1. A method for using computer-aided diagnosis (CAD) for digital tomosynthesis mammograms (DTM) comprising using a processor for: retrieving a DTM image file having a plurality of DTM image slices; applying a three-dimensional analysis to the DTM image file to detect lesion candidates defining a location; identifying a volume of interest and locating its center at the location, which is one of high gradient convergence or high response; segmenting an object corresponding to the volume of interest by a three dimensional method; extracting one or more object characteristics from the object corresponding to the volume of interest, the object characteristics being one of a morphological feature, a gray level feature, or a texture feature; and invoking a classifier to determine if the object corresponding to the volume of interest is a breast lesion or normal breast tissue.
2. The method of claim 1 , further comprising invoking the classifier to determine if the breast lesion is malignant or benign.
3. The method of claim 1 , including applying a three-dimensional Hessian matrix analysis to the DTM image file to detect lesion candidates.
4. The method of claim 1 , including applying a three-dimensional gradient field analysis to the DTM image file to detect lesion candidates; and averaging a plurality of image voxels to obtain a smoothed volumetric data set, wherein the volumetric data set corresponds to the DTM image file.
5. The method of claim 1 , including applying a three-dimensional gradient field analysis to the DTM image file to detect lesion candidates; and calculating the three-dimensional gradient field analysis in a spherical region 5 to 10 mm in radius.
6. The method of claim 5 , including determining a gradient vector at a smoothed voxel in the spherical region, and projecting the gradient vector to a radial direction from the center of the spherical region to the smoothed voxel.
7. The method of claim 1 , including segmenting the object corresponding to the volume of interest by one of a three-dimensional active contour method or level set method.
8. The method of claim 1 , including segmenting the object corresponding to the volume of interest by a three-dimensional growing method.
9. The method of claim 8 , wherein segmenting the object corresponding to the volume of interest includes using the location of high gradient convergence as a starting point and allowing the object corresponding to the volume of interest to grow across at least a portion of the plurality of DTM image slices in the DTM image file.
10. The method of claim 8 , including guiding the three-dimensional region growing method by a radial gradient magnitude.
11. The method of claim 10 , including terminating the growth of the object corresponding to the volume of interest where the radial gradient reaches a threshold value.
12. The method of claim 1 , including extracting a morphological feature from the object corresponding to the volume of interest, wherein the morphological feature is selected from the group of morphological features consisting of: a volume of the object, a volume change before and after three-dimensional opening by a spherical element with a predetermined voxel radius, a surface area of the object, a maximum perimeter of the object among all of the DTM image slices in the DTM image file intersecting the object, a longest diameter of the object, and a compactness of the object.
13. The method of claim 1 , including extracting a gray level feature from the object corresponding to the volume of interest, wherein the gray level feature is selected from the group of gray level features consisting of: a contrast of the object relative to the surrounding background of the object, a minimum gray level, a maximum gray level, a skewness characteristic, a kurtosis characteristic, an energy characteristic, and an entropy characteristic.
14. The method of claim 1 , including extracting one or more object characteristics from the object corresponding to the volume of interest, wherein the object characteristic is one of a spiculation feature, a boundary sharpness, or a shape irregularity.
15. A method for using computer-aided diagnosis (CAD) for digital tomosynthesis mammograms (DTM) comprising using a processor for: applying a pre-screening analysis to a DTM image file to detect a potential breast cancer lesion, the DTM image file including a plurality of DTM image slices; identifying a volume of interest; segmenting the volume of interest by a three dimensional growing method by using one of a location of high gradient convergence or high response as a starting point and allowing an object corresponding to the volume of interest to grow across at least a portion of the plurality of DTM image slices in the DTM image file; extracting a morphological feature from the object corresponding to the volume of interest, wherein the morphological feature is selected from the group of morphological features consisting of: a volume of the object, a volume change before and after three-dimensional opening by a spherical element with a predetermined voxel radius, a surface area of the object, a maximum perimeter of the object among all of the DTM image slices in the DTM image file intersecting the object, a longest diameter of the object, and a compactness of the object; and extracting a gray level feature from the object corresponding to the volume of interest, wherein the gray level feature is selected from the group of gray level features consisting of: a contrast of the object relative to the surrounding background of the object, a minimum gray level, a maximum gray level, a skewness characteristic, a kurtosis characteristic, an energy characteristic, and an entropy characteristic.
16. The method of claim 15 , including invoking a classifier to determine if the object corresponding to the volume of interest is a breast cancer lesion.
17. The method of claim 16 , wherein if the classifier determines that the object corresponding to the volume of interest is a breast lesion, invoking a classifier to determine if the lesion is malignant or benign.
18. The method of claim 15 , including extracting a texture feature from the object corresponding to the volume of interest, wherein the texture feature is extracted from the lesion analyzed in a number of planes, including one or more of the following planes: (1) along the reconstructed high-resolution DTM slices that are parallel to the detector plane, (2) a set of planes that cut through the centroid and the south-north poles, and (3) a set of planes that cut through the centroid and the east-west poles; and the texture feature is selected from the group of texture measures consisting of: run-length statistics textures, spatial gray level dependence textures, gray level dependence statistics texture, or Laws textures.
19. The method of claim 15 , including averaging a plurality of image voxels to obtain a smoothed volumetric data set, wherein the volumetric data set corresponds to the DTM image file.
20. The method of claim 15 , including applying a three-dimensional Hessian matrix analysis to the DTM image file to detect lesion candidates.
21. The method of claim 15 , including applying a three-dimensional gradient field analysis to the DTM image file to detect lesion candidates.
22. The method of claim 21 , including determining a gradient vector at a smoothed voxel in a spherical region, and projecting the gradient vector to a radial direction from the center of the spherical region to the smoothed voxel.
23. The method of claim 22 , including determining a gradient field convergence to produce a three-dimensional gradient field image.
24. The method of claim 15 , including segmenting the volume of interest by one of a three-dimensional active contour method or level set method.
25. The method of claim 15 , including segmenting the volume of interest by a three-dimensional growing method.
26. The method of claim 25 , wherein segmenting the volume of interest includes using the location of high gradient convergence of high response as a starting point and allowing the object corresponding to the volume of interest to grow across at least a portion of the plurality of DTM image slices in the DTM image file.
27. A method for identifying a breast cancer lesion comprising using a processor for: using a digital tomosynthesis mammogram (DTM) image file having a plurality of DTM image slices; applying a computerized three-dimensional analysis to the DTM image file to detect a potential breast cancer lesion; averaging a plurality of image voxels to obtain a smoothed volumetric data set, wherein the volumetric data set corresponds to the DTM image file; Identifying a volume of interest; segmenting the volume of interest extracting one or more object characteristics from the object corresponding to the volume of interest, the object characteristics being one of a morphological feature, a gray level feature, a texture feature, a spiculation feature, a boundary sharpness, or a shape irregularity; invoking a first classifier to determine if the object corresponding to the volume of interest is a breast cancer lesion or normal breast tissue; and invoking a second classifier to determine if the lesion is malignant or benign if the first classifier determines that the object corresponding to the volume of interest is a breast lesion.
28. The method of claim 27 , including applying one of a three-dimensional Hessian matrix analysis or a three-dimensional gradient field analysis to the DTM image file to detect lesion candidates.
29. The method of claim 27 , including invoking a feature selection method to select the most effective features and reduce the dimensionality of the feature space for the classifier.
30. A method for determining whether a volume of interest in a digital tomosynthesis mammogram (DTM) image contains a malignant or benign lesion comprising using a processor for: using a background correction filtering algorithm for reducing dense tissue adjacent to a lesion in a volume of interest to estimate a low frequency background gray level image from a shell of voxels at a periphery of the volume of interest; segmenting the lesion in the volume of interest; extracting morphological, gray-level, texture, and spiculation features from the lesion and its segmented boundaries; invoking a classifier to determine at least one of whether the lesion in the volume of interest is malignant or benign, or an estimate of the lesion's likelihood of malignancy.
31. The method of claim 30 , including segmenting the lesion using a k-means clustering algorithm.
32. The method of claim 31 , including using a 3D active contour segmentation method that uses a K-means segmentation result as the initial contour.
