This is a continuation-in-paπ application of Serial No. 08/344,144 filed 1 1/23/94 and Serial No. 08/530,791 filed 10/3/95. The present invention relates to digital x-ray imaging devices and specifically to digital x-ray mammography devices.

BACKGROUND OF THE INVENTION

Most x-ray imaging devices involve directing a beam of x-rays through an object onto a phosphor screen which converts each x-ray photon into a large number of visible photons. The visible photons expose a sheet of photographic film placed close to the phosphor thus forming an image of the attenuation of x-rays passing through the object.

There are several limitations to film-screen x-ray devices. A major limitation is that the film serves the combined purpose of both the image acquisition function and the image display function. In addition, the range of contrast or latitude of the film is too limited to display the entire range of contrast in many objects of interest. Because of the limited latitude and dual acquisition/display function of film, a film-screen x-ray is often overexposed in one area and underexposed in another area due to the thickness and composition variations of the object across the image. The gray-scale level of x-ray film has a sigmoidal response as a function of exposure which results in difficulties in distinguishing contrast differences at the extremes of the exposure range; that is, in the most radiodense and in the most radiolucent areas of the image.

Digital x-ray techniques have been proposed as a technology which replaces the phosphor/film detector with a digital image detector, with the prospect of overcoming some of the limitations of film-screens in order to provide higher quality images. A potential advantage of digital x-ray technology involves the separation of the image acquisition function from the image display function. Digital detectors also provide a much greater range of contrast than film and the contrast response function is linear over the entire range. This would allow a digital detector to more easily distinguish subtle differences in attenuation of x-rays as they pass through various paths of the object. Differences in attenuation due to thickness and composition variations across the object can be subtracted out of the digital data in the computer and the residual contrast can then be optimized for the particular viewing mechanism, be it film or computer monitor. The residual contrast differences can then be analyzed to search for things of interest. Other advantages of digital x-ray technology include digital image archival and image transmission to remote locations for viewing purposes.

Two recent patents disclose systems which digitally image a small area of the breast in order to facilitate needle placement for needle-core biopsy. In the system manufactured by LORAD Medical Systems and described in U.S. Patent No. 5,289,320 (issued 2/22/94 to Pellegrino, et al.), a light emitted phosphor screen is coupled to a CCD array with a commercially available lens system. In a system manufactured by Fisher Imaging Corporation and described in U.S. Patent No. 5,078,142 (issued 1/7/92 Siczek, et al.), light emitted from a phosphor screen is coupled to a CCD array with a fiber-optic taper.

Current digital x-ray devices have fairly limited resolution and so they are limited in their applications. What is needed is a better digital x-ray device.

SUMMARY OF THE INVENTION

The device includes an x-ray source approximating a point source which is, for a single full breast image, held effectively in a fixed position with respect to the breast. The device also includes an aperture and a digital x-ray detector, both of which are movable with respect to the source and the breast. A positioning system positions the aperture and the detector in harmony to provide overlapping x-ray beam paths, each path radiating radially from the x-ray source, through the apertures through the breast to the detector. Digital image data is acquired by a data acquisition system representing images of overlapping breast regions. A digital computer programmed with a stitching algorithm computes an image of the full breast or a large portion of the full breast from image data obtained with respect to the overlapping regions.

In a preferred embodiment two apertures and two detectors provide two simultaneous beam paths and the positioning system positions these apertures and these detectors to provide the overlapping beam paths.

In a preferred embodiment of the present invention, the x-ray detector CCD array comprises a scintillator comprised of a flat single crystal Csl crystal supported on an optically transparent support plate. The opposite surface, an x-ray illumination surface, of the crystal is coated with an x-ray transparent optically reflecting material to provide an x-ray scintillation sandwich having an optical mirror at the x-ray illumination surface of the Csl crystal. An optical camera is focused on the illumination surface of the Csl crystal. Our preferred camera utilizes a Schmidt optical system to focus the scintillator image on to a 1024 x 1024 pixel. In this preferred embodiment an optical grade adhesive is used at the x-ray illumination surface of the scintillator to attach the reflector and to reduce Fresnel reflections.

Alternatively, index matching fluid matched optically to the Csl crystal is trapped between the crystal and the optical mirror at the illumination surface of the crystal.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic drawing showing the principal elements of a preferred embodiment of the invention fabricated by inventors and their co-workers.

FIG. 2 is a drawing showing the principal parts of the x-ray detector assembly in the first preferred embodiment of the invention.

FIG. 3a, 3b and 3c show three methods of fabricating an efficient x-ray to a visible light converter.

FIG. 4 is a sketch of a second embodiment of the present invention.

FIG. 5 is a sketch of a third embodiment of the present invention.

FIG. 6A through 6E are views of a scintillator.

FIG. 7 shows how to focus the camera in a preferred embodiment.

FIGS. 8a and 8b show how to fabricate a preferred scintillator sandwich.

FIG. 9 is a schematic drawing showing elements of another preferred embodiment of the invention.

FIG. 10 is a side view of the FIG. 9 embodiment showing the different positions of the digital detector assembly.

FIG. 11a shows the different image pane positions of a preferred embodiment of the invention.

FIG. 1 lb shows the placement of x-ray attenuating material surrounding the breast.

FIG. 12a shows two positions of a variable thickness x-ray apodizer.

FIG. 12b shows the thickness dimension of the x-ray apodizer.

FIG. 12c shows the spatial distribution of x-ray exposure for each of two image panes.

FIG. 13 illustrates distortion correction of barrel distortion for an image pane.

FIGS. 14a and 14b illustrate the registration of the four distortion corrected image panes onto a single full grid.

FIG. 15 is a side view of one embodiment the invention which shows the different positions of the digital detector assembly.

FIG. 16 is an end view of the invention shown in FIG. 15 which shows the different positions of the digital detector assembly.

FIG. 17a shows the different image pane positions of the second preferred embodiment of the invention.

FIG. 17b shows x-ray paths for two camera positions.

FIGS. 18a and 18b show the front and side view of one embodiment of the invention.

FIGS. 19a and 19b illustrate the principle of the moving x-ray apodizer.

FIGS. 20a and 20b compares a film screen breast image to a stitched digital x-ray image of a female breast acquired with a preferred embodiment of the invention.

FIG. 21 shows a block diagram demonstrating the functioning of a preferred stitching algorithm.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENT

Preferred embodiments of the present invention are described below by reference to the figures.

First Preferred Embodiment

A schematic of the key elements of a first preferred embodiment of a digital ma mography device is shown in FIG. 1. The device consists of an x-ray source 40, a conventional breast compression mechanism 50, 51, and a digital detector system 54.

The preferred embodiment utilizes an x-ray source 40 which incorporates a standard Model Bl 10/M149 Varian Eimac x-ray generation tube with tungsten anode. High voltage power is applied to the x-ray source 40 with power supply 41. An x-ray filter wheel 84 has different x- ray filters, fabricated from aluminum, silver, iodine and rhodium, for example. A specific filter in the filter wheel 84 is automatically selected by filter wheel drive 86 which is linked to the central computer 72. This embodiment locates the x-ray source 40 at 0 elevation (relative position), aperture 45 at 15 cm, the breast tray 50, 51 at 60 cm, and the front surface of the scintillator 55 at 63 cm.

The digital detector assembly 54 displayed in FIGS. 1 and 2 consists of a scintillator assembly 55, and a//0.83 Schmidt optical system consisting of a field lens 56, flat mirror 58, aspherical Schmidt corrector plate 59, and spherical primary mirror 60 which focuses the light into a doublet lens 62, finally forming an image on a CCD array 63. The entire digital detector assembly 54 is enclosed in a sealed housing 68 to eliminate dust and ambient visible light.

The first preferred embodiment utilizes a commercially available Model KAF-1000 CCD array 63 (Kodak Corporation), as shown in FIG. 1, containing an array of 1024 x 1024 pixels. The size of each pixel is 24 microns x 24 microns resulting in a 2.5 cm x 2.5 cm imaging area. The Schmidt optical system provides a magnification ratio of 2.75 between the CCD array 63 and the front surface of the scintillator 55. The separation distance of about 3 cm between the breast 49 and the x-ray absorbing surface of the scintillator 55 produces a slight geometrical magnification of 1.05 of an object of the breast tray 50, 51. In addition, this separation distance contributes to the reduction of the background signal produced by x-rays scattered by the breast 49. This results in an equivalent pixel size of 66 microns x 66 microns at the breast tray 50, 51. The imaging area is then 6.7 cm x 6.7 cm at the scintillator assembly 55 and 6.5 cm x 6.5 cm at the breast tray 50, 51. The spectral response characteristics of the CCD array 63 are selected to provide the most efficient detection of the visible photons emitted by the scintillator assembly 55. Other magnification ratios can be used.

A drive/preamplifier electronics assembly 65 is provided at the CCD 63, with cables leading to external detector electronics assembly 70 which contains analog-to-digital conversion circuitry to convert the analog CCD data into 12-bit digital values at a 5 MHz readout rate. This digital data is then stored in the central computer 72. The central computer 72, a commercially available 586AT, is equipped with a high resolution display monitor 76, a very high resolution display monitor 77 to view the final images, and a media storage device 80 to store the images. A separate control panel 74 used to control the x-ray source 40 is linked with the central computer 72.

FIG. 3a, 3b and 3c discloses three methods for fabricating the scintillator assembly 55. Each x-ray photon striking the scintillator assembly 55 is converted into a large number of visible photons. The preferred embodiment allows a high percentage of x-rays to be absorbed by the scintillator, a corresponding high number of visible photons to be converted for each x-ray photon, and a high percentage of visible photons to exit the rear surface of the scintillator.

FIG. 3a displays a preferred method of fabricating the scintillator assembly 55. We use a 7 cm x 7 cm x 0.25 cm thick optically transparent scintillator crystal 37. The preferred scintillator material 37 is a thallium-doped cesium iodide Csl (Tl) crystal which is "optically surfaced" on both sides of the thickness dimension. Another possible scintillator crystal is thallium-doped sodium iodide. The fragile Csl scintillator 37 is bonded to a 0.25 cm thick sheet of optically transparent polycarbonate 38 to provide structural rigidity. A separate 0.1 cm thick sheet of polycarbonate 35 is coated with a visible high reflectance coating 36 such as a thin aluminum, in order to provide reflectance for visible light. The reflector coated side of the polycarbonate sheet 35 is bonded to the top of the scintillator 37. The polycarbonate sheet 35 is then machined to a thickness of about 0.025 cm in order to minimize the attenuation of x-rays passing through the sheet 35. We calculate that for 17 to 30 keV x-ray photons such as those used for x-ray mammography imaging, for example, that greater than 98% of the x-rays striking the scintillator assembly 55 pass through the polycarbonate sheet 35 and the reflector coating 36 and are absorbed in the first 200 microns of the scintillator 37 which converts each x-ray photon into a large number of visible photons. These visible photons are emitted into 4 π steradians. Photons hitting the reflector coating 36 are reflected back towards the optical system, thus effectively doubling the visible light collected by the CCD array 63. A visible light image representing the attenuation of x-rays through the sample 49 is therefore produced at the front surface of the scintillator 37.

FIG. 3b disclosed an alternate method for fabricating the scintillator assembly 55. This method involves coating a 7 cm x 7 cm x 0.25 mm thick sheet of beryllium or polycarbonate 1 14 with a thin layer of visible high reflectance coating 1 15 such as aluminum to provide efficient collection of visible photons. A 200 micron thin layer of dendritic cesium iodide 1 16 is then coated onto the reflector coated 1 15 side of the beryllium sheet. This allows a minimum number of x-ray photons to be absorbed by the beryllium and a very high number of x-ray photons to be converted each to a large number of visible photons. The reflector coating 115 helps to direct a high percentage of these visible photons towards the rear of the scintillator assembly 55.

FIG. 3c discloses a third method for fabricating the scintillator assembly 55. This embodiment provides a 200 micron thick layer of dendritic cesium iodide 118 on a 7 cm x 7 cm x 0.1 cm thick sheet of optical glass 1 19. A thin layer of visible light reflector such as

aluminum is then coated onto the dendritic cesium iodide. This allows a high percentage of x-ray photons to be converted into visible light and allows these visible photons to pass through the optical glass towards the rear of the scintillator assembly 55 with very little attenuation all of the dimensions in FIGS. 3a, 3b and 3c can be varied as desired.

Second Preferred Embodiment

A second preferred embodiment of the present invention, which can be made very small and portable, is described by reference to FIG. 4 and FIGS. 3a, b, and c. A 40 kV, 10 - 60 watt x- ray source 2 (such as is supplied by Fine Focus Corporation) provides an x-ray beam 4 which is collimated by collimator 6 to produce collimated beam 8. An object 10 to be x-rayed is placed in the path of beam 8. A portion of the x-ray photons in beam 8 pass through object 10 and produce an x-ray shadow image on Csl (Tl) scintillator 55. Scintillator 55 can be any of the three scintillators shown in FIGS. 3a, 3b or 3c. Scintillator 55 produces a visible light scintillator image corresponding to the x-ray shadow image. Lens system 12 in turn images the visible light scintillator image on to CCD detector array 14 of CCD camera 16.

The lens system we prefer for this embodiment is a NAVITAR f/0.95 25 mm CC TV lens and our choice of CCD cameras is a commercially available high-sensitivity, low noise miniature CCD camera such as those produced by Panasonic or Sony. A leaded glass window may be provided between scintillator 55 and CCP array 14 to protect x-ray sensitive electronic equipment.

Third Preferred Embodiment

A third preferred embodiment of the present invention as shown in FIG. 5 is an x-ray camera for examining integrated circuit chips. As in the second preferred embodiment we use an x- ray source 2 described above. The chips being examined are positioned at 20 in FIG. 5.

In this application the scintillator 55 is any of the scintillators shown in FIGS. 3a, 3b and 3c and described above. A flat fold mirror 15 directs light from the visible light scintillator image through lens system 12 (in this case a NAVITAR 25 mm f/0.95 CC TV lens) onto the CCD array of camera 16 (in this case a COHU 4915 CCD camera). Enclosure 19 is preferably a lead shielded light-tight aluminum enclosure. Fold mirror 15 permits camera 16 to be positioned out of the direct path of x-rays from source 2.

Csl Sandwich

FIG. 6a through 6e display, in greater detail than that of FIG. 3a, 3b and 3c, our currently preferred method for fabricating the scintillator assembly 55. It is very important to produce scintillators having a very good optical quality reflecting surface. This is a problem because producing a very flat surface on Csl crystals is difficult. We use a 7 cm x 7 cm 0.25 cm thick optically transparent single crystal scintillator 94. The preferred scintillator material is a thallium-doped cesium iodide Csl (Tl) crystal which is surfaced on both sides to the thickness dimension desired (in this case about 0.25 cm) using a diamond fly cutting procedure or any other procedure which produces an optical quality surface with less than about 100 angstroms of surface roughness. We then bond an optical quality polycarbonate plate 95, which is about 0.40 cm thick, to the Csl crystal as shown in FIG. 6e. We choose an optical grade adhesive 90 which is index-matched as well as possible to the Csl index of refraction. A preferred adhesive is Summers Labs UV74 mixed with 9-vinyl carbazole monomer which is cured with UV light. Its index of refraction when cured is 1.6. The polycarbonate plate 95 provides structural rigidity over the entire surface area of the crystal. The index of refraction of the polycarbonate plate (1.59) closely matches that of the Csl crystal and the adhesive closely matches both materials. Therefore, we minimize light scatter and other boundary interface artifacts in the final light image. Fresnel reflections at these interfaces cause losses through the sandwich as well as contribute to scattered light that can degrade image quality. A separate 0.1 cm thick sheet of polycarbonate 91 is coated with a thin layer of aluminum 92 to provide both very high reflectance of visible light within the crystal and stop any outside light from entering the crystal. The aluminum coated side of the polycarbonate sheet 91 is then bonded, using the same adhesive 90, to the top of the Csl crystal 94. Polycarbonate sheet 91 is then machined at the other side to a thickness of about 0.025 cm in order to minimize the attenuation of x-rays passing through the sheet 91. We calculate that for 17 to 30 keV x-ray photons such as are used for x-ray mammography imaging, for example, greater than 98% of the x-rays striking the scintillator assembly 55 pass through the polycarbonate sheet 91 and the aluminum coating 92 and are absorbed in the first 200 microns of the Csl crystal 94 which converts each x-ray photon into a large number of visible light photons. These visible photons are emitted into 4 π steradians and the photons hitting the aluminum coating are reflected back towards the optical system thus effectively doubling the visible light collected by the CCD array shown in FIGS. 1, 2, 4, 5, or 7. A focused, visible light image representing the attenuation of x-rays through the object being x- rayed is therefore produced at the surface between the scintillator 94 and the aluminum coating 92.

Essential to the usefulness of any general purpose scintillator is adequate structural integrity as well as resistance to any potentially damaging moisture while exposed to expected

environmental conditions. The Csl (Tl) and other related crystals are typically hygroscopic and therefore require a barrier between their outer surfaces and nearly all environments. We accomplished this sealing through the implementation of optical-quality polycarbonate plastic plates. Polycarbonate was chosen because its Coefficient of Thermal Expansion (CTE) in addition to its optical indexes is relatively close to that of Csl.

The substantially rigid polycarbonate plate 95 which is placed on the optical side of the sandwich is also designed to enhance the structural integrity as well as seal out the moisture. The plate is relatively thick (~4 mm) and is anti-reflection coated with coating 98 to minimize Fresnel reflections from its outer surface. As indicated by the following formula optical indices of adjoining materials should be closely matched to reduce unwanted reflections:

where nj = index of material, τ\ ₂ = index of material 2 and R is the Fresnel reflection.

For our Csl crystal, the index of refraction at the peak scintillation wavelength (of 550 nm) is 1.793. The index of refraction for our optical adhesive is 1.6. This gives a Fresnel reflection of about 0.4% at the x-ray illumination surface of the crystal. It is important that this reflection be kept low especially at this junction. The reflection here should preferably be kept less than about 0.5%. For some applications we have learned that the reflection problem can become acute if the Fresnel reflection exceeds about 1%.

The overall thickness of our preferred scintillator sandwich is slightly larger than 3.5 mm consisting of the following layers starting at the x-ray incident side:

Polycarbonate Top Layer 0.25 mm

Aluminizing Reflector Layer 0.01 mm

Optical Adhesive 0.05 mm

Csl Crystal 1.50 mm

Optical Adhesive 0.05 mm

Polycarbonate Bottom Layer 4.00 mm

Anti-Reflectant Coating 0.01 mm

Our single-crystal scintillator provides substantial advantages over prior art dendritic (needle- type) crystals. Better x-ray conversion is also possible due to the allowable thicker scintillator depth, before degrading resolution beyond a usable extent. Use of a single crystal (as opposed to needle-type crystal which must be very thin for good resolution) permits us to focus the optical portion of our camera system at the reflector - Csl interface 99 (in FIG. 6c). This provides an extremely good image with very high resolution.

Sandwich With Index Matching Fluid

FIGS. 8 A and 8B demonstrate another preferred scintillator sandwich incorporating the principals of the present invention. In this case the Csl crystal 122 is contained between polycarbonate base plate 120 and polycarbonate cover plate 121. Cover plate 121 as above is coated with a thin aluminum layer 128 to provide an x-ray transparent optically reflecting surface. The spaces 127 between the crystal and the reflecting surface 128 of cover plate 121 and between the crystal 122 and the base plate 120, are filled with an index matching fluid having an index refraction almost exactly matching that of the Csl crystal. We used in both spaces Cargille ηd = 1.70, B-series index matching fluid. The thickness of the fluid was about 20 μm microns compared to a crystal thickness of about 1.5 mm O-ring 129 assures a good seal. Note in FIG. 8b the thicknesses of the spaces 127 filled with the fluid is exaggerated. Note also we have emphasized the flatness of the mirror surface at the bottom of reflective layer 128 and the jaggedness of the upper and lower surfaces of Csl crystal 122 in order to indicate the importance of the index matching fluid in improving the optical performance of the sandwich. As indicated in FIG. 8b we focus our camera on the reflective surface 128 which provides a very precise image of all scintillations in crystal 122 including the light reflected off the mirror 128. Because of the close match of the fluid and the crystal, there are virtually zero reflections from the rough surface of the Csl crystal.

Focusing the Optical System

Each x-ray photon typically generates one scintillator spot as it is absorbed in the Csl (Tl) crystal. Referring to FIGS. 6a - 6e, the greatest absorption location is at the point of x-ray entrance into the crystal, just down stream of the reflecting aluminum surface 92 as shown in FIG. 6b. However, many x-ray photons are absorbed at greater depths into the crystal. Spot locations within Csl crystal 94 are depicted at 30 and 31 in FIG. 7 as representing scintillations from about 20 absorptions. Each of these produce real images. Mirror 92 produces virtual images of these spots as represented at 32 and 33 in FIG. 7. Our optical system focal plane is at the mirror - Csl crystal interface as shown at 99 on FIG. 7 and we have provided an optical system with a depth of field that includes about 86% of the real and virtual scintillation spots. As shown at 100 in FIG. 7, large number of lined up spots (real and virtual in scintillator 55, which would be representative of narrow holes in the object being x-rayed) are imaged as points on CCD array 40.

Stitching Images to Obtain Full Breast Image

A schematic of the key elements of a preferred embodiment of a full breast digital mammography device which has been fabricated and tested is shown in FIG. 9. The primary

components of the device consist of an x-ray source 3, a conventional breast compression mechanism 2, and a digital detector system 15.

X-ray source 3 incorporates a standard Model B 1 10 M149 Varian/Eimac x-ray generation tube with a tungsten anode. High voltage power is applied to the x-ray source 3 with a commercially available high voltage power supply 4. The ideal x-ray source has x-ray emitted from a point source of substantially small areal extent. Our x-ray source emits x-rays from an area typically 300 microns in diameter. An x-ray filter 34 fabricated from a moderate x-ray attenuator such as a 50 micron thick sheet of silver, for example, provides a narrower energy spectrum of x-rays to provide a higher quality x-ray image. This embodiment locates the x-ray tube 3 at 0 elevation, aperture at 32, cm, a breast tray 24 at 64 cm, and a phosphor screen 8 at 87 cm.

The breast 1 is compressed between the breast tray 24 and an adjustable breast compression paddle 2. This immobilizes the breast 1 during x-ray exposure, helps to equilibrate the x-ray exposure across the breast, and reduces the effects of x-ray scatter in the breast.

The digital detector system 15 consists of a phosphor screen 8, pellicle mirror 9, x-ray window 7 which is transparent to x-rays and opaque to visible light, and a Schmidt optical system 17. The entire digital detector assembly 15 is enclosed in a sealed housing to eliminate dust and ambient visible light. The pellicle mirror 9 is comprised of a 9 micron thick sheet of optical grade nitrocellulose stretched over a metal frame. The pellicle mirror 9 has a very thin (approximately 1 micron) layer of aluminum silicate deposited on the underside. In the preferred embodiment, phosphor material on phosphor screen 8 is a 10.5 cm x 7.7 cm sheet of terbium-activated gadolinium oxysulfide (Lanex, by Kodak Corporation). Each x-ray photon striking the phosphor screen 8 is converted into a large number of visible photons. A visible light image of the breast 1 is therefore produced on the phosphor screen 8. X-rays pass with minimal amount of attenuation through the pellicle 9, while the visible light image generated on the phosphor screen 8 is reflected from the underside of the pellicle mirror 9 directly towards the Schmidt optical system 17.

The visible light image is digitally imaged by a Schmidt optical system 17 which consists of a curved primary mirror 1 1, Schmidt aspheric corrector plate 10, doublet lens 12, and charged coupled device (CCD) array 13. Schmidt cameras are disclosed in Modern Optical Engineering, by Warren Smith. McGraw Hill, New York, 1990, pg. 446. These cameras are known for maximum light collection efficiency (the preferred embodiment has f number f/0.83) with a minimum amount of optical aberration.

The preferred embodiment utilizes a commercially available Model KAF-6300 CCD array 13 (supplied by Kodak Corporation) which contains an array of 2048 x 3072 light sensitive pixels. The dimensions of each pixel are 9 microns x 9 microns, resulting in 1.84 cm x 2.76 cm imaging area. The Schmidt optical system provides a de-magnification ratio of 6: 1 between the phosphor screen 8 and the CCD array 13. The separation distance of 23 cm between the breast tray 24 and the phosphor screen 8 produces a geometrical magnification of 1.35 between the breast tray 24 and the phosphor screen 8. In addition, this separation distance contributes to the reduction of the background signal produced by x-rays scattered by the breast 1. The corresponding pixel size is 54 micron x 54 micron at the phosphor 8 and 40 microns x 40 microns at the breast tray 24. This results in a 7.7 cm x 10.5 cm image at the breast tray 24.

A driver/preamplifier electronics assembly 14 is provided at the CCD array 13, with cables leading to external detector electronics assembly 16 containing analog-to-digital conversion circuitry to convert the analog CCD data into 12-bit digital values at a 5 MHz readout rate. Digital data from the CCD array 13 is stored in the computer 18 for display on the high resolution display monitor 20. Control panel 19 used to control the x-ray source is linked to the data acquisition functions through the computer 18. Parameters associated with control panel 19 are displayed on control display 21.

The x-ray source 3, aperture 6, and the digital detector assembly 15 move relative to the clamped breast 1 to sequentially image four individual sections or quadrants of the breast 1 as shown in FIG. 1 1a. The quadrants overlap by approximately 10% resulting in a final composite image which is 20.5 cm x 15 cm. FIG. 10 shows a side view of the different positions of the detector assembly 15. Motion of detector assembly 15 is parallel to the breast tray 24 utilizing a pair of LM Corporation Type HK ball slides 48 mounted on HK Corporation rails 44 which are in turn mounted on the frame (not shown) of the mammography unit. Motion control is accomplished with American Precision Industries stepper motors, 37488 power supplies and B341-01 brakes, and Warner R050 ball screw assemblies (not shown). A mechanical linkage system 5 sequentially positions the x-ray source 3 as the position of detector assembly 15 changes. Aperture 6 is sequentially positioned with respect to the detector assembly 15 and the x-ray source 3 by the linkage system 5 in order to confine the x-ray energy within the area defined by the phosphor 8.

FIG. 1 la demonstrates that a typical breast does not cover the entire area comprised on the individual image panes 1 through 4. The approximately 100 to 1 variation in x-ray intensity between the non-breast and breast regions results in two problems. The predominant problem involves excessive conversion of x-rays to visible light from the surfaces and edges of components in the Schmidt optical system 17. This scattered visible light commonly known

as "veiling glare" contributes to spatially dependent noise in the breast images. In addition, "blooming" or overexposure of the pixels in the CCD 13 which are in the non-breast region and close to the breast results in a spillover of electric charge into the neighboring pixels. These problems are alleviated in the first preferred embodiment by the placement of an x-ray attenuating material 26 in the areas around the breast 1 as shown in FIG. 1 lb. The preferred embodiment uses a bolus material commonly used for radiation therapy which is marketed under the name "Superflab." This material approximates the x-ray attenuation of the breast and adjusts the intensity of the x-rays passing through the non-breast region to an acceptable level.

X-ray dose limitations to the patient are a primary concern for a screening procedure such as mammography. The x-ray exposure level for each of the four image panes do not exceed the allowable exposure levels. However, because the image panes overlap slightly, the exposure levels in the overlap regions can exceed the average exposures by a factor of two, and a factor of four in the case of the region where the four panes overlap. We have developed an apodizer technique for reducing the exposure level specifically in the overlap region for each of the four image panes in order to reduce the total exposure in the overlap to an acceptable level. FIG. 12a, 12b and 12c illustrate an apodizer we have fabricated and implemented in our invention. The apodizer is placed at the aperture 6 as shown in FIG. 9. The preferred embodiment uses a variable thickness x-ray attenuator 30, fabricated from aluminum which is a moderate attenuator of x-rays, which is fabricated in the shape which resembles a picture frame. The outer dimensions of the variable attenuator 30 are 4.8 cm x 6.3 cm so that the shadow cast by x-rays passing through this apodizer line up with the edges of each image pane. The inner dimensions of the attenuator 30 are 10% smaller than the outer dimensions and the thickness dimension 32 is shaped so that the attenuation of x-rays passing through the attenuator 30 increases linearly as one moves away from the inner dimension of the attenuator. Attenuator 30 is surrounded by a separate attenuator of 0.5 cm thickness of lead which substantially attenuates x-rays in the periphery. The resultant x-ray exposure per image pane falls off linearly in the overlap region as shown in FIG. 12c. Attenuator 30 is centered with respect to each image pane before each x-ray exposure by moving the aperture 6 with the linkage system 5. FIG. 12c also demonstrates that the total exposure in the overlap region between two image panes, which is a sum of exposures of the individual image panes, is reduced to the average exposure level of each image pane, except where the four panes overlap, in which the exposure is reduced to twice the average exposure level of each image pane.

Data Acquisition

The breast 1 is positioned between breast tray 24 and breast compression paddle 2. Bolus material 26 is positioned around the breast 1 as shown in FIG. 1 lb. The x-ray exposure level is predetermined from exposure level charts dependent on breast size and composition. X-ray exposure is determined by a combination of x-ray tube voltage and current, and x-ray exposure time. Each image plane 1 through 4, shown in FIG. 1 1a, is sequentially imaged. Typical x-ray exposure times are 1 second per image pane and approximately 1 second is required to move the detector assembly from image pane to image pane. Therefore, approximately 8 seconds are required to image the full breast. For each image pane, the x-ray beam passes through the aperture 6, breast paddle 2, breast 1 , x-ray window 7, pellicle mirror 9, and strikes the phosphor screen 8. Visible light from the phosphor screen is collected on the CCD array 13 by Schmidt optical system 17. Digital data from the CCD array 13 is readout by the electronics assembly 16 while the detector assembly is moving to the next image pane.

The CCD camera output data is in a 1024 x 1024 pixel array in the illustrated embodiment. The value of the luminance from the CCD camera has a resolution of 12 bits or 4096 luminance values. These four image panes are stored in the computer as I ^raw _m ( Xj _, y _j )(ij = 1,N) (m = 1,4) where N = 1024.

Combining of Image Panes

Our device produces four digital x-ray image panes which overlap by approximately 10%. As shown in FIG.1 la, the area of these four image panes contain a breast and surrounding areas outside the breast. We describe here a computer processing procedure which we call "stitching" which identifies identical features in the overlap areas of the image panes and uses this information to register the panes in order to form a single seamless image of the full breast.

We implement the stitching procedure by the following five steps: 1) characterization and calibration of system, 2) preprocessing of the four image panes, 3) distortion correction of each image pane, 4) correlation analysis of at least a subsection of each overlap region to determine the vertical and horizontal offsets of each pane relative to the neighboring panes, 5) registration of the four image panes on a single grid using the offset information, and 6) adjustment of the pixel values in the overlap areas (blending) in order to provide a seamless image.

1) Characterization and Calibration of System

The digital mammography device is initially characterized in order to identify hardware imperfections in the x-ray source 3, phosphor screen 8, Schmidt optical system 17, and CCD array 13. These hardware imperfections include dead or weakly responding pixels in the CCD, electronic noise in the CCD array, spatial imperfections in the phosphor, spatially varying illumination of the x-ray source, gain variations of the pixels in the CCD array, distortions introduced by the optical system 17, and rotations of the detector assembly 15 as it moves to the four image pane positions. The characterization information of the hardware imperfections is used to preprocess the breast images produced by the mammography device in order to improve the registration of the four image panes to produce a seamless image and to maximize the final image quality.

The optical center of the Schmidt optical system 17 is determined as part of the assembly procedure of the optical 17. CCD array 13 is bonded to the doublet lens 12 to be reasonably symmetric with respect to this optical center. Finally, the pixel location corresponding to the optical center is determined and stored in the computer 18 as a two dimensional pixel location OC. For the preferred 1024 x 1024 CCD array 13, the optical center OC should be reasonably close to the pixel location (512, 512).

The characterization procedure requires the acquisition of eight dark field images and eight white field images for each of the four image pane positions shown in FIG. 11a. Dark field images are acquired with the x-ray source 3 turned off in order to characterize the electronic noise in the system. White field images are acquired by placing a 2 cm thick sheet of Lucite on the breast tray 24 and illuminating the phosphor 8 with x-ray source 3 in order to characterize the spatially varying illumination of the x-ray source 3.

We subtract a dark field frame from a white field frame to produce a residual frame and average eight of these residual frames together to form one calibration frame per image pane. Small imperfections in the phosphor 8, and dead or weak pixels in the CCD array 13, defined as having greater than 15% variation in luminance, are identified as defective pixels in the calibration frame. Defective pixels are corrected in the calibration frames by interpolation of eight nearest neighbors for point defects and six nearest neighbors for column or row defects. These four calibration frames (one for each pane position) are then stored in the computer 18 as I^m _. x-.y _j ) 0 _> j = 1, N) (m = 1,4) and the positions of each defective pixel are stored as a defect map. We also average the six million pixel values in each of the four calibration frames and store this information as MEAN _m (m = 1 ,4). A single dark field image is stored as I ^dark m (x _i , y _j ) (i, j = l _. N) (m = l,4).

Image distortions complicate the registration of the four image panes and are visible in the final image. These image distortions include pincushion or barrel distortion due to the Schmidt optical system 17 and distortions due to the imaging geometry such as relative rotation of the individual image panes arising from the position changes of the digital detector assembly 15. An example of barrel distortion is shown in FIG. 13. The stitching algorithm characterizes the image distortions and uses this information to correct each breast image pane. The characterization is accomplished with a 24 cm x 18 cm calibration grid comprised of a two-dimensional array of 1 mm diameter copper dots spaced 0.5 cm apart on a standard electrical circuit board. The calibration grid is placed on the breast tray 24 so as to fully cover all of the four pane positions. An x-ray image is acquired at each of the four image pane positions.

For each image pane, the pixel coordinates (XJ , y _j ) (i,j = 1 ,N) of the calibration grid that can be related to observed distorted pixel coordinates (Xj, Y _j ) (i,j = 1,N) observed by the CCD array 13. The preferred embodiment uses a third-order polynomial mapping

j = ao + aj xj + a2 j + a3 Xj ² + --4 xj yj + a5 j ² +--6 yj ² a7 xi yi ² + &% xj ² + yi + ag yi ³ Yj = bo + bι Xj + b2 yi + b3 xj ² + b4 xj yj + bs j ² +b6 yi ³ + b ₇ x- yi ² + bs xj ² + yi + b9 yi ³ (1)

where a„ (n = 0, 1, 2,...,9) and b _m (m = 0, 1, 2,...,9) are the distortion calibration constants. Reasonable accuracy can be achieved with a third-order polynomial fit in (XJ, y _j ) as displayed in equation (1), although higher accuracy can be achieved by including more orders in the polynomial fit. The distortion calibration constants a _j (i = 1, 2,...,9) and b _j (i = 1, 2, ...,9) depend on the image distortion produced by the Schmidt optical system 17, distortions and rotations introduced by moving the digital detector assembly 15 to the four pane positions, and also the overall position and rotation of the calibration grid with respect to the CCD pixel array when the calibration grid is placed on the breast tray 24. A rotation angle for each image pane is extracted from equations (1) and the overall rotation angle of the calibration grid is obtained by summing the rotation angles of the four image panes and dividing by four. This overall rotation angle is removed from the measured data (Xj, Y _j ). The distortion constants a _n (n = 0, 2,...,9) and b _m (m = 0, 1, 2, ...,9) are derived with respect to the independently measured optical center OC of the Schmidt camera assembly 17. Separate sets of distortion constants are derived for each of the four image pane positions. These distortion constants are calculated to subpixel accuracy using a standard least-squares procedure which is described in detail in Section 3.6 (pg. 61-75) of "Digital Image Warping," by George Wolberg, IEEE Computer Society Press, Los Alamitos, 1990. The distortion constants a _n (n = 0, 2,...,9) and b _m (m = 0, 1 , 2, ...,9) for each of the four image panes (72 values) are stored in the computer 18.

2) Processing of Images

We acquire four image panes of the breast called I ^ra _m ( ^χ i> yj) (i.j = 1 ,N) (m = 1, 4) as discussed in the data acquisition section. Dead pixels in each image, identified by the defect map, are corrected by the interpolation procedure discussed in the calibration section. The images are then corrected for gain variations of the CCD pixels and spatial variations of the x-ray source by a procedure commonly known as "flat fielding,"

1° - (xi _. Yj) (i.j = 1,N) (m = 1, 4) = [I™*.. _{(x i} , _{yj )} . ιdark _{m (χ j> yj )} ]*MEAN _m /lca- _{m(xi r yj ) (2)} In equation (2), I ^cor _m (xj, y _j ) (i,j = 1,N) ( = 1,4) are the corrected image panes which are stored in the computer 18, * denotes a scalar multiplication, and / denotes a pixel-by-pixel division.

The flat fielding procedure also corrects for the spatially varying x-ray exposure of each image pane in the overlap region. The luminance values in the overlap regions of the raw images I ^ra _m (XJ, y _j ) fall off linearly from as one approaches the edge of each image pane due to the variable attenuator 30. From equation (2), the flat fielding procedure automatically adjusts the luminance values in the overlap regions to produce an image I ^cor _m (xi, y _j ) of each image pane which appears to be taken with no variable attenuator 30 in place. The flat fielding procedure effectively removes the effects of the variable attenuator 30 by adjusting the luminance values of the pixels in the overlap regions to appear that if the breast 1 were removed these pixels would each receive an equivalent amount of x-ray photons.

Due to slight variations in the total x-ray exposure per image pane, the four image panes exhibit an average luminance which varies from image pane to image pane. To correct for this effect for image pane 1 and image pane 2, we first calculate average luminance values G i and G ₂ in the overlap region of I ^cor ₁ (x,, y _j ) and I ^cor ₂ ( j. yj), respectively, by summing the luminance values of the pixels in the overlap region of each respective image pane and then dividing by the number of pixels in the overlap region. We then normalize image pane 2 to image pane 1 by calculating

porm _{2 (xij yj) = [Gl} /G ₂ ] * I ^cor ₂ (Xi, y _j )- (3)

We repeat this procedure by normalizing image pane 3 to image pane 2 and then image pane 4 to image pane 3 so that all four image panes appear to have been acquired with the same total x-ray exposure per image pane.

3) Distortion Correction of Images

Each image Inorm _m ( _Xjj yj) j _s corrected for distortions and relative rotation by mapping the distorted image I ^norm _m (XJ , yj) onto a regular grid using the distortion constants a _n (n = 1, 2,...,9) and b _m (n = 1, 2,...,9) and the optical center location OC stored in the computer 18. The distortion correction, shown graphically in FIG. 13 for the case barrel distortion, is accomplished such that the location of the optical center OC remains fixed. The distortion correction also rotates each image pane such that the axes of each image pane are parallel to each other. The location of each distortion corrected pixel location generally falls between the regularly spaced pixel locations. We utilize a bilinear interpolation in order to remap each corrected point to a point on the regular grid. This procedure is discussed in detail in Section 3.5.1 (pg. 58-61) of "Digital Image Warping," by George Wolberg, IEEE Computer Society Press, Los Alamitos, 1990. Higher accuracy can be achieved by using a higher-order Lagrange polynomial interpolation as discussed in Section 3.6 of the same text. The distortion corrected images for each image pane 1 through 4 are stored in the computer 18 as II (x_, yj), l2(*i, yj), i, yj), and I ₄ (xi, yj) (i,j = 1 ,N). Each image I _m (xi, yj) is an N x N matrix of 12-bit digital values.

4) Determination of Relative Offsets

We have now produced four image panes which are corrected for distortions and are rotated such that their image axes are parallel to each other as shown in FIG. 14a. Accurate registration of these images requires a determination of the relative offsets in the x and y directions of each image pane relative to the other image panes. We then position each image pane in the computer 18 such that the four image panes correctly align with one another to form a full seamless image as shown in FIG. 14b. The preferred embodiment calculates the offsets by cross-correlating fiducial marks 81 which are located at the edges of each image pane as shown in FIG. 14a. Fiducial marks with sharp edges in both directions, such as the words FULL BREAST DIGITAL IMAGE, for example, are preferred for the correlation procedure.

We begin by determining the relative integer offsets between image pane 1 and image pane 2. Image pane 1 and image pane 2 are mathematically defined as I] (XJ, yj) and l2(xj, yj), respectively, where xi = Δx*i and xj = Δy*j (i,j = 1,N) where Δx and Δy are the dimension of each pixel (Δx = Δy = 40 microns at the breast case for the preferred embodiment). We calculate the correlation function C (xm, yn) between image pane 1 and image pane 2 as

C(χ _m ,y _n ) = + X m'yj + ^χ _m .yj + yn ) (4)

C. m- yn) (m,n = 1, N) is a two-dimensional grid of numerical values. C(x _m , y _n ) has a sharp maximum value C(xι', yj') at the pixel location (x _\ , y _\ '). This location is also denoted as integer values (ii', ji') where the two image pane align themselves.

The value C(xι', y i') provides the nearest integer offset (x ι\ y ι') between image pane 1 and image pane 2. Higher registration accuracy can be obtained by calculating a residual fractional offset (dx i', dy i') which further maximizes C(x ι', + d\ _\ ', y _\ ', dy )- We use an inteφolation procedure which fits C(x j', + dx , y _\ ', dyj ') to a quadratic function and then calculates the coordinate xj', + d\ at which C(x j', + dx i', yj') is maximized while holding the y-axis constant at y i'. We then calculate the coordinate yi', dyj' at which C(x j', + y _\ dyj') is maximized while holding the x-axis constant the coordinate yj', dyj' at which C(xι', + , y _l '. dy i') is maximized while holding the x-axis constant at i'. Following this procedure, the fractional offset dxj', dy y). Following this procedure, the fractional offset (dx _\ dyi') is calculated as:

where (x _\ , y ) is the integer offset.

We calculate the integer offset (X ₂ ', + dx ₂ ', y ₂ ', dy ₂ ') and (X ₃ ', + dxβ', y ₃ ', dy ₃ ') between image pane 2 and image pane 3, and image pane 3 and image pane 4, respectively in the same manner by correlating the fiducial marks 81 located in the overlap region between these respective panes. The offsets dyj') (i = 1,3) are stored in the computer 18.

5) Registration of Images

We now register the four image panes on a full breast grid iF .x-.y _s ) (r,s = 1, 2N) which has 2N x 2N regularly spaced grid points. We begin by mapping image pane 1 given by I ι(xj,y _j ) (i = 1 ,N) directly to the upper left corner of the full grid such that iF .Xi.yj) = I ι(xi,y _j ) (i,j = 1,N) as shown in FIG. 14a. We then map image pane 2 given by I ₂ (xi,y _j ) (i,j = 1,N) onto the full grid. This is done in two steps: 1) remapping of the fractional offsets (dxj', dyj') and 2)

mapping of the integer offsets (XJ ', yj'). We first correct for the fractional offsets (dxj', dyj') by remapping Ϊ2( i- xj', y _j '-dyj') (i,j = 1,N) which generally falls between the pixel locations onto a regularly spaced grid I2 (xj-yj) (i,j = 1,N). This is done by calculating the value of I2 (xj-yj) at the integer pixel location (xj-y _j ) nearest to the non-integer pixel location (xj- xj', yj, dyj'). The preferred embodiment calculates this with a bilinear inteφolation procedure as discussed in detail in Section 3.5.1 (pg. 58-61) of "Digital Image Waφing," by George Wolberg, IEEE Computer Society Press, Los Alamitos, 1990. Higher accuracy can be achieved by using a higher-order Lagrange polynomial inteφolation as discussed in Section 3.6 of the same text. We then map image pane 2, 12 (xi-yj) (i,j = 1,N), onto the full grid such that I _F (xi,yj) = l2(xi-N*Δx +x +l,yj +y +l) (i = N= i , ^, ,2N;j = -j ,N).

We repeat the registration procedure in a similar manner for image pane 3 and image pane 4. First we correct for the fractional offsets (dx2', dy ₂ ') and (dx3 ^r , dy3') by remapping I3(xj-dx2', yj-dy2') (i, j = 1, N) and -_ ₄ (xj- x3', yj-dy3') (i, j = 1, N), each which generally falls between the pixel locations, onto a regularly spaced grid I3(XJ, yj) (i, j = 1, N), onto the full grid such that I _F (XJ, yj) = I ₃ (XJ-N*ΔX +xι'+l .yj -N*Δy + y ι '+l) (i = N-ij', 2N; j = N-jι\2N). We then map image pane 4,

L. ( i, yj) (i,j = 1,N), onto a full grid such that I _F (xi, yj) = l4(Xi+x ι' + 1, yj-N*Δy+y ι'+l) (i = -i _l ', N;j = N-jι',2N). The final image is then fully registered as displayed in FIG. 14b.

6) Blending

The luminance values of the pixels in the final image are finally adjusted to provide a seamless image. As discussed in preprocessing section, the luminance values of each image pane have been normalized to each other. A sum of the two luminance values at the pixel location where two image panes overlap produces a final image where the luminance values in the overlap regions are a factor of two higher than the luminance values in the non-overlap regions. A simple blending procedure involves dividing the luminance values in the overlap regions by a factor of two to produce a seamless image.

We blend together the overlapping regions by a slightly more elaborate algorithm. We linearly weight the sum of the luminance values of the pixels in the overlap regions between image pane 1 and image pane 2 to provide a 100% weight to pixels of image pane 1 at the side of the overlap region closest to image pane 1 and a 100% weight to pixels of image pane 2 at the side of the overlap region closest to image pane 2. This is mathematically expressed as I _F (xj,yj) = [(N-iV ri xj.yj) + [(iι ^, -N+i)i, ^, ]*I ₂ (xiN+xι',yj+yl') (i = N-ij', 2N;j = -j f,N). This procedure seems to produce a higher quality image with a less noticeable overlap region. We repeat this procedure in a similar manner for the overlap region between image pane 2 and image pane 3 and the overlap region between image pane 3 and image pane 4.

7) Image Enhancement

The seamless digital image I _F (x _r , y _s ) (r,s = 1, 2N) of the full breast is processed in the computer 18 in order to optimize the contrast between features in the breast. The preferred image enhancement procedure involves a preferential enhancement of the high spatial frequency components of the image while maintaining a good balance between the low and high spatial frequency components. The enhancement procedure starts with a logarithm transform Ii _og (x _r , y _s ) = In[I _F (x _r , y _s )] (r,s = 1, 2N) of the image in order to reduce the contrast differences of the image and to linearize the computer calculations. The high spatial frequency components of the image Iι _0g (x _r , y _s ) are enhanced by a procedure commonly known as "unshaφ masking" where we subtract a fraction of a blurred image from the original image, I _en h = I-og - I.iurred, where typically a = 0.25. Ibiurred obtained by successively convolving the image I] _og (x _r ,y _s ) with three top hot pixel distributions of various sizes (preferred top hat pixel dimensions are 100 x 100, 75 x 75, and 125 x 125). This procedure is described in Section 7.4 of "Fundamentals of Digital Image Processing," by anil K. Jain, Prentice Hall, New Jersey, 1989.

The contrast of the image I _e - _h ( ^χ _r , y _s ) i adjusted to produce the final image. The preferred procedure uses a sigmoid function, I _f]na i (xr.y _s ) = P-exp (-σϊl _enh frr.ys M)" ¹ where M is the mean value of the image I _enh ( r.y _s ) and σ is a gain parameter. The parameters M and s are presently chosen by visual inspection of each image.

A block diagram showing the principal steps of our process for stitching digital images is shown in FIG. 21.

Second Preferred Full Breast Embodiment

A schematic of the key elements of a second preferred embodiment of a full breast digital mammography device is currently being fabricated by the inventors and their fellow workers and is shown in FIGS. 1, 15 and 6.

The second preferred embodiment uses two identical digital detector assemblies 54 which simultaneously image different areas of the breast in order to reduce the imaging time. FIGS. 15 and 16 show the side and end view of the detector assemblies 54. The x-ray source 40, aperture 45, and the two digital detector assemblies 54 move relative to the clamped breast 49 (shown in FIG. 1) to sequentially image sixteen individual sections (eight sections per camera) of the breast 49. FIGS. 15 and 16 show the different positions of the two detector assemblies 54. The front surface of the scintillator assemblies 55 in the two digital detector assemblies 54 move along a plane while the front surfaces of the scintillator assemblies 55 tilt

so as to always remain normal to the x-ray paths. This feature results in improved image quality.

FIGS. 16 and 17 show the different positions of detectors 54. The two detectors 54 are rigidly mounted to a frame 122 which moves with LM Coφoration Type HK ball slides 88 mounted on HK Coφoration rails 89 which are mounted to the mammography unit frame. A separate drive 46 (shown in FIG. 1 ) is used to sequentially position the x-ray source 40 as the position of each detector assembly 54 changes. The drive 46 also positions aperture 45 with respect to the detectors 54 and the x-ray source 40. Aperture 45 contains two apertures to confine the x-ray beam within the areas defined by the two scintillator assemblies 55. Rapid camera motion is produced in both directions by servo motors 200 driving high lead angle ball screws 202. The servo motors are controlled by the central computer 72 through driver 204. Preferred embodiment operations employ Reliance Electro-Craft E-3622-H-FOOAN brushless servomotors with 1 ,000 line encoders, Electro-Craft BDC-12 and BDC-25 brushless drives 200. Accurate position sensing, provided by encoders 208, is provided to control the detector positions. Limit switches 210 are used for travel limits and over travel protection.

FIG. 17 demonstrates the sixteen image pane positions with respect to the breast tray 50, 51. The two detectors simultaneously image the breast 49 in the first position and then sequentially move to each of the next seven positions, imaging the breast at each position. The imaging areas of the sixteen image panes, corresponding to the area of the scintillator assembly 55, overlap by 10% to facilitate the stitching algorithm in order to provide a single seamless image of the full breast. FIG. 17b shows the different x-ray paths provided by the dual aperture 45 which strike the two scintillator assemblies 55.

Data acquisition is same as described previously. The sixteen image panes are then stitched as described previously. The detector positions in the second preferred embodiment are accurately repeatable so that the offset and distortion constants are determined only once and not for each breast image.

FIG. 18a and 18b show the front and side view of the packaged invention. The entire x-ray source 40, compression device 50, and detector assemblies 54 are mounted to a C arm 102. The C arm pivots to allow mediolateral and cranial-caudal x-ray views commonly acquired in mammography examinations. In the preferred embodiment, the C arm 102 is attached to the base assembly 106 via a pair of rotary bearings 1 10 which are mounted on a shaft 1 12 attached to a slide mounted vertical travel carriage 108. The vertical travel carriage 108 allows the height adjustment of the C arm 102 for patient interface.

FIG. 20a shows a conventional film screen mammogram of a female breast. It is left MLO projection of the breast. Notice very dark areas in which details of the image are lost. FIG. 20b shows the same view of the same breast taken with an embodiment of the present invention. Note should be taken of enhanced visualization small white dots (images of microcalcifications) are much more apparent in the digital image.of all regions of the breast. Eight panes were stitched for this image. No reconstruction artifacts are apparent. This comparison show that the present invention provides substantially higher quality breast images.

Alternate Embodiments

The attenuator 30 described in FIGS. 12a, 12b and 12c has functioned well in the invention. However, the moderately broad x-ray spectrum from the x-ray source passing through the attenuator 30 results in a phenomenom commonly known as beam hardening which is a variation of x-ray attenuation as a function of x-ray energy. This has resulted in final images which display a slight variation of the luminance in the overlap region compared to the non- overlap region (see combining of images). FIG. 19a illustrates an alternate attenuator 34 which we have fabricated and implemented which is independent of x-ray energy, thereby alleviating the problem of beam hardening. We have tested this attenuator 34 in the first preferred embodiment of the invention with excellent results. The attenuator 34 is positioned at the aperture 6 as shown in FIG. 1. Attenuator 34 is fabricated with 0.5 cm thickness of lead which substantially blocks x-rays. The inner dimension of attenuator 34 is 7.7 cm x 10 cm as displayed in FIG. 19a. The attenuator 34 is initially positioned before each image pane exposure so that the shadow cast by the attenuator 34 completely obscures the overlap region of the image pane. The outer dimensions of the attenuator 34 are large enough to completely block all x-rays in the periphery. The attenuator 34 moves at a constant speed v in the direction illustrated in FIG. 19a during the approximately 1 sec x-ray exposure. This results in an x-ray exposure which decreases linearly as one moves through the overlap region, as shown in FIG. 19b. Attenuator 34 is positioned with respect to each image pane before each image pane exposure by moving the aperture 6 with the linkage system 5. The total exposure in the overlap regions which a sum of exposures of the individual image panes is then reduced to the average exposure level of each pane. Even where the four panes overlap the exposure is reduced to the average exposure level of each image pane.

The present invention requires manual selection of x-ray exposure times for a given x-ray energy and filter combination. The exposure time also depends on the breast size and composition. An alternate embodiment incoφorates an x-ray source 40 and a power supply 41 controlled by an automatic exposure control (AEC) circuit 42 (shown in FIG. 1) utilizing feedback from a pre-exposure pulse. The AEC circuit is presently used on LORAD's

commercially available MS mammography unit. The approximately 1 msec pre-exposure pulse is acquired directly before the main exposure sequence by positioning the one of the detectors 54 at the center of the breast. The computer 72 uses the luminance value of the pre- pulse exposure to decide the correct x-ray exposure for a given x-ray energy and filter combination.

An alternate embodiment to the bolus material described with reference to FIG. 1 lb involves an adjustable x-ray mask 130, located above aperture 6 shown in FIG. 9. The mask 130 is fabricated from a material which substantially attenuates x-rays such as 0.5 cm of lead, for example. The mask 130 is adjusted after the breast 1 is positioned and compressed by the compression mechanism 2 so as to block x-rays at the periphery of the breast 1. The mask 130 is designed to be continuously deformable in order to position around the periphery of the various breast sizes and shapes.

There are several alternate embodiments for improving the quality of the stitching procedure and reducing the stitching time. The correlation analysis can be performed only on the small area surrounding the fiducial marks. The fiducial marks could be put either directly on the breast compression plate, which is subject to deformation, or on a separate dedicated frame. The distortion correction and the registration can be done at the same time to reduce the stitching time.

The preferred embodiment performs the correlation integral C(xι' + dx i', yi' + dy ) in the spatial domain. This is computationally efficient when the positioning uncertainty between panes is limited to a few pixels. The correlation procedure can also be done in the spatial frequency domain by multiplying two-dimensional Fourier transforms of the areas.

The relative position of the panes can also be determined by fiducial mask permanently placed in the x-ray path passing through the breast which superimposes a low level contrast pattern to the breast image. After stitching the image panes, the pattern due to the fiducial mask can be removed from the final image without noticeable image degradation because of the low contrast of the mask. This is done by dividing the breast image by the fiducial mask image resampled to align its grid with the breast image grid, and multiplying by the average luminance of the fiducial mask image.

There are several alternate embodiments with regard to the design features of the second preferred embodiment. The x-ray source 40 and aperture 45 may be provided by a mechanical linkage such as a chain or cable drive. The detectors 54 can be positioned using curved rails with grooved wheels and driven with AC or DC motors with linear position sensors. The detectors 54 can be moved with standard ball screw actuators, belt or cable

dries, or pneumatic cylinder actuators. An alternate embodiment for the C arm 102 consists of a large rotation shaft and bearing set, with the detectors 54 inside the base unit and the optical path passing inside the bearing set.

While the above description contains many specifications, the reader should not construe these as limitations on the scope of the invention, but merely as exemplifications of preferred embodiments thereof. Those skilled in the art will envision many other possible variations are within it scope. CCD camera 16 could be any of many commercially available cameras which could produce either digital images or an analog image. An index matching fluid could be used as the interface between the illumination surface of the Csl crystal and the reflective surface of the reflector plate. For example, CARGILLE Company distributes an index matching fluid that closely matches the index of refraction of Csl the scintillator sandwich can be made as large as available crystal permits. Crystals as large as 24 inches by 24 inches are currently available. Good quality crystals as large as 12 inches by 12 inches are currently available. Dimensions specified herein may be varied if desired. Accordingly, the reader is requested to determine the scope of the invention by the appended claims and their legal equivalents, and not by the examples which have been given.