Medical image segmentation
View Patent ↗A method of automatically segmenting a patient's colon in a medical image data set is provided. The method comprises two main stages. The first stage is to classify voxels in the data set as not belonging to the colon based on their radiodensities and/or positions within the volume data set. In taking account of the voxels' positions in the data set, account may be taken of their positions relative to other voxels. For example, this may include determining whether voxels are within connected groups of voxels of similar characteristics, i.e. within connected regions or domains. Subsequent to the first stage of classifying voxels as not belonging to the colon, a second stage is performed in which all remaining voxels are classified as belonging to the colon. Thus the method allows for automatic segmentation of the colon without needing any particular voxel, such as a seed voxel for region growing, to first be positively identified as being within the colon.
1. A method of segmenting a colon in a medical image volume data set comprising an array of voxels having associated signal values, the method comprising:
defining a segmentation data structure for maintaining segmentation information; and
classifying voxels in the segmentation data structure by performing a first step of classifying voxels as not belonging to the colon based on their associated signal value and/or position in the data set, and performing a second step of classifying remaining voxels as belonging to the colon, thereby automatically segmenting the colon without first requiring a voxel to be identified within it.
2. A method of segmenting a colon according to claim 1 , wherein the first step includes classifying voxels having signal values greater than an intensity threshold as not belonging to the colon.
3. A method of segmenting a colon according to claim 2 , wherein the intensity threshold is selected from the group consisting of −200, −300, −400, −500, −600, −700 and −800 Hounsfield Units.
4. A method of segmenting a colon according to claim 1 , wherein the first step includes classifying voxels with a position within a connected region of voxels contained wholly within a pre-determined sub-portion of the data set as not belonging to the colon.
5. A method of segmenting a colon according to claim 4 , wherein the sub-portion comprises a region of the data set extending away from the posterior coronal boundary of the data set.
6. A method of segmenting a colon according to claim 4 , wherein the sub-portion comprises a fraction of selected from the group consisting of 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45% and 50% of the voxels in the medical image data set.
7. A method of segmenting a colon according to claim 1 , wherein the first step includes classifying voxels in connected regions which intersect with the sagittal and coronal boundaries of the data set as not belonging to the colon.
8. A method of segmenting a colon according to claim 1 , wherein the first step includes selecting first and second transverse planes in the volume data set and classifying voxels in connected regions which intersect with the first transverse plane but not with the second transverse plane as not belonging to the colon.
9. A method of segmenting a colon according to claim 1 , wherein the first step includes classifying voxels in connected regions having a volume less than a threshold number of voxels as not belonging to the colon.
10. A method of segmenting a colon according to claim 9 , wherein the threshold number of voxels is selected from the group consisting of 50, 100, 150, 200, 250 and 300 voxels.
11. A method of segmenting a colon according to claim 1 , further comprising displaying an image which represents the segmented colon to a user based on the contents of the segmentation data structure.
12. A non-transitory computer readable storage medium with an executable program stored thereon, wherein the program instructs a processor to perform the method of claim 1 .
13. A system for segmenting a colon in a medical image volume data set comprising an array of voxels having associated signal values, the system comprising:
a processor and memory; and
logic stored in memory and executable by the processor, said logic including logic that defines a segmentation data structure for maintaining segmentation information, and
logic that classifies voxels in the segmentation data structure by performing a first step of classifying voxels as not belonging to the colon based on their associated signal value and/or position in the data set, and performing a second step of classifying remaining voxels as belonging to the colon, thereby automatically segmenting the colon without first requiring a voxel to be identified within it.
14. The system according to claim 13 , further comprising a computer network.
15. An apparatus for segmenting a colon in a medical image data set comprising an array of voxels having associated signal values, the apparatus comprising:
a processing unit coupled to a source of medical image data, wherein the processing unit is operable to load a medical image data set from the source of medical image data, and further operable to define a segmentation data structure for maintaining segmentation information; and
classify voxels in the segmentation data structure by performing a first step of classifying voxels as not belonging to the colon based on their associated signal value and/or position in the data set, and performing a second step of classifying remaining voxels as belonging to the colon, thereby automatically segmenting the colon without first requiring a voxel to be identified within it.