Sparse Appearance Learning-based Segmentation

• US 20150235360A1
• Filed: 02/06/2015
• Published: 08/20/2015
• Est. Priority Date: 02/18/2014
• Status: Active Grant

• Current Assignee: Siemens Healthineers AG (Siemens AG)
• Original Assignee: Siemens AG
• Inventors: Zheng, Yefeng, Lu, Shiyang, Huang, Xiaojie
• Granted Patent: US 9,594,976 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06T 2207/10072 Tomographic images
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/10088 Magnetic resonance imaging [MRI]
• CPC Class Codes: G06T 2207/20072 Graph-based image processing
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30048 Heart; Cardiac
• CPC Class Codes: G06T 2207/30101 Blood vessel; Artery; Vein; Vascular
• CPC Class Codes: G06T 2207/30172 Centreline of tubular or elongated structure
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06T 7/162 involving graph-based methods

Abstract

The coronary sinus or other vessel is segmented by finding a centerline and then using the centerline to locate the boundary of the vessel. For finding the centerline, a refinement process uses multi-scale sparse appearance learning. For locating the boundary, the lumen is segmented as a graph cut problem.

Claims

1. A method for segmentation of a vessel, the method comprising:

extracting a centerline for a vessel represented in medical imaging data representing a patient, the extracting being a function of a cost term;

calculating the cost term as a function of similarity of patches of the medical imaging data to machine-learned appearance patterns at multiple scales relative to the medical imaging data, the cost term indicating membership as in the vessel or not in the vessel of different locations for the patches;

segmenting the vessel as represented in the medical imaging data, the segmenting being a function of the centerline; and

generating an image of the vessel segmented from the medical imaging data.

2. The method of claim 1 wherein extracting comprises extracting the centerline for a coronary sinus.

3. The method of claim 1 wherein extracting comprises computing a shortest path as a function of the cost term.

4. The method of claim 1 wherein extracting comprises extracting as a function of the cost term and a path length.

5. The method of claim 1 wherein calculating comprises forming an appearance vector for each patch of voxels of the medical imaging data, calculating reconstruction residues of the appearance vectors with the machine-learned appearance patterns, and classifying the patches as a function of the reconstruction residues.

6. The method of claim 1 wherein calculating comprises applying appearance dictionaries for the vessel and for background tissue to the patches.

7. The method of claim 1 wherein calculating comprises calculating as a multi-scale sparse appearance model.

8. The method of claim 1 wherein calculating comprises calculating the similarities at each of the multiple scales for each of the different locations, and weighted averaging the similarities from the multiple scales.

9. The method of claim 1 wherein segmenting comprises:

defining planes along and orthogonal to the centerline and resampling the medical imaging data along the planes; and

detecting a boundary of the vessel in each of the planes.

10. The method of claim 1 wherein segmenting comprises determining a boundary probability for different points being at a boundary of the vessel.

11. The method of claim 10 wherein segmenting further comprises determining a minimally cut graph for a Markov Random Field.

12. The method of claim 1 wherein generating the image comprises generating a three-dimensional rendering of the vessel identified from the segmenting.

13. In a non-transitory computer readable storage medium having stored therein data representing instructions executable by a programmed processor for segmentation of a vessel, the storage medium comprising instructions for:

determining probabilities of locations in a volume of a patient represented by medical imaging data being at a vessel boundary;

constructing a Markov random field graph from the probabilities;

extracting a contour of the vessel boundary from the Markov random field graph; and

segmenting the vessel with the vessel boundary.

14. The non-transitory computer readable storage medium of claim 13 wherein determining the probabilities comprises applying a machine-learnt classifier.

15. The non-transitory computer readable storage medium of claim 13 wherein constructing comprises constructing the Markov random field graph as a network graph of vertices and edges labeled with the probabilities.

16. The non-transitory computer readable storage medium of claim 13 wherein extracting comprises calculating a maximum flow-minimum cut.

17. The non-transitory computer readable storage medium of claim 13 wherein extracting comprises finding a minimum cut of the Markov random field graph.

18. The non-transitory computer readable storage medium of claim 13 further comprising defining the locations relative to a straightened centerline of the vessel and warping the contour as a function of a curved centerline of the vessel in the patient.

19. The non-transitory computer readable storage medium of claim 13 further comprising locating a centerline of the vessel with multi-scale sparse appearance;

and

wherein determining the probabilities comprises determining the probabilities as a function of the centerline.

20. A method for segmentation of a coronary sinus vessel, the method comprising:

extracting a centerline of the coronary sinus vessel in a volume represented by computed tomography data, the extracting being a function of sparse appearance modeling at multiple scales;

identifying voxels of the volume belonging to the coronary sinus vessel as a function of the centerline and a graph optimization; and

generating an image from the voxels of the volume identified as belonging to the coronary sinus vessel.