Detection of infarcts using trained network
A system and method includes acquisition of a first plurality of images, determination, for each of the first plurality of images, of whether an infarct is depicted in the image and generating a label associated with the image based on the determination, acquisition of a second plurality of non-contrast-enhanced computed tomography images, determination, for each of the second plurality of non-contrast-enhanced computed tomography images, of a corresponding one of the first plurality of images, association of each of the second plurality of non-contrast-enhanced computed tomography images with the label associated with its corresponding one of the first plurality of images, and training of a neural network to output a network-generated label, the training based on the second plurality of non-contrast-enhanced computed tomography images and associated labels.
1. A computing system comprising:
a storage system;
one or more processors to execute processor-executable process steps stored on the storage system to cause the computing system to:
acquire a first plurality of magnetic resonance slice images;
for each of the first plurality of magnetic resonance slice images, determine whether an infarct is depicted in the image and generate a label associated with the image based on the determination;
acquire a second plurality of contrast-enhanced computed tomography slice images;
for each of the second plurality of contrast-enhanced computed tomography slice images, determine a corresponding one of the first plurality of magnetic resonance slice images;
for each of the second plurality of contrast-enhanced computed tomography slice images, associate the contrast-enhanced computed tomography slice image with the label associated with the corresponding one of the first plurality of magnetic resonance slice images;
acquire a third plurality of non-contrast-enhanced computed tomography slice images;
for each of the third plurality of non-contrast-enhanced computed tomography slice images, determine a corresponding one of the second plurality of contrast-enhanced computed tomography slice images;
for each of the third plurality of non-contrast-enhanced computed tomography images, associate the non-contrast-enhanced computed tomography slice image with the label associated with the corresponding one of the second plurality of contrast-enhanced computed tomography slice images; and
train a network to generate a label based on an input non-contrast-enhanced computed tomography slice image by iteratively inputting the third plurality of non-contrast-enhanced computed tomography slice images into the network, operating the network to output labels corresponding to the input third plurality of non-contrast-enhanced computed tomography slice images, comparing the output labels with the labels associated with each of the third plurality of non-contrast-enhanced computed tomography slice images, and modifying the network based on the comparison.
2. A computing system according to claim 1 , the one or more processors to execute processor-executable process steps stored on the storage system to cause the computing system to:
operate the trained network to receive a first non-contrast-enhanced computed tomography image and output a label indicating whether an infarct is depicted in the first non-contrast-enhanced computed tomography image.
3. A computing system according to claim 1 , wherein determination of a corresponding one of the first plurality of magnetic resonance slice images for each of the second plurality of contrast-enhanced computed tomography slice images comprises:
detecting first myocardium structure within the first plurality of magnetic resonance slice images;
detecting second myocardium structure within the second plurality of slice images; and
registering the first plurality of slice images with the second plurality of slice images.
4. A computing system according to claim 1 , wherein the label indicates a location of an infarct.
5. A computing system according to claim 1 , the one or more processors to execute processor-executable process steps stored on the storage system to cause the computing system to:
operate the trained network to receive a first non-contrast-enhanced computed tomography image and output an image including the first non-contrast-enhanced computed tomography image and a graphic overlaying a location of an infarct depicted in the first non-contrast-enhanced computed tomography image.
6. A computer-implemented method comprising:
acquiring a first plurality of magnetic resonance slice images;
for each of the first plurality of magnetic resonance slice images, determining whether an infarct is depicted in the image and generating a label associated with the image based on the determination;
acquiring a second plurality of contrast-enhanced computed tomography slice images;
for each of the second plurality of contrast-enhanced computed tomography slice images, determining a corresponding one of the first plurality of magnetic resonance slice images;
for each of the second plurality of contrast-enhanced computed tomography slice images, associating the contrast-enhanced computed tomography slice image with the label associated with the corresponding one of the first plurality of magnetic resonance slice images;
acquiring a third plurality of non-contrast-enhanced computed tomography slice images;
for each of the third plurality of non-contrast-enhanced computed tomography slice images, determining a corresponding one of the second plurality of contrast-enhanced computed tomography slice images;
for each of the third plurality of non-contrast-enhanced computed tomography images, associating the non-contrast-enhanced computed tomography slice image with the label associated with the corresponding one of the second plurality of contrast-enhanced computed tomography slice images; and
training a neural network to generate a label based on an input non-contrast-enhanced computed tomography slice image by iteratively inputting the third plurality of non-contrast-enhanced computed tomography slice images into the network, operating the network to output labels corresponding to the input third plurality of non-contrast-enhanced computed tomography slice images, comparing the output labels with the labels associated with each of the third plurality of non-contrast-enhanced computed tomography slice images, and modifying the network based on the comparison.
7. A method according to claim 6 , further comprising:
operating the trained network to receive a first non-contrast-enhanced computed tomography image and output a label indicating whether an infarct is depicted in the first non-contrast-enhanced computed tomography image.
8. A method according to claim 1 , wherein determining a corresponding one of the first plurality of magnetic resonance slice images for each of the second plurality of contrast-enhanced computed tomography images comprises:
detecting first myocardium structure within the first plurality of magnetic resonance slice images;
detecting second myocardium structure within the second plurality of slice images; and
registering the first plurality of images with the second plurality of slice images.
9. A method according to claim 6 , wherein the label indicates a location of an infarct.
10. A method according to claim 6 , further comprising operating the trained network to receive a first non-contrast-enhanced computed tomography image and output an image including the first non-contrast-enhanced computed tomography image and a graphic overlaying a location of an infarct depicted in the first non-contrast-enhanced computed tomography image.