Generating pseudo-CT image volumes from ultra-short echo time MR

1. A method for generating a pseudo-computed tomography (CT) image volume, comprising:

acquiring a first magnetic resonance (MR) image volume (UTE 1 ) using an ultra-short echo time within a range of from 0 to 100 ms;

acquiring a second MR image volume (UTE 2 ) using a conventional echo time that is longer than 100 ms;

normalizing the acquired UTE 1 and UTE 2 image volumes;

creating a first mask for an anatomical structure featured in the normalized UTE 1 image volume and creating a second mask for the anatomical structure featured in the normalized UTE 2 image volume;

segmenting bone regions from the normalized UTE 1 and UTE 2 image volumes using the created first and second masks and one or more trained classifiers; and

constructing a pseudo-CT image from the normalized UTE 1 and UTE 2 image volumes, the created first and second masks, and the segmented bone regions.

2. The method of claim 1 , wherein the anatomical structure is a head.

3. The method of claim 1 , wherein the first MR image volume UTE 1 and the second MR image volume UTE 2 are acquired using a single UTE MR protocol.

4. The method of claim 1 , wherein the acquired UTE 1 and UTE 2 image volumes are normalized to standardize intensity spreads between a main background region and a foreground region.

5. The method of claim 1 , wherein normalization of the acquired UTE 1 and UTE 2 image volumes includes histogramming the UTE 1 and UTE 2 image volumes and identifying a first intensity value at a first major peak attributable to a background region and a second intensity value attributable a second major peak attributable to a foreground region and performing linear remapping on the UTE 1 and UTE 2 images to normalize the first and second intensity values.

6. The method of claim 1 , wherein creating the mask first and second masks of the anatomical structure includes:

thresholding the normalized UTE 1 and UTE 2 image volumes;

using morphological methods to remove noise due to stray intensity patches;

applying a morphological closing operation to remove small gaps; and

performing region labeling to detect small unconnected regions which are then labeled as background.

7. The method of claim 1 , wherein the one or more trained classifiers use features based on one or more of the following:

a difference between the normalized UTE 1 and UTE 2 image volumes;

a first localization map proving voxel distance inward from an outer surface of the anatomical structure; and

a second localization map providing voxel distance downward from a top plane of the anatomical structure.

8. The method of claim 1 , wherein the one or more trained classifiers are trained using training data including:

UTE 1 image volumes;

UTE 2 image volumes; and

corresponding CT scans that have been registered to serve as ground truth.

9. The method of claim 8 , wherein training the trained classifiers includes normalizing the UTE 1 and UTE 2 image volumes and extracting classifier features.

10. The method of claim 9 , wherein extracting classifier features includes:

calculating a difference between the normalized UTE 1 and UTE image volumes;

generating a first localization map providing voxel distance inward from an outer surface of the anatomical structure; and

generating a second localization map providing voxel distance downward from a top plane of the anatomical structure.

11. The method of claim 1 , wherein constructing the pseudo-CT image includes assigning appropriate Hounsfield scale values based on the anatomical map and the segmented bone regions.

12. The method of claim 1 , additionally including:

acquiring a positron emission tomography (PET) scan;

generating an approximate attenuation correction (AC) map based on the constructed pseudo-CT image; and

calibrating an intensity of the PET result using the generated AC map.

13. A method for generating a pseudo-computed tomography (CT) image volume, comprising:

acquiring a set of training data including a first magnetic resonance (MR) image volume (UTE 1 ) acquired using an ultra-short echo time within a range of from 0 to 100 ms, a second MR image volume (UTE 2 ) acquired using a conventional echo time that is longer than 100 ms, and corresponding reference CT images;

normalizing the UTE 1 and UTE 2 images of the training data;

extracting one or more features from the normalized UTE 1 and UTE 2 images of the training data;

training one or more classifiers based on the normalized UTE 1 and UTE 2 images of the training data and the one or more extracted features;

acquiring test case data including UTE 1 and UTE 2 images;

normalizing the acquired UTE 1 and UTE 2 images of the test case data;

creating a first mask for an anatomical structure featured in the normalized UTE 1 image of the test case data and creating a second mask for the anatomical structure feature in the normalized UTE 2 image of the test case data;

segmenting bone regions from the normalized UTE 1 and UTE 2 images of the test case data using the created mask and one or more trained classifiers; and

constructing a pseudo-CT image from the normalized UTE 1 and UTE 2 images of the test case data, the created first and second masks, and the segmented bone regions.

14. The method of claim 13 , wherein normalization of the acquired UTE 1 and UTE 2 image volumes of the training data and the test data includes histogramming the UTE 1 and UTE 2 image volumes and identifying a first intensity value at a first major peak attributable to a background region and a second intensity value attributable a second major peak attributable to a foreground region and performing linear remapping on the UTE 1 and UTE 2 images to normalize the first and second intensity values.

15. The method of claim 13 , wherein the features extracted from the normalized UTE 1 and UTE 2 images of the training data are based on one or more of the following:

a difference between the normalized UTE 1 and UTE 2 images of the training data;

a first localization map proving voxel distance inward from an outer surface of the anatomical structure; and

a second localization map providing voxel distance downward from a top plane of the anatomical structure.

16. A method for generating a pseudo-computed tomography (CT) image volume, comprising:

acquiring a first magnetic resonance (MR) image volume (UTE 1 ) of a head using an ultra-short echo time within a range of from 0 to 100 ms;

acquiring a second MR image volume (UTE 2 ) of the head using a conventional echo time that is longer than 100 ms;

normalizing the acquired UTE 1 and UTE 2 image volumes;

creating a first head mask in the normalized UTE 1 image volume and creating a second head mask in the normalized UTE 2 image volume;

segmenting bone regions from the normalized UTE 1 and UTE 2 image volumes using the created first and second head masks and one or more trained classifiers; and

constructing a pseudo-CT image from the normalized UTE 1 and UTE 2 image volumes, the created first and second head masks, and the segmented bone regions.

17. The method of claim 16 , wherein constructing the pseudo-CT image includes:

assigning a first Hounsfield scale value to the segmented bone regions;

assigning a second Hounsfield scale value to air regions determined using the head map; and

assigning a third Hounsfield scale value to brain regions determined using the normalized UTE 2 image volume.

18. A computer system comprising: a processor; and a non-transitory, tangible, program storage medium, readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for generating a medical image, the method comprising:

acquiring a first magnetic resonance (MR) image volume (UTE 1 ) using an ultra-short echo time within a range of from 0 to 100 ms;

acquiring a second MR image volume (UTE 2 ) using a conventional echo time that is longer than 100 ms;

normalizing the acquired UTE 1 and UTE 2 image volumes;

segmenting bone regions from the normalized UTE 1 and UTE 2 image volumes, a first segmentation mask produced from the UTE 1 image, a second segmentation mask produced from the UTE 2 image, and one or more trained classifiers; and

constructing a pseudo-CT image by assigning appropriate Hounsfield scale values based on the normalized UTE 1 and UTE 2 image volumes and segmented bone regions.

19. The computer system of claim 18 , wherein the anatomical structure is a head and the first and second segmentation masks are created from the normalized UTE 1 and UTE 2 image volumes, respectively, and the generated first and second head masks are used in segmenting the bone region and constructing the pseudo-CT image.

20. The method of claim 18 , wherein the one or more trained classifiers are trained using:

training data including UTE 1 image volumes, UTE 2 image volumes, and corresponding CT scans that have been registered to serve as ground truth; and

features based on one or more of the following: a difference between the normalized UTE 1 and UTE 2 image volumes, a first localization map proving voxel distance inward from an outer surface of the anatomical structure, and a second localization map providing voxel distance downward from a top plane of the anatomical structure.