signal associated with an area in the first energy x-ray image corresponding to the rib area, and a second, different rib signal associated with an area in the second energy x-ray image corresponding to the rib area; and

creating the first energy rib suppressed image by subtracting the first rib signal from the first energy x-ray image, or creating the second energy rib suppressed image by subtracting the second rib signal from the second energy x-ray image.

8. The non-transitory computer-readable storage device of claim 7 , where generating the set of pre-decomposition images includes registering the set of energy dependent images or the set of rib suppressed images using rigid registration, piece-wise registration, perspective registration, non-rigid registration, or sliding registration.

9. The non-transitory computer-readable storage device of claim 8 , the method further comprising generating a binary mask that defines an area inside a heart silhouette and an area outside the heart silhouette, where generating the binary mask includes:

computing an initial heart silhouette based on the first energy x-ray image or the first energy rib suppressed image, where the initial heart silhouette is computed using dynamic programming;

detecting a deformed heart silhouette in the second energy x-ray image or the second energy rib suppressed image; and

generating an enlarged heart silhouette by dilating the initial heart silhouette using a margin, where the enlarged heart silhouette covers the deformed heart silhouette.

10. The non-transitory computer-readable storage device of claim 9 , where registering the set of energy dependent images or the set of rib suppressed images using sliding registration includes:

segmenting the first energy x-ray image or the first energy rib suppressed image into a first inside region and a first outside region using the binary mask, where the first inside region defines an area of the first energy x-ray image or the first energy rib suppressed image inside the heart silhouette defined by the binary mask, and the first outside region defines an area of the first energy x-ray image or the first energy rib suppressed image outside the heart silhouette defined by the binary mask;

segmenting the second energy x-ray image or the second energy rib suppressed image into a second inside region and a second outside region using the binary mask, where the second inside region defines an area of the second energy x-ray image or the second energy rib suppressed image inside the heart silhouette defined by the binary mask, and the second outside region defines an area of the second energy x-ray image or the second energy rib suppressed image outside the heart silhouette defined by the binary mask;

generating a registered inside region by registering the first inside region with the second inside region using free form deformation (FFD) registration;

generating a registered outside region by registering the first outside region with the second outside region using rigid deformation registration; and

stitching the registered inside region with the registered outside region using soft weighting.

11. The non-transitory computer-readable storage device of claim 10 , where registering the first inside region with the second inside region using FFD registration includes parameterizing a motion field between the first energy x-ray image and the second energy x-ray image, or between the first energy rib suppressed image and the second energy rib suppressed image, by interpolating a series of B-spline basis functions with a mesh of uniform spacing control points, and by minimizing a masked FFD registration cost function, where the masked FFD registration cost function is masked using the binary mask.

12. The non-transitory computer-readable storage device of claim 11 , where generating the set of selectively filtered images includes filtering the first energy x-ray image or the first energy rib suppressed image using a Gaussian filter.

13. The non-transitory computer-readable storage device of claim 11 , where generating the set of selectively filtered images includes filtering the second energy x-ray image or the second energy rib suppressed image using a vesselness operator using morphological suppression, or filtering the second energy x-ray image or the second energy rib suppressed image using a directional low pass filter.

14. The non-transitory computer-readable storage device of claim 13 , where filtering the second energy x-ray image or the second energy rib suppressed image using a directional low pass filter includes:

generating a set of directional band-pass filters;

producing a vector of filter responses by convolving the second energy x-ray image or the second energy rib suppressed image with the set of directional band-pass filters, where a filter response is associated with a channel;

producing a transformed vector of filter responses by multiplying a filter response in the vector of filter responses by a weighting coefficient, where the weighting coefficient amplifies a selective width associated with the filter response or a selective direction associated with the filter response, or where the weighting coefficient suppresses the selective width associated with the filter response or the selective direction associated with the filter response; and

filtering the second energy x-ray image or the second energy rib suppressed image based, at least in part, on the transformed vector of filter responses.

15. The non-transitory computer-readable storage device of claim 1 , where generating the set of dual energy subtraction images includes correcting the set of selectively filtered images by removing scatter from a member of the set of selectively filtered images.

16. The non-transitory computer-readable storage device of claim 1 , where generating the post-decomposition image includes applying un-sharp masking to the set of dual energy subtraction images, where the un-sharp masking is based, at least in part, on the set of dual energy subtraction images, an un-sharp coefficient, and a low-pass filtering operation, where the value of the un-sharp coefficient is within [0, 1).

17. The non-transitory computer readable storage device of claim 1 , where generating the post-decomposition image includes selectively adjusting the dynamic range of the set of dual energy subtraction images to match, within a threshold amount, the dynamic range of a display device format or a storage device format.

18. The non-transitory computer-readable storage device of claim 1 , where computing the coronary calcium score includes:

extracting a set of features from the post-decomposition image;

relating the set of features to a set of vessel-territory specific computed tomography (CT) calcium scores; and

computing the coronary calcium score based, at least in part, on the relation of the set of features to the set of vessel-territory specific CT calcium scores.

19. The non-transitory computer-readable storage device of claim 18 , where the set of features includes an exposure time feature, a vesselness filtering feature, a texture feature, a shape feature, or a curvedness feature.

20. The non-transitory computer-readable storage device of claim 18 , where relating the set of features to the set of vessel-specific CT calcium scores is performed using a support vector machine (SVM) regression.

21. The non-transitory computer-readable storage device of claim 1 , where the set of energy dependent images is acquired using switching, where the switching has a timing, where the timing of the switching is controlled by electrocardiograph (EKG) triggering.

22. The non-transitory computer-readable storage device of claim 21 , where the EKG triggering is based upon a percentage of a running average of R-R wave intervals.