System and Method for Lesion Segmentation in Whole Body Magnetic Resonance Images

A method for lesion segmentation in 3-dimensional (3D) whole body digital images comprising the steps of:

• providing a digitized 3D whole body image, said image comprising a plurality of intensities associated with a 3D grid of voxels;
• selecting a 2D region of interest (ROI) from said 3D image, said ROI containing a suspected tumor lesion;
• extending said ROI into 3-dimensions forming a volume of interest (VOI);
• iterating a foreground and background voxel competition wherein an initial segmentation of foreground voxels from background voxels is obtained;
• initializing a surface between the foreground and background voxels;
• computing a surface curvature and statistics of 3D Gaussian kernels in domain regions around each voxel on the surface to calculate an overall surface evolution speed for each voxel along the zero level set;
• moving the surface by altering the level set function;
• and
• updating statistics of said 3D Gaussian kernels in the domain regions around each voxel on the surface, and updating said surface, until an energy functional associated with said surface converges in value, wherein said surface segments said suspected lesion from said whole body image.

A method for lesion segmentation in 3-dimensional (3D) digital images comprising the steps of:

• providing a digitized 3D image, said image comprising a plurality of intensities associated with a 3D grid of voxels;
• selecting a 2D region of interest (ROI) from said 3D image, said ROI containing a suspected lesion and extending said ROI into 3-dimensions forming a volume of interest (VOI), wherein voxels on a border of said VOI are initialized as background voxels and voxels in an interior of said VOI are initialized as foreground voxels, and wherein a subset of said foreground voxels are initialized as center voxels, and a subset about a center of said lesion are initialized as surrounding voxels;
• iterating a voxel competition wherein an initial segmentation of voxels into background, center, and surrounding voxels is obtained;
• and
• evolving surfaces between said center, surrounding, and background voxels in 3D until an energy functional associated with each said surface converges in value, wherein said surfaces segment said suspected lesion from said image and a surrounding region from a center region in said lesion.

A program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method steps for lesion segmentation in 3-dimensional (3D) digital images, said method comprising the steps of:

• providing a digitized 3D image, said image comprising a plurality of intensities associated with a 3D grid of voxels;
• selecting a 2D region of interest (ROI) from said 3D image, said ROI containing a suspected lesion, extending borders of said ROI to 3D forming a volume of interest (VOI), wherein voxels on said borders of said VOI are initialized as background voxels and voxels in an interior of said VOI are initialized as foreground voxels;
• propagating a foreground and background voxel competition wherein for each voxel in said VOI, having each neighbor voxel in a neighborhood of said voxel attack said voxel, and, if said attack is successful, updating a label and strength of said voxel with that of said successful attacking voxel;
• and
• evolving a surface between the foreground and background voxels in 3D until an energy functional associated with said surface converges in value, wherein said surface segments said suspected lesion from said image.