Method and System for Image Normalization

Abstract

The present invention relates to a method and system for transforming an original mammogram into a standardized representation, wherein pixel values are independent of imaging conditions. The method performed includes: contrast enhancement for improving the visibility of breast tissue components, thereby segmenting regions of the breast and applying a contrast stretching algorithm to the segmented regions to preferably create an enhanced original image or mammogram; performing a local maximum transform, whereby a two-dimensional first filter is designed to extract the maximum pixel value from a region of interest (ROI) to preferably create a local maximum mapped image or map; and deriving a ratio map, whereby the pixel values of the ratio map are a measure of the relative response of the pixel relative to its local maximum, thereby capturing differences between breast components regardless of mammogram variations.

Description

Technical Field

[0001] The present invention relates to a method and system for transforming raw mammograms into a standardized representation, where pixel values are independent of imaging conditions. Background Art

[0002] Mammography is a diagnostic and screening process in which X-rays are used to examine the breast, mainly for breast cancer detection. Mammography can also identify abnormalities associated with other diseases, such as chronic kidney disease, reduced bone mineral density, diabetes, metabolic syndrome, hypertension, coronary artery disease, and stroke.

[0003] Breast screening requires obtaining two views of each breast, i.e., four mammograms per patient. Recently, to improve diagnosis, a series of two-dimensional (2D) projections have been captured and used to reconstruct three-dimensional (3D) images, which are called tomosynthesis or 3D mammography.

[0004] Digital mammography typically involves two types of images: "raw" images and "processed" images. "Raw" images are the acquired images with some technical adjustments such as pixel calibration and non-uniformity correction; "processed" images are images derived from the raw images that have been modified to improve spatial resolution (e.g., contrast equalization or enhancement).

[0005] Basically, mammography images are radiation attenuation maps, and due to the inconsistencies in the physical parameters (e.g., radiation dose, X-ray tube voltage, filtering material, etc.) applied to the imaging process to generate the raw images, the display of raw mammography images varies significantly between different imaging modalities and each mammography device (i.e., devices manufactured by different manufacturers).

[0006] CAD systems automatically detect and highlight the locations of potential cancers in mammograms. CAD systems use artificial intelligence techniques, pattern recognition techniques, or image processing techniques for automatic analysis, and then the analysis is reviewed by radiologists for final diagnostic decisions.

[0007] The purpose of CAD is to reduce the false positive rate and improve the sensitivity of mammograms. However, the performance of CAD is affected by the presence of breast arterial calcification (BAC). BAC is calcium deposits along the arterial walls in the breast. BAC is often observed on screening mammograms and is one of the most common false positives marked by CAD systems. Examples of BAC are shown in Figure 1 where, in Figure 1 (d), BAC is seen as a bright linear trace in the lower half of the breast.

[0008] The detection of BACs in mammograms is a challenging task, especially due to their topological complexity (BACs can cross and overlap each other). This topological complexity gives rise to various patterns, and this diversity makes it difficult to define a model to describe BACs.

[0009] In addition, the intensity values of BACs are non-uniform, and their appearance sometimes resembles healthy breast tissue. Therefore, a robust method is needed to distinguish and segment BACs from mammograms, so that CAD algorithms can achieve more consistent performance on raw images from different modalities and manufacturers.

[0010] The present invention provides such a solution. Summary of the Invention

[0011] The present invention relates to a method and system for distinguishing and segmenting BACs from mammograms, so that CAD algorithms can achieve more consistent performance on raw images from different modalities and manufacturers. The method can be used to standardize raw images. According to the method and system, the raw images are standardized by means of pixel values independent of imaging conditions: the filtered pixel values represent the relative amplitude between the pixel in the raw image and its adjacent pixels.

[0012] According to an aspect of the present invention, the method comprises the following three steps as given.

[0013] 1) Contrast enhancement , for improving the visibility of breast tissue components, thereby segmenting the region of the breast and applying a contrast stretching algorithm to the segmented region to preferably create an enhanced raw image or mammogram;

[0014] 2) Perform local maximum transformation , whereby a 2D first filter is designed to extract the maximum pixel value from the region of interest (ROI) to preferably create a local maximum map or mapping; and

[0015] 3) Derive ratio map , (wherein "ratio mapping" means the pixel-by-pixel ratio between the enhanced raw image and its local maximum mapping), whereby the pixel value of the ratio mapping is a measure of the relative response of the pixel with respect to its local maximum, thereby capturing the differences between breast components regardless of the mammogram variations.

[0016] The method is a means for standardizing raw images for distinguishing and segmenting BACs from other tissues in mammograms. The method reduces the false positive rate and improves the sensitivity of mammograms.

[0017] A standardized image can be derived from the original mammogram according to the above three steps. The standardized image can show the per-pixel ratio between the per-pixel contrast enhancement value from step 1 above and the corresponding local maximum from step 2 above.

[0018] In step 2, the local maximum can be derived by filtering the contrast enhancement value determined in step 1.

[0019] A local maximum map can be derived from the contrast-enhanced mammogram by using a local maximum transform to filter the contrast-enhanced mammogram. Preferably, the local maximum transform is implemented by a two-dimensional first filter.

[0020] A customized second filter can be applied to the ratio map to assist or perform the extraction of the BAC. Preferably, the customized second filter is of the type using Hessian-based multi-scale filtering. The customized second filter can be applied to the ratio map to generate a probability image of the BAC. A map or image showing the prediction of the binary BAC can be obtained by thresholding the probability image of the BAC.

[0021] The contrast stretching algorithm can include segmenting the breast region in the original image and scaling the pixel values in the breast region.

[0022] The contrast stretching algorithm includes segmenting the breast region in the original image and scaling the pixel values in the breast region. Preferably, the contrast stretching algorithm includes saturating the scaled pixel values in the breast region and / or determining the saturated scaled pixel values in the breast region. The contrast stretching algorithm can include performing a correction on the scaled pixel values of the original image. The original image can be a saturated image or an image in which the breast region has saturated scaled pixel values.

[0023] Preferably, the correction is performed using a low saturation parameter and / or a high saturation parameter indicating relatively low and relatively high pixel values. The low saturation value and / or the high saturation value can be the pixel values of the pixels at a preselected low percentile and / or a preselected high percentile of the pixels in the breast region sorted from lowest to highest according to the pixel values. The correction can be of the gamma correction type.

[0024] Gamma correction emphasizes the differences in pixel values, thus helping to distinguish them. For example, two pixels may have pixel values of 1 and 2 respectively in the original image. After gamma correction, the gamma-corrected image may have pixel values of 10 and 50 respectively for the same two pixels. Therefore, the contrast between the two pixels is "stretched". Preferably, gamma correction includes a calculation using a saturation parameter that quantifies a low saturation threshold and a high saturation threshold for pixels in the breast region of the original image.

[0025] Preferably, the low saturation threshold is the pixel value of a preselected low percentile of the pixels in the breast region. Preferably, according to the pixel values, the preselected low percentile is the lowest 0.5% or 1% or 2% or 5% of the pixels. Similarly, preferably, the high saturation threshold is the pixel value of a preselected high percentile of the pixels in the breast region. Preferably, according to the pixel values, the preselected high percentile is the highest 0.5% or 1% or 2% or 5% of the pixels.

[0026] The application of the contrast stretching algorithm can be gamma correction of the saturated original image, where the saturation parameter is determined based on the scaled pixels in the breast region.

[0027] According to another aspect of the present invention, there is provided a system for transforming an original mammogram including a breast image for standardized presentation, the system including a processor in communication with a memory, the memory being arranged to:

[0028] 1) Create a contrast-enhanced mammogram to improve the visibility of breast tissue components, where the region of the breast image in the original mammogram stored in the memory is segmented, and the processor is arranged to implement a contrast stretching algorithm applied to the segmented region;

[0029] 2) Create a local maximum map of the contrast-enhanced mammogram stored in the memory by performing a local maximum transform, where a two-dimensional first filter implemented by the processor extracts the maximum pixel value from the region of interest (ROI); and

[0030] 3) Use the processor to derive a ratio map and store the ratio map in the memory, where the ratio map includes the per-pixel ratio between the contrast-enhanced mammogram and the local maximum map stored in the memory, where the pixel value of the ratio map is a measure of the relative response of the pixel pair of the ratio map to its local maximum value in the original mammogram.

[0031] The system implements a method for creating maps and images independent of imaging conditions.

[0032] Preferably, the system includes a display for showing a ratio map or a normalized image derived from the map. Thus, various features in the breast region can be clearly seen using the system.

[0033] Preferably, the processor is arranged to implement a contrast stretching algorithm to map pixel values in the breast region to new values to create a contrast-enhanced mammogram such that a portion of the new values saturates at low and high intensities of the input data. Thus, the full range of pixel values between the high saturation limit and the low saturation limit can be "stretched" to see features within that range with clarity and contrast.

[0034] In the system, preferably, the processor is arranged to process a two-dimensional first filter on a contrast-enhanced mammogram of size m rows by n columns and divide it into overlapping ROIs; convert it to another representation with the same image size; and generate a local maximum map in which its pixels represent local maximum pixel values from adjacent pixels in the contrast-enhanced mammogram. In this way, the two-dimensional filter can extract the maximum pixel values from the region of interest.

[0035] Preferably, the processor is arranged to read pixel values of the ratio map ranging from 0 to 1 and rescale the pixel values to 8-bit, 16-bit, 32-bit or 64-bit integers for storage in a memory as a normalized image. In this way, the normalization magnifies the differences between breast compositions, so that the BAC has better contrast compared to other breast tissues.

[0036] Preferably, the processor is arranged to apply a customized second filter to the ratio map, where the second filter is a vessel enhancement filter of the type using Hessian-based multi-scale filtering, and the vessel enhancement filter is tuned to adapt to the characteristics of the BAC and the mammogram to obtain a measure of vessel probability to generate a probability image of the BAC. Thus, false alarms and misunderstandings by experts are reduced because the customized second filter enhances the tubular and elongated structures in the ratio map, which correspond to the BAC pattern.

[0037] Other features of the invention are disclosed in the claims. The invention will now be described by way of example only with reference to the accompanying drawings. Description of the Drawings

[0038] Figure 1 Shows the normalization of the original mammogram, where Figure 1 The normalized image of (d) is Figure 1 The contrast-enhanced mammogram of (b) (i.e., Figure 1 The contrast-enhanced version of the original mammogram shown in (a)) and Figure 1The per-pixel ratio between the local maximum mappings of (c), where (c) is obtained by filtering (b) using a local maximum transform implemented with a two-dimensional first filter. Figure 1 (b) to obtain Figure 1 (c), where the local maximum transform is implemented with a two-dimensional first filter.

[0039] Figure 2 Shows two pairs of images: the original mammogram Figure 2 (a) and the tomosynthesis combination Figure 2 (b), which are produced by imaging devices from the same manufacturer (top row); the original mammograms from the same modality but different manufacturers, namely, Siemens Figure 2 (c) and Philips Figure 2 (d); the ratio map normalization (bottom row) removes the differences between each pair of original images and further enhances the tissue structure.

[0040] Figure 3 Shows the BAC extraction using Hessian-based multiscale filtering. The mammogram is normalized to a ratio map 3(a); a custom second filter is applied to generate the BAC probability image Figure 3 (b); the binary BAC prediction is obtained by thresholding Figure 3 (b) to obtain Figure 3 (c) (blue). The corresponding ground truth mask (red) is shown in (d).

[0041] Figure 4 Shows the original image (A) and its breast region segmentation (B). Here, the original image is monochrome 2, so the pixel values in the breast region are less than those in the background region.

[0042] Figure 5 Shows the application of pixel saturation. In the scaled pixels in the breast region (see scaling equation (1)), the low saturation threshold and the high saturation threshold are determined by the top and bottom 1% of these pixels in (a). The saturation thresholds are applied to the entire image as in (b), resulting in a saturated image (c).

[0043] Figure 6 Shows gamma correction, in which, after the shown non-linear transformation, the correction maps low saturation to low output and high saturation to high output. The gamma correction applied to Figure 5 (c) results in Figure 6 (b).

[0044] Figure 7The local maximum transformation is shown. Taking a 3×3 image region (a) as an example, the local maximum transformation implemented by the first filter extracts the maximum pixel value from this region. The first filter replaces the central pixel value with the regional maximum value, as shown in (b). Figure 6 The local maximum transformation on (b) produces Figure 7 (c).

[0045] Figure 8 It shows the ratio image derived from the per-pixel ratio between Figure 6 (b) and Figure 7 (c), where the pixel values outside the breast region are set to zero.

[0046] Figure 9 It shows the workflow of normalizing the original mammogram into a ratio image. The ratio image (f) is the per-pixel ratio between the gamma-corrected image (c) and the local maximum image (d). The former is obtained from the gamma correction of the 2% saturation image (b), and the latter is derived by filtering (c) using the local maximum first filter. Detailed implementation

[0047] In the implementation, the method and system for differentiating and segmenting BAC from mammograms include the following three steps.

[0048] 1) Contrast enhancement , for improving the visibility of breast tissue components, thereby segmenting the breast region, and a contrast stretching algorithm is applied to the segmented region. This algorithm maps the pixel values in the breast region to new values such that a part of the data saturates at low and high intensities of the input data.

[0049] 2) Perform local "maximum" transformation , whereby a 2D first filter is designed to extract the maximum pixel value from the region of interest (ROI). An image of m rows and n columns can be divided into m×n overlapping ROIs, where each ROI is centered at the pixel position (m, n). Thus, the first filter converts the image into another representation with the same image size. Applying the first filter to the enhanced original mammogram produces the following image, in which its pixels represent the local maximum pixel values from adjacent pixels in the enhanced original mammogram.

[0050] 3) Derive ratio map , whereby the pixel value of the ratio mapping is a measure of the relative response of the pixel to its local maximum. This correlation captures the differences between breast compositions regardless of the mammogram variations. Since the range of the pixel values of the ratio mapping is from 0 to 1, rescaling to 8-bit or 16-bit integers obtains a normalized image.

[0051] The pixel values in the normalized image represent the relative magnitude between that pixel and its neighboring pixels in the original image regardless of the imaging conditions. Additionally, normalization amplifies the differences between breast components, so BAC has better contrast compared to other breast tissues. As Figure 1 shown, compared to the original mammogram ( Figure 1 (a)), the visibility of BAC is significantly improved in the normalized image ( Figure 1 (d)). This clarity helps in developing BAC detection algorithms.

[0052] Using the Frangi vessel enhancement second filter tuned to accommodate the characteristics of BAC and mammograms, the multi-scale second-order local structure (i.e., the Hessian matrix) is examined, and a measure of the vessel probability is obtained from the eigenvalues of the Hessian matrix. The second filter enhances the tubular and elongated structures in the ratio map, and the tubular and elongated structures correspond to the BAC pattern.

[0053] Refer to Figure 3 , image normalization ( Figure 3 (a)) amplifies the image gradient, thereby enhancing the contrast between BAC and the surrounding tissues. Applying the Frangi vessel enhancement second filter ( Figure 3 (a)) results in a BAC probability image ( Figure 3 (b)). Then, an adaptive thresholding algorithm extracts the final BAC mask from the filtered image ( Figure 3 (c)). The extracted BAC ( Figure 3 (c)) is in good agreement with the ground truth ( Figure 3 (d)) manually labeled by an experienced radiologist. The origin of the ground truth is objective and verifiable data.

[0054] Directly applying the Frangi vessel enhancement second filter on the original mammogram cannot enhance the BAC structure compared to other breast tissues, which further makes it impossible for the thresholding algorithm to segment BAC.

[0055] Another illustrative example of the three steps in this method is given below.

[0056] 1) Contrast enhancement

[0057] a) The breast region is segmented as follows:

[0058] b) A binary image is derived from the original mammogram image of Figure 4 (a) as shown in Figure 4 (b), where non-zero pixels indicate the position of the breast in the corresponding original image.

[0059] c) Scale the pixel values within the breast region of the original image to the range between 0 and 1, and then perform a logarithmic transformation using Equation (1).

[0060] Scaled Raw Image=log(Raw Image / 65535) (1)

[0061] d) Determine the image saturation thresholds, i.e., guided by the segmentation Figure 4 (b), sort the scaled pixels in the breast region of the original image Figure 4 (a) in ascending order based on the pixel values. Count these pixels from the higher values to the lower values, and the top and bottom 1% of the pixels are determined as the saturation thresholds. The illustration of determining the saturation thresholds from Figure 5 (a) is shown in Figure 4 (a) and Figure 4 (b). In the original image, the pixel values less than the low saturation threshold are set to the low saturation threshold, and the pixel values greater than the high saturation threshold are set to the high saturation threshold. The saturated image has a pixel value range between the low saturation threshold and the high saturation threshold, as shown in Figure 5 (b). For example, Figure 5 (c) is the saturated image of Figure 4 (a).

[0062] e) Gamma correction maps the pixel values of the saturated image to a non - linear range between a specific bottom output threshold and a top output threshold according to Equation (2) and Figure 6 (a). The bottom output threshold and the top output threshold are set as parameters that respectively produce the best contrast between adipose tissue and dense (fibroglandular) tissue, such as 0.01 and 1, and the gamma correction factor is set to 0.5. After gamma correction, Figure 5 (c) becomes Figure 6 (b).

[0063]

[0064] where I represents the saturated image pixel value; as shown in Figure 5 (a) and Figure 6 (a), low in and high in are the low saturation threshold and the high saturation threshold respectively, low out and high out are the specific bottom output threshold and top output threshold, such as "low output" and "high output" in Figure 6 (a), and γ controls the weight between the low output and the high output.

[0065] 2) Local maximum transformation

[0066] The two-dimensional first filter is designed to extract the maximum pixel value from a 5 mm × 5 mm region of interest (ROI). Applying this first filter to the gamma-corrected image produces an output in which its pixels represent the maximum pixel value from the ROI centered at that pixel location. A demonstration of this local maximum transformation is shown in Figure 7 (a) and Figure 7 (b), and the transformed Figure 7 (c) is shown in Figure 6 (b).

[0067] 3) Ratio map derivation

[0068] Here, a ratio map is used to describe the per-pixel ratio between the original gamma-corrected image and its local maximum map. For example, Figure 8 is the result of Figure 6 (b) (per-pixel) divided by Figure 7 (c). After scaling the floating-point ratio map to a 16-bit image, the final normalized image can be obtained. The entire workflow for deriving the normalized ratio map image from the original image is depicted in Figure 9 .

[0069] The pixel values in the normalized image represent the relative magnitude between that pixel and its neighboring pixels in the original image regardless of the imaging conditions. As shown in Figure 2 , the proposed normalization algorithm reduces the differences between mammograms from different modalities and manufacturers. Additionally, normalization naturally amplifies the image gradients, thereby enhancing breast tissue with sharp edges. This can further facilitate the extraction of important tissue features such as breast arterial calcifications.

[0070] The invention has been described only by way of example, and modifications and alternatives will be apparent to those skilled in the art. All such embodiments and modifications are intended to fall within the scope of the claims.

Claims

1. A method for transforming a raw mammogram including a breast image into a standardized representation, the method comprising the following steps: 1) Create a contrast-enhanced mammogram to improve the visibility of breast tissue composition, wherein, Segmenting a breast region of the breast image in the raw mammogram within the breast region, and scaling pixel values in the breast region by a contrast stretching algorithm, and mapping the pixel values in the breast region to new values by correction such that a portion of the new values saturates at low and high intensities of the pixel values; 2) Creating a local maximum mapping of the contrast-enhanced mammogram by performing a local maximum transform, wherein a 2D first filter extracts maximum pixel values from a region of interest (ROI); and 3) Deriving a ratio map including a per-pixel ratio between the contrast-enhanced mammogram and the local maximum mapping, wherein pixel values of the ratio map are a measure of the relative response of pixels of the ratio map to their local maxima in the raw mammogram.

2. The method according to claim 1, wherein In step 1), the breast region is segmented from the raw mammogram and a binary image is derived as follows, in which non-zero pixels indicate the position of the breast in the corresponding raw mammogram image.

3. The method according to claim 1, wherein, In step 1), pixel values within the breast region in the raw mammogram are scaled to a range between 0 and 1, and then a logarithmic transform is performed.

4. The method according to claim 1, wherein, The contrast stretching algorithm includes performing the correction on scaled pixel values of an image in which the breast region has saturated scaled pixel values.

5. The method according to claim 1, wherein In step 1), an image saturation threshold is determined such that, guided by the segmentation, scaled pixels in the breast region of the raw mammogram are sorted in ascending order based on pixel values.

6. The method according to claim 5, wherein, The correction includes a saturation parameter that quantifies low and high saturation thresholds for pixels in the breast region.

7. The method according to claim 6, wherein, The top percentile and bottom percentile of pixel values in the breast region are determined as saturation thresholds.

8. The method according to claim 7, wherein, The low saturation threshold and the high saturation threshold are in the range of 0.5% to 5% of the lowest and highest pixels of the pixel, respectively, according to the pixel value of the pixel.

9. The method according to claim 1, wherein, In the raw mammogram, pixel values less than the low saturation threshold are set to the low saturation threshold, and pixel values greater than the high saturation threshold are set to the high saturation threshold to form a saturated image.

10. The method according to claim 9, wherein, The saturated image has a pixel value range between the low saturation threshold and the high saturation threshold.

11. The method according to claim 9, wherein, The contrast stretching algorithm includes performing gamma correction on the saturated raw mammogram, wherein the saturation parameter is determined from scaled pixels in the breast region.

12. The method according to claim 11, wherein, Gamma correction maps pixel values of the saturated image to a non-linear range between a specific bottom output threshold and a top output threshold.

13. The method according to claim 12, wherein The low intensity and high intensity of the pixel value where a part of the new value saturates are the bottom output threshold and the top output threshold, and the bottom output threshold and the top output threshold are set to parameters that generate a higher contrast between adipose tissue and dense fibroglandular tissue compared to the contrast in the original mammogram.

14. The method according to claim 1, wherein In step 2), wherein the ROI is 5 mm × 5 mm.

15. The method according to claim 11, wherein, Applying the two-dimensional first filter to the gamma-corrected image produces an output, wherein the pixel of the output represents the maximum pixel value from the ROI centered at the pixel position.

16. The method according to claim 1, wherein The local maximum map can be derived from the contrast-enhanced mammogram by filtering the contrast-enhanced mammogram using a local maximum first filter.

17. The method according to claim 1, wherein, An image of size m rows and n columns is divided into m×n overlapping ROIs, each ROI centered at the pixel position (m, n); and the first filter converts the image into another representation with an unchanged image size.

18. The method according to claim 17, wherein The image of size m rows and n columns is the contrast-enhanced mammogram, and the other representation is the local maximum map.

19. The method according to claim 1, wherein Applying the first filter to the contrast-enhanced mammogram produces the local maximum map, in which the mapped pixel represents the local maximum pixel value from adjacent pixels in the contrast-enhanced mammogram.

20. The method according to claim 1, wherein The first filter extracts the maximum pixel value from a 3×3 image region and replaces the central pixel value with the regional maximum value.

21. The method according to claim 11, wherein In step 3), the ratio map describes the per-pixel ratio between the gamma-corrected image and the local maximum map.

22. The method according to claim 1, including arranging the ratio map to show the per-pixel ratio between the per-pixel contrast enhancement value from step 1) and the corresponding local maximum value from step 2).

23. The method according to claim 1, wherein, Applying a customized second filter to the ratio map to generate a BAC probability image.

24. The method according to claim 23, including thresholding the BAC probability image to obtain a map or image showing the BAC prediction.

25. The method according to claim 23, wherein The customized second filter is of the type using Hessian-based multi-scale filtering.

26. The method according to claim 23, wherein, The customized second filter uses a Frangi vessel enhancement filter tuned to adapt to the characteristics of BAC and mammogram to obtain a measure of vessel probability.

27. The method according to claim 26, wherein, The mammogram is the original mammogram or the contrast-enhanced mammogram.

28. The method according to any one of claims 1 to 27, wherein After scaling the ratio map to an 8-bit, 16-bit, 32-bit, or 64-bit image, a final normalized image is obtained.

29. The method according to any one of claims 1 to 27, wherein, When the range of the pixel values of the ratio map is from 0 to 1, rescaling to an 8-bit, 16-bit, 32-bit, or 64-bit integer image obtains a normalized range.

30. A system for transforming an original mammogram including a breast image for a standardized presentation, the system including a processor in communication with a memory, the memory arranged to: 1) Create a contrast-enhanced mammogram to improve the visibility of breast tissue composition, wherein, Segment the breast region of the breast image in the original mammogram stored in the memory in the breast region, and the processor is arranged to implement a contrast stretching algorithm to scale the pixel values in the breast region, and implement a correction to map the pixel values in the breast region to new values such that a portion of the new values saturate at the low and high intensities of the pixel values; 2) Create a local maximum map of the contrast-enhanced mammogram stored in the memory by performing a local maximum transform, wherein a two-dimensional first filter implemented by the processor extracts the maximum pixel value from the region of interest (ROI); and 3) Use the processor to derive a ratio map and store the ratio map in the memory, wherein the ratio map includes a pixel-by-pixel ratio between the contrast-enhanced mammogram and the local maximum map stored in the memory, and wherein the pixel value of the ratio map is a measure of the relative response of the pixel of the ratio map to its local maximum value in the original mammogram.

31. The system according to claim 30, including a display for showing the ratio map as a normalized image.

32. The system according to claim 30, wherein, The processor is arranged to implement a contrast stretching algorithm to map the pixel values in the breast region to new values to create the contrast-enhanced mammogram such that a portion of the new values saturate at the low and high intensities of the input data.

33. The system according to claim 30, wherein, The processor is arranged to process a two-dimensional first filter on the contrast-enhanced mammogram of size m rows and n columns and divide it into overlapping ROIs; convert it into another representation with unchanged image size; and generate a local maximum map in which its pixels represent the local maximum pixel values from adjacent pixels in the contrast-enhanced mammogram.

34. The system according to any one of claims 30 to 33, wherein, The processor is arranged to read the pixel values of the ratio map ranging from 0 to 1 and rescale the pixel values to 8-bit, 16-bit, 32-bit or 64-bit integers for storage in the memory as a normalized image.

35. The system according to any one of claims 30 to 33, wherein, The processor is arranged to apply a customized second filter to the ratio map, wherein the second filter is a type of vessel enhancement filter using Hessian-based multi-scale filtering, and the vessel enhancement filter is tuned to adapt to the characteristics of the BAC and the mammogram to obtain a measure of the vessel probability and thereby generate a probability image of the BAC.

Images