Method and system for segmentation of the liver in magnetic resonance images using multi-channel features
View Patent ↗A method and system for fully automatic liver segmentation in a multi-channel magnetic resonance (MR) image is disclosed. An initial liver boundary in the multi-channel MR image, such as an MR Dixon scan. The segmented initial liver boundary in the multi-channel MR image is refined based on features extracted from multiple channels of the multi-channel MR image using a trained boundary detector. The features may be extracted from an opposed channel and a water channel of an MR Dixon scan.
1. A method for liver segmentation in a multi-channel magnetic resonance (MR) image, wherein the multi-channel MR image is obtained using an MR Dixon scan, comprising:
segmenting an initial liver boundary in the multi-channel MR image; and
refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector.
2. The method of claim 1 , wherein segmenting an initial liver boundary in the multi-channel MR image comprises:
segmenting the initial liver boundary in the multi-channel MR image based on features extracted from multiple channels of the multi-channel MR image.
3. The method of claim 2 , wherein segmenting the initial liver boundary in the multi-channel MR image based on features extracted from multiple channels of the multi-channel MR image comprises:
segmenting the initial liver boundary in the multi-channel MR image using marginal space learning (MSL) segmentation based on the features extracted from multiple channels of the multi-channel MR image.
4. The method of claim 1 , wherein refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
for each of a plurality of boundary points on the initial liver boundary,
calculating a probability response for each of a set of candidate points along a normal direction from the boundary point using the trained boundary detector based on features extracted from the opposed channel and the water channel of the multi-channel MR image; and
moving to boundary point to a candidate point having the highest probability response.
5. The method of claim 1 , wherein refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
generating a hierarchical shape model based on the segmented initial liver boundary, the hierarchical shape model having a plurality of levels and a respective liver mesh at each of the plurality of levels represents the initial liver boundary at a respective one of a plurality of image resolutions; and
adjusting boundary points of the respective mesh at each of the plurality of levels of the hierarchical shape model from a coarsest level to a finest level based on features extracted from the opposed channel and the water channel of the multi-channel MR image at the respective image resolution using a respective trained boundary detector for each respective image resolution.
6. The method of claim 1 , wherein refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
refining the segmented initial liver boundary based on steerable features extracted from the opposed channel and the water channel of the multi-channel MR image using a training boundary detector.
7. The method of claim 1 , wherein segmenting an initial liver boundary in the multi-channel MR image comprises:
segmenting the initial liver boundary in the multi-channel MR image based on features extracted from the opposed channel and the water channel of the multi-channel MR image.
8. The method of claim 1 , wherein segmenting an initial liver boundary in the multi-channel MR image comprises:
segmenting the initial boundary in the multi-channel MR image based on features extracted from a single channel of the multi-channel MR image.
9. An apparatus for liver segmentation in a multi-channel magnetic resonance (MR) image, wherein the multi-channel MR image is obtained using an MR Dixon scan, comprising:
means for segmenting an initial liver boundary in the multi-channel MR image; and
means for refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector.
10. The apparatus of claim 9 , wherein the means for segmenting an initial liver boundary in the multi-channel MR image comprises:
means for segmenting the initial liver boundary in the multi-channel MR image based on features extracted from multiple channels of the multi-channel MR image.
11. The apparatus of claim 9 , wherein the means for refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
means for calculating, for each of a plurality of boundary points on the initial liver boundary, a probability response for each of a set of candidate points along a normal direction from the boundary point using the trained boundary detector based on features extracted from the opposed channel and the water channel of the multi-channel MR image; and
means for moving each of the plurality of boundary points to a candidate point having the highest probability response.
12. The apparatus of claim 9 , wherein the means for refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
means for generating a hierarchical shape model based on the segmented initial liver boundary, the hierarchical shape model having a plurality of levels and a respective liver mesh at each of the plurality of levels represents the initial liver boundary at a respective one of a plurality of image resolutions; and
means for adjusting boundary points of the respective mesh at each of the plurality of levels of the hierarchical shape model from a coarsest level to a finest level based on features extracted from the opposed channel and the water channel of the multi-channel MR image at the respective image resolution using a respective trained boundary detector for each respective image resolution.
13. The apparatus of claim 9 , wherein the means for segmenting an initial liver boundary in the multi-channel MR image comprises:
means for segmenting the initial liver boundary in the multi-channel MR image based on features extracted from the opposed channel and the water channel of the multi-channel MR image.
14. The apparatus of claim 9 , wherein the means for segmenting an initial liver boundary in the multi-channel MR image comprises:
means for segmenting the initial boundary in the multi-channel MR image based on features extracted from a single channel of the multi-channel MR image.
15. A non-transitory computer readable medium storing computer program instructions for liver segmentation in a multi-channel magnetic resonance (MR) image, wherein the multi-channel MR image is obtained using an MR Dixon scan, the computer program instructions when executed by a processor cause the processor to perform operations comprising:
segmenting an initial liver boundary in the multi-channel MR image; and
refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector.
16. The non-transitory computer readable medium of claim 15 , wherein segmenting an initial liver boundary in the multi-channel MR image comprises:
segmenting the initial liver boundary in the multi-channel MR image based on features extracted from multiple channels of the multi-channel MR image.
17. The non-transitory computer readable medium of claim 16 , wherein segmenting the initial liver boundary in the multi-channel MR image based on features extracted from multiple channels of the multi-channel MR image comprises:
segmenting the initial liver boundary in the multi-channel MR image using marginal space learning (MSL) segmentation based on the features extracted from multiple channels of the multi-channel MR image.
18. The non-transitory computer readable medium of claim 15 , wherein refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
for each of a plurality of boundary points on the initial liver boundary, calculating a probability response for each of a set of candidate points along a normal direction from the boundary point using the trained boundary detector based on features extracted from the opposed channel and the water channel of the multi-channel MR image; and
moving to boundary point to a candidate point having the highest probability response.
19. The non-transitory computer readable medium of claim 15 , wherein refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
generating a hierarchical shape model based on the segmented initial liver boundary, the hierarchical shape model having a plurality of levels and a respective liver mesh at each of the plurality of levels represents the initial liver boundary at a respective one of a plurality of image resolutions; and
adjusting boundary points of the respective mesh at each of the plurality of levels of the hierarchical shape model from a coarsest level to a finest level based on features extracted from the opposed channel and the water channel of the multi-channel MR image resolution at the respective image resolution using a respective trained boundary detector for each respective image resolution.
20. The non-transitory computer readable medium of claim 15 , wherein refining the segmented initial liver boundary in the multi-channel MR image based on features extracted from an opposed channel and a water channel of the multi-channel MR image using a trained boundary detector comprises:
refining the segmented initial liver boundary based on steerable features extracted from the opposed channel and the water channel of the multi-channel MR image using a training boundary detector.
21. The non-transitory computer readable medium of claim 15 , wherein segmenting an initial liver boundary in the multi-channel MR image comprises:
segmenting the initial liver boundary in the multi-channel MR image based on features extracted from the opposed channel and the water channel of the multi-channel MR image.
22. The non-transitory computer readable medium of claim 15 , wherein segmenting an initial liver boundary in the multi-channel MR image comprises:
segmenting the initial boundary in the multi-channel MR image based on features extracted from a single channel of the multi-channel MR image.