Method and system for whole body bone removal and vascular visualization in medical image data

Patent metadata

Field	Value
Publication number	US-10037603-B2
Application number	US-201514703132-A
Country	US
Kind code	B2
Filing date	May 4, 2015
Priority date	May 4, 2015
Publication date	Jul 31, 2018
Grant date	Jul 31, 2018

Abstract

A method and apparatus for whole body bone removal and vasculature visualization in medical image data, such as computed tomography angiography (CTA) scans, is disclosed. Bone structures are segmented in the a 3D medical image, resulting in a bone mask of the 3D medical image. Vessel structures are segmented in the 3D medical image, resulting in a vessel mask of the 3D medical image. The bone mask and the vessel mask are refined by fusing information from the bone mask and the vessel mask. Bone voxels are removed from the 3D medical image using the refined bone mask, in order to generate a visualization of the vessel structures in the 3D medical image.

First claim

The invention claimed is:

1. A method for bone removal and vessel visualization in a 3D medical image of a patient, comprising:
segmenting bone structures in the 3D medical image, resulting in a bone mask of the 3D medical image;
segmenting vessel structures in the 3D medical image, resulting in a vessel mask of the 3D medical image, wherein segmenting bone structures in the 3D medical image and segmenting vessel structures in the 3D medical image are performed independently with respect to each other;
refining the bone mask and the vessel mask by fusing information from the bone mask and the vessel mask; and
removing bone voxels from the 3D medical image using the refined bone mask to generate a visualization of the vessel structures in the 3D medical image, wherein segmenting bone structures in the 3D medical image comprises:
detecting a plurality of landmarks in the 3D medical image,
calculating landmark-based features for each of a plurality of voxels in the 3D medical image based on the detected landmarks in the 3D medical image,
calculating image-based features for each of the plurality of voxels in the 3D medical image, and
classifying each of the plurality of voxels in the 3D medical image as bone or non-bone based on the landmark-based features and the image-based features calculated for each of the plurality of voxels using a trained voxel classifier.

2. The method of claim 1 , wherein removing bone voxels from the 3D medical image using the refined bone mask to generate a visualization of the vessel structures in the 3D medical image comprises:
subtracting the refined bone mask from the 3D medical image.

3. The method of claim 1 , further comprising:
enhancing the vessel structures in the visualization of the vessel structures in the 3D medical image using the refined vessel mask.

4. The method of claim 1 , wherein the 3D medical image is a 3D computed tomography angiography (CTA) image and removing bone voxels from the 3D medical image using the refined bone mask to generate a visualization of the vessel structures in the 3D medical image comprises:
subtracting the refined bone mask from a binary mask resulting from performing intensity-based thresholding on the 3D CTA image.

5. The method of claim 1 , wherein the plurality of voxels in the 3D medical image corresponds to a plurality of voxels having intensities greater than a predetermined intensity threshold.

6. The method of claim 1 , wherein calculating landmark-based features for each of a plurality of voxels in the 3D medical image based on the detected landmarks in the 3D medical image comprises:
calculating, for each of the plurality of voxels, a respective L1 distance from that voxel to each of the plurality of landmarks detected in the 3D medical image, a respective L2 distance from that voxel to each of the plurality of landmarks detected in the 3D medical image, and axial projections of offsets between that voxel and each of the plurality of landmarks detected in the 3D medical image.

7. The method of claim 1 , wherein calculating image-based features for each of the plurality of voxels in the 3D medical image comprises:
calculating, for each of the plurality of voxels, at least one of Haar features or relative shift intensity difference (RSID) features in an image patches centered at that voxel.

8. The method of claim 1 , wherein the trained voxel classifier is a random forest classifier trained to select backup decision criteria using a penalized measure of overlap.

9. The method of claim 1 , wherein segmenting bone structures in the 3D medical image further comprises:
fitting a body cavity mesh having a plurality of mesh points to the 3D medical image; and
calculating mesh-based features for each of a plurality of voxels in the 3D medical image based on the mesh points of the body cavity mesh fitted to the 3D medical image.

10. The method of claim 9 , wherein classifying each of the plurality of voxels in the 3D medical image as bone or non-bone based on the landmark-based features and the image-based features calculated for each of the plurality of voxels using a trained voxel classifier comprises:
classifying each of the plurality of voxels in the 3D medical image as bone or non-bone based on the landmark-based features, the image-based features, and the mesh-based features calculated for each of the plurality of voxels using a trained voxel classifier.

11. The method of claim 10 , wherein calculating mesh-based features for each of a plurality of voxels in the 3D medical image based on the mesh points of the body cavity mesh fitted to the 3D medical image comprises:
calculating, for each of the plurality of voxels, a respective L1 distance from that voxel to each of the plurality of mesh points of the body cavity mesh, a respective L2 distance from that voxel to each of the plurality of mesh points of the body cavity mesh, and axial projections of offsets between that voxel and each of the plurality of mesh points of the body cavity mesh.

12. The method of claim 1 , wherein segmenting bone structures in the 3D medical image further comprises:
automatically refining the segmentation of the bone structures in the 3D image to fill in holes in the segmented bone structures.

13. The method of claim 12 , wherein automatically refining the segmentation of the bone structures in the 3D image to fill in holes in the segmented bone structures comprises:
identifying a voxel not labeled as bone that neighbors at least one voxel labeled as bone;
perform region growing up to N voxels for the identified voxel, where N is a predetermined maximum size hole to fill; and
if a hole size for the identified voxel is less than N, label the grown region for the identified voxel as bone.

14. The method of claim 1 , wherein segmenting vessel structures in the 3D medical image comprises:
segmenting an aorta in the 3D medical image using intensity-based thresholding and morphological operations; and
segmenting remaining vessels in the 3D medical image at least one of vessel tracing or slice-based vessel segmentation.

15. The method of claim 14 , wherein segmenting an aorta in the 3D medical image using intensity-based thresholding and morphological operations comprises:
performing intensity-based thresholding on the 3D medical image to generate a binary mask having connected components of bright voxels;
performing morphological erosion on the connected components to disconnect connected components representing the aorta from connected components representing vertebrae;
classifying each of the connected components resulting from the morphological erosion as aorta or non-aorta; and
dilating the aorta connected components back to their original size.

16. The method of claim 14 , wherein segmenting remaining vessels in the 3D medical image at least one of vessel tracing or slice-based vessel segmentation comprises:
detecting connected components of bright pixels in 2D horizontal slices of the 3D medical image; and
labeling the connected components of bright pixels in the 2D horizontal slices as vessel or non-vessel based on a circleness measure of each connected component of bright pixels.

17. The method of claim 1 , wherein segmenting vessel structures in the 3D medical image comprises, for at least one vessel in the 3D medical image:
estimating a centerline and boundary for the vessel in the 3D medical image by recursively generating a plurality of layers of a curve linear grid, wherein each layer of the curve linear grid corresponds to a cross-section of the vessel at a point along

Assignees

• Siemens Healthcare Gmbh

Inventors

• Lay Nathan
• Liu David
• Zhou Shaohua Kevin
• Geiger Bernhard
• Zhang Li
• Ordy Vincent
• Xu Daguang
• Schwemmer Chris
• Wolber Philipp
• El-Zehiry Noha Youssry

Classifications

• G16H30/40

  for processing medical images, e.g. editing · CPC title

• G06V10/446

  using Haar-like filters, e.g. using integral image techniques · CPC title

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title

• G06V10/50

  by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title