A SYSTEM AND METHOD FOR THE QUANTIFICATION OF CONTRAST AGENT

Abstract

The present invention relates to contrast-enhanced radiographic imaging, the quantification of contrast agent in tissue and the assessment of the radiographic image quality. The invention provides a radiographic system-agnostic method to assess tissue administered with a radio-opaque contrast agent. The method is a system-agnostic means to accurately quantify contrast agent content in normal tissue and in cancerous tissue from contrast-enhanced radiographic images, and to assess and verify image quality and the efficacy of a clinical assessment from these images.

Claims

1. A method to assess the uptake of a radio-opaque contrast agent in organ tissue(s) wherein low energy (LE) and high energy (HE) radiographic images (RI) of the organ tissue(s) are used to quantify the radio-opaque contrast agent content in the organ tissue(s).

2. The method according to claim 1 including a step of x-ray scatter correction of the LE RI and HE RI.

3. The method according to claim 1 including producing a thickness map(s) of the organ using the LE RI to determine a volumetric composition map.

4. The method according to claim 1, including quantifying the radio-opaque contrast agent only where interesting tissue has a thickness greater than zero, the interesting tissue including fibroglandular tissue and any lesions of the organ tissue.

5. The method according to claim 1 including adopting a relation that additional concentration of contrast agent relative to contrast agent in adipose material occurs only in interesting tissue, the interesting tissue including fibroglandular tissue and any lesions of the organ tissue

6. The method according to claim 1 including adopting a relation that the organ tissue comprises: interesting tissue which includes active tissue and a portion of non-active tissue, and adipose tissue which includes another portion of the non-active tissue and excludes all active tissue.

7. The method according to claim 6 including quantifying the radio-opaque contrast agent only where interesting tissue has a thickness greater than zero, the interesting tissue including fibroglandular tissue and any lesions of the organ tissue, and including adopting a relation that the total concentration of contrast agent in the active tissue is the sum of the concentration of the contrast agent in the non-active tissue plus an additional concentration of the contrast agent in the interesting tissue according to location in the RI.

8. The method according to claim 7 including adopting a relation that the concentration of the contrast agent in the non-active tissue is less that that of the concentration of contrast agent in active-tissue

9. The method according to claim 7 including determining the thickness of the interesting tissue by an adopting a relation that in the LE RI there is no difference between the concentration of contrast agent in the active tissue versus in the non-active tissue.

10. The method according to claim 6 including adopting a relation for the LE RI, but not for the HE RI, that the additional concentration of the contrast agent in the interesting tissue is zero.

11. The method according to claim 6 including determining the additional contrast agent concentration in interesting tissue from a non-active fatty reference pixel value found from the HE image invariant of the location in the HE RI.

12. The method according to claim 6 including correlating the additional contrast agent concentration in interesting tissue, but excluding lesions, with benign parenchymal enhancement to determine an objective measure of benign parenchymal enhancement that may be continuous, or categorised into two or more descriptive classifications.

13. The method according to claim 1 including assessment of CE RI image quality wherein an estimate of the uncertainty in the quantification of the additional contrast agent concentration in interesting tissue is made and incorporated in a CE RI specific quality measure.

14. The method according to claim 1 wherein image corrections are made to the raw LE and/or HE images according to image quality assessment and are used to produce a raw DE image, wherein the weighting factor is determined from a first difference between the linear attenuation coefficient of the interesting tissue and the adipose tissue in the HE RI, and a second difference between the linear attenuation coefficient of the interesting tissue and the adipose tissue in the LE RI and determining the ratio of the first difference to the second difference.

15. The method according to claim 7, including producing a map in visual form of the additional contrast agent concentration in interesting tissue according to location in the CE RI.

16. A system arranged to implement the method according to claim 1, including: an image input device to input image pixel values and corresponding locations in at least one LE RI and at least one HE RI; a data input device LE and HE imaging system and contrast agent data; a map producing device to produce a segmentation map, an organ thickness map, a scatter corrected LE and HE maps, and a volumetric organ density map; and a contrast agent uptake assessment subunit to quantify radio-opaque contrast content in the organ tissue.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0059] FIG. 1 shows iodine concentration in the normal tissue vs time after intravenous contrast agent administration estimated from measurements made on single-energy contrast-enhanced mammograms (Jong et al (2003))

[0060] FIG. 2A is a first low-energy (LE) radiographic image (RI);

[0061] FIG. 2B is a second low-energy (LE) radiographic image (RI);

[0062] FIG. 2C is a high-energy (HE radiograph image (RI);

[0063] FIG. 2D is an illustration of an intermediate step segmentation;

[0064] FIG. 2E is an illustration of an intermediate step of breast thickness estimate;

[0065] FIG. 2F is an illustration of an intermediate step of x-ray scatter-to-primary ratio (SPR) estimation in a LE image;

[0066] FIG. 2G is an illustration of an intermediate step of x-ray scatter-to-primary ratio (SPR) estimation in a HE image;

[0067] FIG. 2H is an ‘interesting tissue’ thickness map of the “interesting tissue” thickness h.sub.int(x,y);

[0068] FIG. 2I is a raw dual-energy image; and

[0069] FIG. 2J is additional contrast agent concentration in the interesting tissue map of the additional contrast agent concentration in the interesting tissue ΔC.sub.int(x,y), relative to concentration of contrast agent in non-active tissue so as to visually localize and quantify contrast agent uptake;

[0070] FIG. 2K is a additional contrast agent concentration in the interesting tissue map showing concentrations above a threshold level so as to threshold areas of maximum enhancement; and

[0071] FIG. 3. shows an overall workflow of system and method for CE RI quantification, where steps of image quality assessment and clinical decision support are shown in optional groups because they are each optional steps that represent additional embodiments when combined with the central quantification method.

DETAILED DESCRIPTION

[0072] By way of illustrative embodiment, prior to contrast agent quantification multiple pre-processing steps are performed to maximize the measurement accuracy.

[0073] System and Method Inputs

[0074] The system and method inputs are CE RI exam images and their image metadata. Unprocessed (“raw,” ‘For Processing’), or processed (‘For Presentation’) CE RI images, or both, may be assessed. In a preferred embodiment quantification of contrast agent content and generation of a ‘raw’ contrast-enhanced, or raw DE, image is made using the unprocessed images.

[0075] The quantification method is demonstrated in FIG. 2 by way of the left craniocaudal view from a CE DM examination. Low-energy (LE: 28 kVp, W/Rh) energy mammograms are shown in FIG. 2A and

[0076] FIG. 2B. In FIG. 2A a fold of tissue is visible at the edge and over the nipple can be seen compared to FIG. 2B. This tissue at the edge corresponds to the segmented region at the edge shown in FIG. 2D.

[0077] A high-energy (HE: 45 kVp, W/Cu) mammograms, shown in FIG. 2C. The LE and HE mammographs shown in FIGS. 2B and 2C respectively were acquired in rapid succession under a single breast compression 4 minutes after intravenous iodinated contrast agent administration. Other anatomical views, such as the mediolateral oblique view are typically acquired and can also be analyzed using the same methods.

[0078] In an embodiment, each of the CE RI images (e.g., LE, HE and DE) are analysed for the presence of motion artefact, such as using the methods described in PCT/162017/054382 except that the current approach is novel for taking into consideration requirements for using the LE and HE together. If substantial motion is found in the recombined image (e.g., DE), but with minimal or no motion in the source images (e.g., LE and HE), then motion compensation is applied, such as in the form of image registration between the source images. If one of the source images is found to contain motion artefact, then this image may be the image selected for transformation in the image registration process, while the sharper image will be used as a reference. If both source images are found to contain substantial motion artefact, then the step of motion compensation may be omitted on the basis that it is unlikely to improve either image quality or quantitative analysis. The following steps of analysis can be performed on either the images after motion compensation has been applied, or the original images, or both, which advantageously allows comparative analysis that can be used to determine uncertainty of interpretation from the uncorrected clinical images.

[0079] Image segmentation maps are generated for each anatomical view (i.e., CC & MLO) for each breast (right and left) using methods known to those skilled in the art, that identify breast tissue, muscle, and image background, for example the Otsu method or machine-learned segmentation. Estimation of the breast area in contact with the compression plates will be made. A model of breast thickness in the peripheral region that is not in contact with the compression plates will be applied to estimate the remaining breast tissue thickness as in PCT/162017/054382 except that the current approach is novel for taking into consideration requirements for using the LE and HE together. It is a preferred embodiment that the image segmentation map is produced from the raw LE image as this image type has the greatest contrast between breast image features.

[0080] Using the image segmentation map and image metadata that describes the acquisition conditions as an input, a breast thickness map is produced. The breast thickness map is an estimate of the total thickness of breast tissue at each image location (x,y). The breast thickness may be produced by methods known to those skilled in the art, including by applying the methods of PCT/TB2018/058663, where the accuracy of image parameters will be assessed to determine a best estimate of the breast thickness.

[0081] FIG. 2D shows an example segmentation of the unprocessed LE image in FIG. 2(a), which delineates an inner breast region in contact with the breast support and compression paddle (white) and a peripheral breast area (grey), where the thickness varies. The breast thickness map produced from these inputs is shown in FIG. 2E.

[0082] Using their associated segmentation and thickness maps, one or more mammographic images are transformed to a volumetric tissue composition map (e.g., by the Volpara density algorithm, or similar method). The set of images will include CE RI study images, but can also include prior conventional (non-contrast) images of the same subject, if available. The validity of, and uncertainty in, the CE RI-derived tissue composition data can be estimated through comparison with the tissue composition data derived from conventional images.

[0083] Integral to the derivation of the volumetric tissue composition map, but also for the generation of accurate DE subtracted images is accurate modelling of the image acquisition process. To develop this model, quantitative descriptive measurements are made from the mammographic images. This includes signal level measurements in one or more image regions. Reference descriptive image and subject data is also extracted from the image DICOM header, acquisition machine, or other database, which may include prior physical measurements or benchmark data. These data include x-ray spectrum information (anode/filter physical properties, kVp, mAs, exposure time, half-value layer, tube output), detector properties (including, but not limited to, detector conversion layer physical properties, detector element size, pixel size, pixel binning, detector pixel offset level, calibration status, number of image rows and columns); acquisition system characteristics (including, but not limited to, anti-scatter grid physical properties and performance characteristics; compression plate physical properties, detector cover physical properties, source-to-detector distance, breast support plate to detector distance; examination related data (including, but not limited to, compressed breast thickness, compression force, compression paddle tilt, contrast agent type and composition, method of contrast agent administration, injection rate, amount of contrast agent administered, start time of injection, image acquisition time).

[0084] In combination with the breast thickness map, these descriptive measurements are used to estimate the x-ray scatter field in each image. Any scatter estimation methods known to those skilled in the art may be applied, such as Monte Carlo modelling, machine-learned scatter models, and convolution with pre-computed scatter point-spread functions. In a preferred embodiment, pre-computed scatter point spread functions, as derived from Monte-Carlo simulations across a clinically relevant range of x-ray beam energies, breast tissue thicknesses and breast compositions, are convolved with the breast thickness map. The resulting scatter estimate is used to correct the image from which it was derived, and is repeated independently for each study image since the image acquisition parameters will differ between the images. For CE RI, a scatter-corrected LE and HE image are separately produced.

[0085] FIG. 2F demonstrates the estimated scatter-to-primary ratio (SPR) without an anti-scatter grid for the LE image. FIG. 2G shows the SPR for the HE image, also without an anti-scatter grid.

[0086] The scatter-corrected CE RI images, with or without motion correction applied, together with the thickness maps, segmentation maps, and volumetric breast composition maps are input to a model for contrast agent content analysis.

[0087] In an embodiment, the model is as follows. In a conventional mammogram (where there is no contrast agent in tissue), image pixel values, g(x,y), (after corrections for image offsets, x-ray scatter, and absorbed energy at the detector) are described by:

g(x,y)=λe.sup.−Hμ.sup.fe.sup.−h.sup.int.sup.(x,y)[μ.sup.int.sup.−μ.sup.f.sup.],

and;

g(x,y)=g.sub.fate.sup.−h.sup.int.sup.(x,y)[μ.sup.int.sup.−μ.sup.f.sup.],

where H represents the compressed breast thickness, h.sub.int (x,y) is the thickness of ‘interesting’ tissue (fibroglandular tissue and any lesions), A accounts for the detected x-ray signal, and μ.sub.int and μ.sub.f, represent the linear attenuation coefficients of interesting and fatty tissues, respectively.

[0088] Points are located in the mammographic image to estimate a reference fat pixel value, g.sub.fat, as described in PCT/GB2010/001472:

g.sub.fat=λe.sup.−Hμ.sup.f

[0089] Under nominal conditions, it is assumed in CE RI that the contrast agent concentration is effectively equal in non-active adipose tissue and non-active fibroglandular tissue.

[0090] But, in metabolically active tissue the total contrast agent concentration, C.sub.total, can be greater than in non-active (na) tissue, resulting in variable amounts of contrast agent across the image. Defining the additional contrast agent concentration to occur within the interesting tissue, we can describe the total contrast agent concentration at locations (x,y) across the image as follows:

C.sub.total(x,y)=C.sub.na+ΔC.sub.int(x,y),

where C.sub.na is the concentration of the contrast agent in ‘non-active’ tissue, and is modelled as a constant value in all tissues, while an amount of additional contrast agent concentration in the interesting tissue, ΔC.sub.int, can vary at pixel locations, (x,y).

[0091] In the case of CE RI of non-active tissue, assuming equal contrast agent concentration in interesting and fatty tissues we have:

g.sub.CA(x,y).sub.na=g.sub.fat_CAe.sup.−h.sup.int.sup.(x,y)[μ.sup.int.sup.−μ.sup.f.sup.]

[0092] Similar to the density estimation procedure in conventional mammographic images, a reference fat pixel value, but this time with a contrast agent (CA) present, “g.sub.fat_CA”, can be found in ‘non-active’ adipose tissue, which gives:

g.sub.fat_CA=λe.sup.−H(μ.sup.f.sup.+μ/ρ.sup.CA.sup.C.sup.na)

where μ/ρ.sub.CA is the mass attenuation coefficient of the contrast agent and the adipose tissue is assumed to be non-active in terms of additional contrast agent uptake. In one embodiment, the method of g.sub.fat_CA, point localisation follows the approach of PCT/GB2010/001472 except that the current approach is novel for taking into consideration effects of the contrast agent and insight into the effect of contrast agent in the LE RI versus the HE RI.

[0093] In the case of CE RI of a breast with active tissue, we observe an increased contrast agent attenuation relative to the fatty non-active reference, which can be described by:

g.sub.CA(x,y)g.sub.fat_CAe.sup.−h.sup.int.sup.(x,y)[μ.sup.int.sup.−μ.sup.f.sup.+μ/ρ.sup.CA.sup.ΔC.sup.int.sup.]

[0094] The non-active fatty reference pixel values, g.sub.fat_CA, can be found in each of the LE and HE images as:

g.sub.fat_CA_LE=λ.sub.LEe.sup.−H(μ.sup.f_LE.sup.+μ/ρ.sup.CA_LE.sup.C.sup.na.sup.)

g.sub.fat_CA_HE=λ.sub.LEe.sup.−H(μ.sup.f_HE.sup.+μ/ρ.sup.CA_HE.sup.C.sup.na.sup.)

wherein the amount of tissue, H, and the contrast agent concentration in the nonactive tissue, C.sub.na, are each assumed to be identical in the LE and HE views.

[0095] The equation for CE RI of a breast with active tissue, g.sub.CA(x,y), can similarly be expressed for each of the LE and HE images, including terms for the non-active fatty reference pixel values, as:

g.sub.CA_LE(x,y)=g.sub.fat_CA_LEe.sup.−h.sup.int.sup.(x,y)[μ.sup.int_LE.sup.−μ.sup.f_LE.sup.+μ/ρ.sup.CA_LE.sup.ΔC.sup.int.sup.(x,y)]

g.sub.CA_HE(x,y)=g.sub.fat_CA_HEe.sup.−h.sup.int.sup.(x,y)[μ.sup.int_HE.sup.−μ.sup.f_HE.sup.+μ/ρ.sup.CA_HE.sup.ΔC.sup.int.sup.(x,y)]

wherein the amount of interesting tissue, h.sub.int (x,y), and the additional contrast agent concentration, ΔC.sub.int(x,y), are each assumed to be identical in the LE and HE views.

[0096] In LE images, the relative sensitivity to contrast agent concentration is low, such that the x-ray attenuation due to normal tissue overwhelms the measured image signal. For example, in typical clinical conditions, around 5% of the image signal, or less, might be attributed to the presence of an iodinated contrast agent. Thus, the relative difference between the added signal from a higher concentration of contrast agent in active tissue compared to the baseline non-active tissue contrast agent concentration is expected to be marginal.

[0097] Thus, minimal error is anticipated if we assume additional contrast agent concentration in interesting tissue is zero, ΔC.sub.int(x,y)=0, such that g.sub.CA(x,y)≈g.sub.CA(x,y).sub.na in the LE RI, and then solve for h.sub.int by re-arranging the above LE image equations to give:

[00001]$h_{int}\left(x,y\right)=\log \left(\frac{g_{\mathrm{fat}\_\mathrm{CA}\_\mathrm{LE}}}{g_{\mathrm{CA}\_\mathrm{LE}}\left(x,y\right)}\right)/\left({\mu }_{\mathrm{int}\_\mathrm{LE}}-{\mu }_{f\_\mathrm{LE}}\right).$

[0098] This output, h.sub.int (x,y), represents an ‘interesting tissue’ thickness map, otherwise known as a density map, similar as described in PCT/GB2010/001472 except that the current approach is novel for taking into consideration effects of the contrast agent and insight into the effect of contrast agent in the LE RI versus the HE RI. In the above procedure, it is advantageous that, despite presence of contrast agent in the tissue in CE RI, knowledge of the non-active contrast agent concentration is not required to solve for the interesting tissue thickness h.sub.int(x,y). An example density map derived from the scatter-corrected LE image is shown in FIG. 2H where the greyscale represents the thickness of fibroglandular tissue.

[0099] To address many of the limitations in the prior art of measuring the contrast agent concentration from the recombined, or subtracted DE image, instead the relative increase in contrast agent concentration in interesting tissue, i.e. the additional contrast agent in interesting tissue ΔC.sub.int(x,y), is calculated from the HE image the non-active fatty reference pixel values, g g.sub.fat_CA_HE, together with the thickness of interesting tissue h.sub.int (x,y). Rearranging the equation for the non-active fatty reference pixel values in the HE RI g.sub.CA_HE(x,y) above, the additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) can be solved for as:

[00002]$\Delta C_{int}\left(x,y\right)=\frac{\frac{\log \left(\frac{g_{\mathrm{fat}\_\mathrm{CA}\_\mathrm{HE}}}{g_{\mathrm{CA}\_\mathrm{HE}}\left(x,y\right)}\right)}{h_{\mathrm{int}}\left(x,y\right)}-\left({\mu }_{\mathrm{int}\_\mathrm{HE}}-{\mu }_{f\_\mathrm{HE}}\right)}{\mu /{\rho }_{\mathrm{CA}\_\mathrm{HE}}}.$

[0100] An output, ΔC.sub.int(x,y) is an additional contrast agent concentration in interesting tissue as determined from the HE RI and the LE RI. The output enables production of an additional contrast agent map in visible form of the locations of, and amount of increased contrast agent concentration in the breast, relative to the contrast agent concentration in fatty tissue, at locations where the thickness of interesting tissue h.sub.int (x,y)>0. It is the active tissue which preferentially takes up the contrast agent and so where the increased contrast agent concentration is present. Active tissue includes tumours which occur in regions of interesting tissue which includes fibroglandular tissue and lesions some of which may be tumours. Hence the limitation that h.sub.int (x,y)>0 is reasonable and not onerous for the purposes of accurately quantifying contrast agent in normal tissue and in cancerous tissue from CE RI and to assess and verify image quality and the efficacy of a clinical assessment from CE RI.

[0101] In one embodiment, using additional model inputs that may include patient size, patient blood volume, patient heart rate, contrast agent injection rate, injection time relative to the CE RI image under analysis, breast compression, compression paddle type, and prior reference measurements such as from FIG. 1, a predictive estimate of the non-active contrast agent concentration could be made to permit estimation of the total contrast agent concentration, C.sub.total(x,y) However, due to the high uncertainty associated with many of these model inputs, it is an advantage of the method that an estimate of the amount and location of additional contrast agent concentration ΔC.sub.int(x,y) can be made without knowledge of the contrast agent concentration in non-active tissue C.sub.na. This improves image quality and clarity since any measurement or estimate of contrast agent concentration in non-active tissue C.sub.na would be inexact and introduce error and uncertainty.

[0102] A flowchart of the main CE RI quantification system and method inputs and steps are shown in FIG. 3.

[0103] In one aspect, several measures can be derived from the output related to active tissue. For example, the additional contrast agent concentration map, ΔC.sub.int(x,y), itself can be used directly as a visualisation of the increased contrast agent uptake in the interesting, or dense, tissue of the breast. Summary statistics about the marginal contrast agent concentration increase can be prepared, such as the average, standard deviation or maximum increased contrast agent concentration in dense tissue. These summary statistics may be used, alone or in a model to define quantifications and/or classifications of additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) that correlate with BPE in an objective manner.

[0104] In a second aspect, measures of the pattern, or complexity of the additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) distribution can be made. For example, by measurement of texture features of the additional contrast agent concentration map. The visual appearance of additional concentration in interesting tissue ΔC.sub.int(x,y) is anticipated to correlate with BPE, and with the potential for masking of lesion contrast-enhancement by BPE. Thus, a CE RI-specific image quality measure related to potential masking by BPE can be determined. More clarity and less blur are both objectives.

[0105] In a third aspect, the additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) map is analysed for information to either predict the presence of disease or characterise known diseased tissue. In one embodiment, manual contouring is used to define one or more regions of interest for quantitative evaluation, especially when disease is already suspected or confirmed. In a preferred embodiment the region(s) for analysis may be identified by automated means. For example, the additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) map can be thresholded to localise areas of interest, or be used in combination with computer aided detection or machine-learned lesion localisation tools to indicate regions with suspicion of disease.

[0106] The resulting additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) map from the quantification method applied to the illustrative CEDM case is shown in FIG. 2K. The additional contrast agent concentration in interesting tissue ΔC.sub.int(x,y) map clarifies where there is active tissue such as tumour tissue which preferentially takes up the iodinated contrast agent compared to normal tissue. Good agreement of the increased iodine concentration observed in FIG. 2J is seen compared to that expected based on FIG. 1. FIG. 2J demonstrates that by thresholding the map, areas of maximum enhancement can be localised, which may indicate an area of disease.

[0107] In an embodiment, the accuracy of the additional contrast agent concentration of interesting tissue ΔC.sub.int(x,y) estimate can be improved, for example by incorporation of contrast agent injection protocol information, such as the injected contrast agent volume and concentration, the injection start time relative to the image acquisition time, and patient factors, such as weight, estimated blood volume, and hormonal status. The use of these parameters, together with reference data on typical contrast agent tissue uptake also permits estimates of total contrast agent concentration, which will include the contrast agent content in the ‘non-active’ adipose tissue.

[0108] The result of DE image decomposition is effectively the HE image with a weighted proportion of the LE image signal removed, and the weighting selected to cancel the appearance of the normal tissue.

[0109] The normal tissue includes fibroglandular tissue which is non-active and fatty tissue. However, vendors may apply additional image corrections and processing, such as to compensate for tissue thickness variation and to maximize contrast enhancement. The image processing may interfere with quantitative interpretation, especially as different amounts of normal tissue suppression can be used in the DE images, which can result in substantially different visualisation of the tissue. The present invention entails a vendor neutral DE image that uses an optimal combination of the LE and HE images after motion compensation and scatter correction.

[0110] In an embodiment, a DE image weighting factor can be calculated as the ratio of the relative difference between the effective linear attenuation coefficients of the interesting and fatty tissue as:

[00003]$w_{DE}=\frac{{\left({\mu }_{int}-{\mu }_f\right)}_{HE}}{{\left({\mu }_{int}-{\mu }_f\right)}_{LE}}.$

[0111] The difference μ.sub.int−μ.sub.f is indicative of the light: dark contrast between interesting tissue and fatty tissue. A DE image can then be produced from the motion-corrected and scatter-corrected LE and DE images by combining the images with the weighting factor, w.sub.DE, as:

g.sub.DE(x,y)=g.sub.HE(x,y)−w.sub.DEg.sub.LE(x,y)

[0112] This DE image is referred to as a ‘raw’ DE image, as no processing for image enhancement has been applied. The ‘raw’ DE image shows the weighted ‘raw’ dual energy (DE) pixel values g.sub.DE(x,y). It is intended to clarify where there is active tissue.

[0113] Use of the raw DE image for diagnostic interpretation and quantitative CE RI analysis has the advantage that the data is vendor neutral and incorporates image corrections that minimize confounding factors for evaluation. It is anticipated that comparison of image quality measurements between the raw DE image and the vendor-supplied DE image will be useful to indicate potential image quality concerns, such as a difference in artefacts and masking potential between the images. Together with independent measures of the vendor-supplied DE image quality, the comparative analysis can be used to determine an overall score, or rating for the DE image quality. Image quality features of interest will include breast positioning, image noise, image contrast, presence of motion artefact, and measurements of the degree of normal tissue cancellation, based on measurements of ‘anatomical noise’ such as via power-law analysis and image texture. These latter measurements of anatomical noise will be included in a measurement and classification of BPE.

[0114] FIG. 2I is an example ‘raw’ DE image, computed using the scatter-corrected LE and DE images and the optimal DE image weighting factor calculated for this case.

[0115] This invention has been described by way of example only, 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 present invention.