Method and apparatus for processing medical images
The present invention provides, among other things, methods of processing medical images for producing images with labeled anatomical features, including obtaining images containing labeled anatomical features, obtaining unlabelled images, comparing and selecting unlabelled images that most closely resemble labeled images, and propagating label data from labeled images to unlabelled images, thereby labeling corresponding anatomical features on unlabelled images. The present invention also provides systems for performing such methods.
1. A method of processing medical images, the method being performed by a computer processor and comprising steps of:
(a) obtaining one or more atlases containing one or more images, the one or more images are characterized in that they are composed of a plurality of voxels wherein at least some of the voxels correspond to one or more anatomical features that have been labelled with label data;
(b) obtaining a plurality of unlabelled images each composed of a respective plurality of voxels;
(c) comparing the one or more labelled images and each unlabelled image of the plurality of unlabeled images, resulting in a comparison;
(d) based on the comparison, selecting one or more unlabelled images that most closely resemble(s) one or more of the labelled images;
(e) propagating label data of the one or more labelled anatomical features from the one or more closest of the labelled images to each of the selected one or more unlabelled images, so that when the corresponding anatomical feature(s) of each of the selected images(s) become labelled, the selected image(s) become labelled image(s); and
(f) iteratively repeating from step (c), thereby labelling, for each iteration, one or more others of the unlabelled images of the plurality of unlabelled images so as to increase the number of labelled images contained in the one or more atlases.
2. The method as claimed in claim 1 , wherein the step of comparing the labelled and unlabelled images comprises embedding the images into a low-dimensional coordinate system.
3. The method as claimed in claim 2 , wherein the low-dimensional coordinate system is a two-dimensional coordinate space.
4. The method as claimed in claim 1 , wherein the step of comparing the labelled and unlabelled images comprises defining a set of pairwise measures of similarity by comparing one or more respective anatomical features for each pair of images in the set of images.
5. The method as claimed in claim 4 , wherein the step of comparing the labelled and unlabelled images further comprises performing a spectral analysis operation on the pairwise measures of similarity.
6. The method as claimed in claim 4 , wherein the pairwise measures of similarity represent the intensity similarity between a pair of images.
7. The method as claimed in claim 4 , wherein the pairwise measures of similarity represent the amount of deformation between a pair of images.
8. The method as claimed in claim 1 , wherein the step of propagating label data comprises propagating label data from a plurality of the closest of the labelled images, based on a classifier fusion technique.
9. The method as claimed in claim 1 , further comprising, after step (e) and before step (f), a step of performing an intensity-based refinement operation on the newly-propagated label data.
10. The method as claimed in claim 1 , wherein the images are of different subjects.
11. The method as claimed in claim 1 , wherein at least some of the images are of the same subject but taken at different points in time.
12. The method as claimed in claim 1 , wherein the images are magnetic resonance images.
13. The method as claimed in claim 1 , further comprising labelling an anatomical feature representative of the presence or absence of a condition and using that feature to derive a biomarker for that condition.
14. The method as claimed in claim 13 , further comprising allocating a subject to a diagnostic category on the basis of the biomarker.
15. The method as claimed in claim 13 , further comprising quantifying a subject's response to treatment on the basis of the biomarker.
16. The method as claimed in claim 13 , further comprising selecting a subject's treatment on the basis of the biomarker.
17. A system, comprising:
a processor; and
a memory storing instructions thereon, wherein the instructions when executed cause the processor:
to obtain at least one atlas containing at least one image, the at least one image is characterized in that it is composed of a plurality of voxels wherein at least some of the voxels correspond to one or more anatomical features labeled with label data;
to obtain a plurality of unlabelled images each composed of a respective plurality of voxels;
to compare the at least one image having one or more anatomical features labeled with label data and the plurality of unlabelled images, resulting in a comparison;
based on the comparison, to select at least one of the plurality of unlabelled images that most closely resembles the at least one image having one or more anatomical features labeled with label data;
to propagate label data from the at least one image having one or more anatomical features labeled with label data from the at least one closest labeled image to each of the selected unlabelled image, so that when the corresponding anatomical feature(s) of each of the selected image(s) become labelled, the selected images(s) become labelled image(s); and
to iteratively repeat for each of the selected unlabelled images, thereby labelling, for each iteration, one or more others of the unlabelled images of the plurality of unlabelled images so as to increase the number of labelled images contained in the at least on atlas.
18. The system of claim 17 , wherein the system is a medical scanner.
19. The system of claim 18 , wherein the system is an MRI scanner.
20. A non-transitory computer readable medium having instructions thereon that, when executed, perform operations comprising:
(a) obtaining one or more atlases containing one or more images, the one or more images are characterized in that they are composed of a plurality of voxels wherein at least some of the voxels correspond to one or more anatomical features that have been labelled with label data;
(b) obtaining a plurality of unlabelled images each composed of a respective plurality of voxels;
(c) comparing the one or more labelled images and each unlabelled image of the plurality of unlabeled images, resulting in a comparison;
(d) based on the comparison, selecting one or more unlabelled images that most closely resemble(s) one or more of the labelled images;
(e) propagating label data of the one or more labelled anatomical features from the one or more closest of the labelled images to each of the selected one or more unlabelled images, so that when the corresponding anatomical feature(s) of each of the selected images(s) become labelled, the selected image(s) become labelled image(s); and
(f) iteratively repeating from step (c), thereby labelling, for each iteration, one or more others of the unlabelled images of the plurality of unlabelled images so as to increase the number of labelled images contained in the one or more atlases.