FAST PREPROCESSING ALGORITHMS FOR DIGITAL MAMMOGRAPHY CAD AND WORKSTATION

Abstract

A method and apparatus are disclosed for an image preprocessing device that automatically detects chestwall laterality; removes border artifacts; and segments breast tissue and pectoral muscle from digital mammograms. The algorithms in the preprocessing device utilize the computer cache, a vertical Sobel filter and a probabilistic Hough transform to detect curved edges. The preprocessing result, along with a pseudo-modality normalized image, can be used as input to a CAD (computer-aided detection) server or to a mammography image review workstation. In the case of workstation input, the preprocessing results improve the protocol for chestwall-to-chestwall image hanging, and support optimal image contrast display of each segmented region.

Description

STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

Not Applicable.

BACKGROUND OF THE INVENTION

The present invention relates generally to the field of medical imaging systems. Particularly, the present invention relates to a method and apparatus for preprocessing digital mammography images in conjunction with mammography CAD server and digital mammography workstation.

Digital mammogram preprocessing includes chestwall laterality detection; border artifact clipping; breast tissue segmentation; pectoral muscle segmentation; and image normalization. The results of the preprocessing are usually used by a computer-aided detection (CAD) server to detect abnormalities within the breast segmented areas of normalized mammogram images. The results of the preprocessing are also used as inputs for a mammography workstation, where the bright borders need to be clipped using the correctly identified laterality for a standard image hanging protocol. The separate segmentations of each region in the breast also improve the image contrast optimization or the intensity inversion on the mammography workstation.

The existing methods for breast segmentation are usually based on the 4 or 8 nearest neighbor pixels within a region that is grown from a seed point (see U.S. Pat. No. 6,091,841) or gradient threshold to determine inside or outside segmentation region (see U.S. Pat. Nos. 5,572,565 and 5,825,91). Processing using this type of algorithms is computationally slow. A region growing method or gradient method also only detects one connected region, so it can not handle a mammography cleavage view, which includes the medial portions of both right and left breast.

An algorithm for segmentation of the pectoral muscle is typically based on a single line Hough Transform to model the edge line between the breast tissue and the pectoral muscle (see U.S. Pat. No. 6,035,056). So the segmentation result cannot accurately represent a curve shaped pectoral muscle. A generalized Hough Transform can be used to model a curve shape; however its calculation is more expensive that the single line approach, resulting in slower processing time.

BRIEF SUMMARY OF THE INVENTION

This invention solves existing problems in image preprocessing for mammography CAD and mammography workstation display by utilizing faster and more accurate segmentation algorithms that automatically detects chestwall laterality; removes border artifacts; and segments breast tissue and pectoral muscle from digital mammograms.

The algorithms in the preprocessing device utilize the computer cache and a vertical Sobel filter to segment breast tissue; and a probabilistic Hough transform to detect curved edges. The preprocessing result, along with a pseudo-modality normalized image, can be used as input to a CAD (computer-aided detection) server or to a mammography image review workstation. In the case of workstation input, the preprocessing results improve the hanging protocol for chestwall-to-chestwall image alignment, and support optimal image contrast display of each segmented region.

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING

The top-level flow-chart for mammography preprocessing is shown in FIG. 1.

Laterality detection is described in FIG. 2.

Bright border edge detection is described in FIG. 3.

Full breast segmentation is described in FIG. 4.

Pectoral muscle segmentation is described in FIG. 5.

An example result of curved shape pectoral muscle segmentation is shown in FIG. 6.

DETAILED DESCRIPTION OF THE INVENTION

The mammography preprocessing steps, as shown in FIG. 1, include (1) detecting chestwall side, or left/right laterality of the mammogram image; (2) flipping the image along the vertical axial if the chestwall is not at left; (3) detecting bright or black gap between chestwall and the edge of the image; (4) segmenting breast tissue from the image background; (5) detecting the curved line separation between breast tissue and pectoral muscle; (6) use segmentation information to normalize the image to a consistent pseudo modality; (7) re-arranging the image so to produce mammography standard chestwall-to-chestwall hanging protocol.

As shown in FIG. 2, the idea of the detection of the chestwall side for left or right laterality is based on the assumption that the breast tissue intensity is mostly distributed at the chestwall side and the low intensity of the air background at the opposite side.

As shown in FIG. 3, continuing from the down-sampled image, and flipping the image if the chestwall detected as on right, use around 10% portion of the image from each side (left, right, top and bottom) to remove (mask) bright or dark gap around the image 4 edges.

FIG. 4 shows flowchart for a fast breast segmentation algorithm. The idea is to use all pixels along the chestwall as seeds to grow the region along the image pixels horizontally in computer cache, so to achieve fast segmentation speed. Multiple seeds also allow to detect multiple non-connected regions, such as, mammography cleavage view, which includes the medial portions of both right and left breast.

The Probabilistic Hough Transform (PHT) was first introduced by N. Kiryati, Y. Eldar and A. M. Bruckshtein in 1990. In the standard implementation of the Duda and Hart algorithm for Hough Transform, the (ρ, θ) plane is divided in N_ρXN_θ rectangular cells and represented by an accumulator array. The algorithm performed in two stages: incrementation (often referred as “voting”) and search stages. The execution of Duda and Hart algorithm requires O(M*N_θ) operations in the incrementation stage and O(N_ρ*N_θ) in the search stage, where M is the number of edge points. Thus the incrementation stage usually dominates the execution time of the algorithm. The difference between Standard and Probabilistic Hough Transform is that only m=aM edge points (m<M) selected at random guide the incrementation stage. As the number of operation at this stage is proportional to m*N_θ, significant computation savings result as m can be made much smaller than M. The idea to apply PHT to segmentations in mammograms is to detect objects that it is sufficient to compute the Hough Transform of only a proportion of the pixels in the image. These pixels are randomly chosen from a uniform probability density function defined over the image. In result, it returns line segments rather than the whole lines. The connected segments form a piece-wise linear curve shape for segmentation. The PHT concept is used in laterality detection, bright border edge detection and pectoral muscle segmentation in this invention, in all those cases, the longer line feature in the image that manifests itself as a significant peak in the accumulator array of the conventional algorithm should be, with high probability, detectable using just m edge points to guide accumulation in the proposed algorithm. FIG. 5 provides the detailed steps how to apply the PHT to detect curved line that separates breast tissue and pectoral muscle. FIG. 6 shows a detection result from this algorithm.

Claims

1. A method to perform image preprocessing for mammography CAD and workstation, comprising the steps of: detect left and right laterality of the chestwall in the mammogram; arrange the mammogram image so chestwall side align the left side; detect bright or black edges along 4-side of the image; segment breast tissue; detect separation curve line between breast tissue and pectoral muscle; perform pseudo-modality image normalization based on breast segmentation and pectoral muscle segmentation; clip edge, mask breast tissue, arrange images to chestwall-to-chestwall layout (hanging protocol).

2. The method of claim 1, wherein said step of detecting chestwall laterality, comprising: down sample of the mammogram image (e.g., 300 um); normalize the pixel dynamic range to a fixed bit (e.g., 16 bit); truncate high and low intensity to a value inside tissue range; calculate the sum of pixel values and the number of pixels of the left half and the right half of the image separately; compare if the left sum bigger than the right sum, and the left number smaller than the right number; yes detection result is that chestwall at left, otherwise compare if the left sum not bigger than the right sum, and the left number bigger than the right number, yes detection result is that chestwall at right, otherwise, the chestwall at neither side

3. The method of claim 1, wherein said step of masking bright or black artifacts around 4 side of the image, comprising: arrange (flip) the image if the chestwall at right; copy 10% of left, right, top and bottom portion of the image separately; apply Sobel filter with 3×3 vertical kernel to the left and the right portions; apply Sobel filter with 3×3 horizontal kernel to the top and the bottom portions; normalize the filtered image to 8 bit; threshold low 25% of the vertical filtered image; threshold low 12.5% of the horizontal filtered image; detect the left and right vertical edges; and the top and bottom horizontal edges using Probabilistic Hough Transform; form the edge mask from the detected edges.

4. The method of claim 1, wherein said step of segmenting breast tissue region, comprising: calculate the pixel value threshold; use all pixels along the chestwall grow regions along the horizontal line so use pixels in computer cache; stop grow if reach the pixels that the value is outside threshold.

5. The method of claim 1, wherein said step of detecting separation curve line between tissue and muscle, comprising: use Gaussian blur filter to smooth pixel profile of the breast skinline; filter the blurred image using Sobel filter with 3×3 vertical kernel; erode the filtered image to remove skinline edges; normalized the eroded image to 8 bit range; threshold low 6% of pixel value in the normalized image; use Probabilistic Hough Transform to detect the pectoral muscle edge; smooth the detected edge line; form the pectoral muscle mask using the detected line and the upper and right borders of the breast segmentation