Classifications

• • G06T3/14—
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T5/00—Image enhancement or restoration
  • G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/10—Segmentation; Edge detection
  • G06T7/174—Segmentation; Edge detection involving the use of two or more images

Definitions

• the present invention relates generally to a computerized method and system provided to aid radiologists in detection of abnormalities, such as lung nodule, pneumothorax, pneumonia, and bulla, in chest radiographs.
• the present invention also generally relates to computerized techniques for automated analysis of digital images, for example, as disclosed in one or more of U.S. Patents 4,839,807; 4,841,555; 4,851,984; 4,875,165; 4,907,156; 4,918,534; 5,072,384; 5,133,020; 5,150,292; 5,224,177; 5,289,374; 5,319,549; 5,343,390; 5,359,513; 5,452,367; 5,463,548; 5,491,627; 5,537,485; 5,598,481; 5,622,171; 5,638,458; 5,657,362; 5,666,434; 5,673,332; 5,668,888; 5,740,268; 5,790,690; and 5,832,103; as well as U.S.
• the present invention includes use of various technologies referenced and described in the above-noted U.S. Patents and Applications, as well as described in the references identified in the appended APPENDIX by the author(s) and year of publication and cross- referenced throughout the specification by numerals in brackets corresponding to the respective references listed in the APPENDIX, the entire contents of which, including the related patents and applications listed above and references listed in the APPENDIX, are incorporated herein by reference.
• Detection of early lung cancers on chest radiographs is a difficult task for radiologists, because subtle lesions tend to be low in contrast and can overlap with ribs and clavicles.
• a temporal subtraction technique has been reported. [1] In this technique, a previous chest image is subtracted from a current chest image to produce a subtraction image. Subtle changes on chest radiographs can be enhanced on the subtraction image, and thus the detection accuracy of interval changes can be improved significantly by use of the temporal subtraction technique. [2] However, the temporal subtraction technique is not applicable in the absence of a previous chest radiograph.
• a novel method, system and computer readable medium for computerized processing of chest images including obtaining a digital first image of a chest; producing a second image which is a mirror image of the first image; performing image warping on one of the first and second images to produce a warped image which is registered to the other of the first and second images; and subtracting the warped image from the other image to generate a subtraction image.
• Another embodiment useful when temporally spaced images are available includes obtaining a digital first image of a chest of a subject; detecting ribcage edges on both sides of the lungs in the first chest image; determining average horizontal locations of the left and right ribcage edges at plural vertical locations; fitting the determined average horizontal locations to a straight line to derive a midline; rotating the chest image so that the midline is vertical; and shifting the rotated image to produce a lateral inclination corrected second image with the midline centered in the lateral inclination corrected image.
• the present invention also includes a computer readable medium storing program instructions by which the method of the invention can be performed when the stored program instructions are appropriately loaded into a computer, and a system for implementing the method of the invention.
• a new contralateral subtraction technique based on one posteroanterior (PA) chest image has been developed. Since the rib structure is nearly symmetrical, the chest image of the right peripheral lung is generally similar to that of the left lung. Taking advantage of this, the technique includes lateral inclination correction by rotating and shifting the original chest image so that the midline of the thorax is aligned with the vertical centerline of the original chest image, lateral reversal of the rotated image to produce a reversed "mirror" image, warping of the mirror image, and subtraction of the warped mirror image from the original image to derive the contralateral subtraction image. Thereafter, additional processing techniques can be successively applied to the initial contralateral subtraction image to acquire improved subtraction images.
• PA posteroanterior
• a contralateral subtraction image can be obtained by subtracting the right/left reversed "mirror" image from a single chest image. Similar to the temporal subtraction technique, the contralateral subtraction scheme can cancel out most of symmetrical skeletal structures and enhance asymmetrical opacities, thus demonstrating subtle abnormalities more clearly. On the other hand, unlike the temporal subtraction technique, a subtraction image can be obtained whenever a single PA chest image is available. Therefore, the contralateral subtraction technique can be of significant clinical importance in some cases.
• Figure 1(a) is an overall scheme of a contralateral subtraction technique for a PA chest image
• Figure 1(b) is an overall scheme for improved temporal subtraction using a midline detection method of a contralateral subtraction technique
• Figure 1(c) is an overall scheme for the removal of false positives from difference images using a contralateral subtraction technique in computerized detection of lung nodules in single chest radiographs;
• Figures 2(a), 2(b), 2(c) and 2(d) are images showing for comparison (a) an original chest radiograph, (b) a contralateral subtraction image including all regions, (c) a contralateral subtraction image with a uniform background over an area outside the two lungs, and (d) a contralateral subtraction image with a "chest" background, which is the original image superimposed over the area outside the two lungs, respectively;
• Figure 3 is an illustration of measures for errors of angle and location of a detected midline relative to a gold standard;
• Figure 4 is a chest radiograph with a gold standard (black solid line) for midline, detected midlines given by a profile based method (dashed line) and by a ribcage edge based method (dotted line);
• Figures 5(a), 5(b) and 5(c) are charts illustrating (a) distribution of errors for angles and locations from a gold standard for individual midlines indicated by three observers, (b) distribution of errors for a midline detected by a profile based method, (c) distribution of errors for a midline detected by a ribcage edge based method, respectively;
• Figures 6(a), 6(b) and 6(c) are images showing for comparison (a) an original chest radiograph, (b) a contralateral subtraction image obtained by using a profile based method for midline detection, and (c) an improved subtraction image obtained by using a ribcage edge based method for midline detection, respectively;
• Figures 7(a), 7(b) and 7(c) are images showing for comparison (a) an original chest radiograph, (b) a contralateral subtraction image obtained by including mediastinal and cardiac regions for shift value analysis, and (c) an improved subtraction image obtained by eliminating the mediastinal and cardiac regions from the shift value analysis, respectively;
• Figures 8(a), 8(b) and 8(c) are images showing for comparison (a) an original chest radiograph, (b) a contralateral subtraction image obtained by using simultaneous fittings of shift values in right and left lungs, and (c) an improved subtraction image obtained by using separate fittings of shift values in the right and left lungs, respectively;
• Figures 9(a) and 9(b) are images showing for comparison (a) an indicated lung nodule, and (b) a contralateral subtraction image with an enhanced nodule, respectively;
• Figure 10 is a graph showing a distribution of contrasts of nodule candidates at 10 mm and 20 mm diameters for nodules and non-nodules;
• Figure 11 is a graph showing a distribution of relative standard deviations for nodule candidates in object and background regions at 10 mm diameter for nodules and non-nodules;
• Figure 12 is a graph showing a relationship between contrast and relative standard deviation at 10 mm diameter for nodules and non-nodules.
• Figure 13 is a schematic illustration of a general purpose computer 100 programmed according to the teachings of the present invention.
• the chest images used in development of the present invention consist of 50 normals and 50 abnormals with solitary lung nodules which were selected from 247 chest images in the Japanese Standard Digital Image Database developed by the Japanese Society of Radiological Technology.
• the images were digitized with a 0.175 mm pixel size, a matrix size of 2048 x 2048 and 12 bits gray levels. However, the matrix size was reduced to 512 x 512 by subsampling of the original image data, and the number of gray levels was decreased to 10 bits.
• Fig. 1(a) there is illustrated a top-level block diagram of a contralateral subtraction technique according to the present invention.
• the lateral inclination which may be caused by an improper patient positioning is corrected by an image rotation technique at step S200.
• the rotated image is laterally (right/left) reversed to produce a reversed "mirror" image at step S300.
• the mirror image is registered to the original image by use of a cross correlation technique, and then deformed based on a nonlinear image warping technique in order to match peripheral ribs in the mirror image with those in the original image at step S400.
• the warped mirror image is subtracted from the original image at step S500 to derive the contralateral subtraction image at step S600.
• Lateral inclination correction (Step S200) is designed particularly to correct lateral inclination of the thorax in the chest image. If the midline of the thorax is inclined slightly with an angle from the vertical direction, then the difference in the angles of the midlines between the original and the reversed chest images will become twice as large, which may lead to a serious misregistration error, and thus result in a poor subtraction image. It is therefore necessary to correct the lateral inclination before the subtraction technique is applied, by rotating the image so that the midline is in a vertical direction and then shifting the midline of the thorax to the vertical centerline of the original chest image, performed step S200.
• the rotating and shifting of the image can be, for example, achieved using conventional image rotation and shifting techniques as is known in the imaging arts.
• CAD computer-aided diagnosis
• the pixels along the midline have local maximum values along a straight line, then this method can give satisfactory results.
• the pixels on the midline are not necessarily local maximum values.
• the pixels in the cardiac region are often greater than those near the midline, and thus the detected midline is commonly shifted to the cardiac region, resulting in an angulated incorrect midline, as will be illustrated later.
• the pixels on the midline have local maximum values, the difference between these values and pixel values in the adjacent area is usually very small, which makes the prior method intrinsically vulnerable to image noise. Therefore, according to the present invention, a new ribcage edge based midline detection method has been devised, next described.
• the ribcage edges on both sides of the lungs are detected.
• the average horizontal locations of the left and right ribcage edges are determined and then fitted to a straight line to derive the midline.
• the ribcage edges are detected by analyzing the first and second derivatives of profiles through chest images.
• the detected ribcage edges are then fitted to a third order polynomial function to form smooth curves and also reduce noise, as described in detail elsewhere.
• the ribcage edge detection method and the previous midline detection method described above both detect edges or peaks by analyzing profiles and fitting them to some functions.
• the results of ribcage edge detection are far more reliable than the results of the previous midline detection because the contrast on the ribcage edges is usually much greater. This is why the present invention preferably employs the ribcage edges for determination of the midline.
• the average horizontal locations (i.e., mid-points) of the right and left ribcage edges at the same vertical position of the chest image are then determined. Due to the nearly symmetrical property of the ribcage edges on both sides of the lungs, most of the mid-points (or average locations) lie on a straight line, and thus these points are fitted to a straight line to produce the expected midline. In other words, a straight line to divide the lungs into two approximately symmetrical and equal parts is determined.
• the mid-points are derived from the average horizontal locations of the right and left ribcage edges, the variance of errors in these mid-points is expected to be a half of that in the detected ribcage edges, which would also contribute to a higher accuracy of the detected midline.
• a right/left reversed mirror image is obtained by laterally reversing the original image after the lateral inclination is corrected at step S200.
• This reversed mirror image will be warped and then subtracted from the original image to produce the contralateral subtraction image.
• a nonlinear image warping technique has been used successfully in the temporal subtraction technique applied to chest images as described in detail previously (see, e.g., Patent application serial Number 09/053,789). According to the present invention, a similar image warping technique to obtain image registration is employed. However, proper modifications are made in image warping as applied to contralateral subtraction as described below.
• the nonlinear image warping technique includes initial global matching, detailed local matching of peripheral ribs, determination of shift values, and coordinate conversion. In addition, nonlinear image warping can be performed iteratively to yield improved results. [5]
• the initial global matching technique which has been already applied to the temporal subtraction technique, is used for aligning the approximate lung area in the reversed mirror image to that in the original image.
• the matrix sizes of two images i.e., the original chest image and the reversed mirror image are reduced by a factor of four, and smoothed by a Gaussian filter to reduce the effect of the fine anatomical structures, such as small vessels, bronchi, devices and catheters on global matching of the two images.
• the lungs are then extracted from each of the two images by using ribcage edge information described above, and the region outside the ribcage edges is ignored. Finally, the upper parts of the lungs are aligned by using a cross correlation technique.
• template ROIs and search area ROIs are automatically located within the lung regions of the original image and the reversed image, respectively, in the same way as those ROIs were located by the temporal subtraction technique.
• the matrix sizes of template and search area ROIs are 32 x 32 and 64 x 64, respectively. It is important to note that with the contralateral subtraction technique, the mediastinal and cardiac regions are excluded for the selection of the ROIs, because the ROIs in the mediastinal and cardiac regions do not contain useful information for reliable image matching for the contralateral subtraction.
• a cross correlation technique is employed to perform the local image matching for the determination of the shift values ⁇ X and ⁇ y in two orthogonal directions, which indicate a shift of the coordinates of the center of a search area ROI in the reversed mirror image, for the best match of the template ROI with the corresponding "matched" area in the search area ROI.
• a contralateral subtraction image is obtained by subtracting the warped reversed image from the original image at step S500.
• the pixel values of the subtraction image may be multiplied by a factor such as 1.5 and 2.0.
• a factor of 1.0 can be used also for maintaining the same contrast of abnormal opacities in the contralateral subtraction image as that in the original image.
• a constant pixel value of 512 is then added to the contrast enhanced subtraction image, and the background region outside lungs is assigned a constant pixel value of 512 in order to display only the contralateral subtraction image of peripheral lungs.
• Fig. 2(a) shows an original chest image
• Figs. 2(b) and 2(c) show, for comparison, contralateral subtraction images without and with the uniform background, respectively.
• the subtraction image with the uniform background can present effectively the asymmetric opacities in lung regions by dark patterns in the ipsilateral side and light patterns in the contralateral side.
• the original PA chest image outside lung areas instead of the constant background, can be displayed to maintain the general appearance of a "chest" background, as shown in Fig. 2(d).
• many radiologists preferred the display of the Fig. 2(d) contralateral subtraction image.
• a five- point rating score which has been used previously for subjective evaluation of the temporal subtraction image, is employed according to the present invention, that is,
• the rating score ranges from -2 to +2 as follows, when the quality of the subtraction image is
• the accuracy of the detected midline is evaluated by determining the errors of angle and location of the midline compared with the gold standard.
• Fig. 3 shows the definitions for the errors of the angle and the location of the detected midline.
• the error of angle between the detected midline and the gold standard is defined by the angle between the orientations of the detected midline and the gold standard.
• the error of location is defined by the horizontal distance of the detected midline from the gold standard at the vertical position of the midpoint between the two ends of the gold standard.
• Fig. 4 shows a chest radiograph with the gold standard (solid line) and midlines detected by the profile based method (dashed line) and the ribcage edge based method (dotted line).
• the detected ribcage edges are also shown by the solid curves.
• the midline by the ribcage edge based method is very similar to the gold standard, whereas the midline by the profile based method is angulated and quite different from the gold standard probably due to the effect of the cardiac region, as described earlier.
• Fig. 5(a) shows the distribution of the errors of angles and locations from the gold standard for individual midlines indicated by three observers. It is apparent that most points are distributed around the origin, which indicates that the errors are relatively small and the midline data given by three observers are quite consistent.
• Figures 5(b) and 5(c) show the distribution of the errors for the detected midlines given by the profile based and the ribcage edge based methods, respectively. It is clear that the ribcage edge based method can provide more accurate results than the profile based method.
• Table 1 shows the comparison of the standard deviations for errors of angles and locations for the midlines measured by observers and by the two midline detection methods. Note that the standard deviations obtained by the ribcage edge based method are considerably smaller than those by the profile based method, and comparable to those by observers.
• Table 2 demonstrates the distribution of the number of cases for different subjective rating scores on the quality of subtraction images, which are obtained with an initial scheme and also by incorporating successively the three techniques.
• Table 3 shows the distribution of the number of cases for the change in subjective rating scores on the quality of subtraction images due to the successive use of the three techniques. At the initial phase of this study, 100 cases of chest radiographs with the initial contralateral subtraction scheme without the three techniques were examined.
• the first technique is the use of an improved midline detection method based on the ribcage edges, instead of the profile based method, to correct for the lateral inclination of the thorax in chest radiographs.
• this technique can improve the contralateral subtraction images considerably, especially for those cases with initially lower subjective rating scores. This can also be observed from the second column (b) of Table 2, where the numbers of cases with rating scores of 1 and 2 are decreased considerably after the new midline detection method is used.
• FIG. 6 shows comparison of (a) an original chest radiograph, (b) a contralateral subtraction image obtained by using a profile based method for midline detection, and (c) an improved subtraction image obtained by using a ribcage edge based method for midline detection. It is apparent in Fig. 6 that the subtraction image is clearly improved by using the new midline detection method.
• the second technique is the elimination of the mediastinal and cardiac region for the shift value analysis.
• This technique is equivalent to the application of the contralateral subtraction technique to lung areas alone.
• the improvement in the contralateral subtraction images is indicated in the second column (b) in Table 3 , and the rating scores for the improved quality of subtraction images are listed in the third column (c) in Table 2.
• Fig. 7 shows comparison of (a) an original chest radiograph, (b) a contralateral subtraction image obtained by including mediastinal and cardiac regions for shift value analysis, and (c) an improved subtraction image obtained by eliminating the mediastinal and cardiac regions from the shift value analysis. It is apparent in Fig. 7 that the subtraction images are further improved by adding the second technique.
• the third technique is the separate fittings of shift values on the right and left lungs independently.
• Tables 2 and 3 show the improvement made by this technique and the final rating scores for the quality of subtraction image, respectively.
• Figure 8 shows comparison of (a) an original chest radiograph, (b) a contralateral subtraction image obtained by using simultaneous fittings of shift values in right and left lungs, and (c) an improved subtraction image obtained by using separate fittings of shift values in the right and left lungs.
• the contralateral subtraction technique can be used to detect asymmetric abnormalities, for example, lung nodule, pneumothorax, pneumonia, and bullae, on single chest radiograph.
• Fig. 9(a) shows a chest radiograph with a lung nodule indicated by an arrow, and its contralateral subtraction image in Fig. 9(b). It is apparent that the lung nodule is enhanced in the subtraction image, since most of symmetric skeletal structures are canceled out.
• the midline of the thorax can be represented approximately by a straight line. It is believed that this assumption is valid in most of PA chest images. However, due to large variations of individual radiographs, this assumption is not always valid to describe midlines in all clinical chest radiographs. In some cases, a nonlinear model to represent the midlines adequately may be needed. For example, a polynomial with the order of 2 or 3 may be suitable for representing the curved midline. Nevertheless, according to the present invention the linear model was used, because the midline is used only for correcting the lateral inclination of the thorax in a chest radiograph. Note that the linear model is simple and adequate for this invention. A complex nonlinear model may complicate the problem and lead to difficulty in determining a proper correction of the midline.
• the subtraction images have been displayed with the small matrix size of 512 x 512, because the cross-correlation technique and the subsequent image warping technique have been applied to this small matrix size.
• a technique to display a high quality, large matrix subtraction image, and equivalently a high quality subtraction image with a small pixel size was devised. This is accomplished by employing a large matrix image such as 1024 x 1024 or 2048 x 2048 for image warping of the reversed mirror image for the contralateral subtraction technique and the previous image in the case of the temporal subtraction technique, as will be further discussed.
• the corresponding shift values for the large matrix size are obtained by scaling using a factor of two or four for the matrix size of 1024 x 1024 or 2048 x 2048, respectively, of the reversed mirror image (or the previous image in the case of temporal subtraction technique).
• the large matrix subtraction image is obtained by subtracting the large matrix warped image from the large matrix original chest image.
• This technique for display of the larger matrix size and thus the use of a small pixel size was very useful for producing high quality subtraction images by elimination of pixel edges and pixel artifacts due to the use of a small matrix size (or a large pixel size) for the temporal and the contralateral subtraction techniques. It is useful also to apply an image processing technique such as a median filter and a Gaussian filter for smoothing the subtraction images by elimination of image noise and relatively small misregistration artifacts.
• the above-described three techniques to improve contralateral subtraction images according to the present invention are applicable to the temporal subtraction technique [1,2] for improvement of the quality of temporal subtraction images.
• the basic scheme of the temporal subtraction technique is briefly described here.
• a number of template ROIs (32 x 32 matrix) and the corresponding search area ROIs (64 x 64 matrix) are selected from the previous and the current images, respectively, to perform local matching of image details in the two images.
• the shift values, ⁇ X and ⁇ y, for all pairs of selected ROIs are then determined by using a cross-correlation technique to find the best matched areas in the search area ROI.
• a two-dimensional surface fitting using a polynomial function is applied to each set of mapped shift values, ⁇ X and ⁇ y, for conversion of the x, y coordinates of the previous image, i.e., for warping of the image.
• the warped previous image is then subtracted from the current image to produce the temporal subtraction image.
• the temporal subtraction image has been shown to be very useful to detect subtle interval changes in chest images.
• a second image warping technique can be employed by using the current and the first warped previous image.
• the final temporal subtraction image is obtained by subtraction of the second warped previous image from the current chest image.
• the image warping technique can be applied repeatedly and thus iteratively; this technique is therefore called an iterative image warping technique.
• Fig. 1(b) is an overall scheme for improved temporal subtraction using the previously described midline detection method of the contralateral subtraction technique (e.g., Fig. 1(a) step S200).
• Fig. 1(b) at step S100' current and previous chest images are obtained.
• step S200' lateral inclination correction is performed on the obtained images using the midline detection method of the contralateral subtraction technique to reduce effects of misregistration.
• step S500' temporal subtraction is performed and at step S600' the temporally subtracted image is displayed.
• Table 4 Distribution of the numbers of chest images for different subjective rating scores on the quality of temporal subtraction images, which are obtained by using profile based and ribcage edge based midline detection method to correct lateral inclination
• Table 5 Distribution of the numbers of chest images for different subject rating scores on the improvement of quality of temporal subtraction images by replacing profile based mid: ine detection method with ribca e ed e based midline detection method
• 181 pairs of current and previous chest images which were obtained from the lung cancer screening program at Iwate prefecture in Japan were employed. These chest images are obtained by using a Fuji computed radiography (CR) system which was installed in a mobile screen unit. Digital image data from the CR system are directly used for calculating the temporal subtraction images using the image warping technique. It is apparent in Tables 4 and 5 that the quality of temporal subtraction images is considerably improved by using the new midline detection technique.
• CR computed radiography
• the technique to display a high quality, large matrix subtraction image as described above, can be applied to temporal subtraction images.
• the contralateral subtraction technique can be incorporated into a computer-aided diagnosis scheme to improve the overall performance. For example, it is very difficult to eliminate some false positives caused by rib-rib crossings or vessel-rib crossings in the computerized detection of lung nodules in chest radiographs using a difference imaging scheme. These false positives can be eliminated by analyzing image features derived from the contralateral subtraction image.
• This difference image technique is considered a generalization of edge enhancement techniques, and in fact the difference image looks very similar to edge-enhanced images. If the two filters are linear filters, then the two can be combined into one filter operation. However, if a nonlinear filter is applied, two filtering operations need to be applied separately in parallel.
• the difference image technique is applicable to computerized detection of isolated abnormal patterns such as microcalcifications and masses in mammograms.
• the initial candidates of lung nodules are identified from relatively round patterns with large pixel values in the difference image, which may include nodules as well as normal structures such as ribs and pulmonary vessels. Therefore, image features of these candidates are extracted and quantified in terms of the size, contrast, and other parameters related to the shape of the candidates. Specifically, the extracted features are related to gray level, morphology, or edge gradient, such as effective diameter, degrees of circularity and irregularity, slopes of the effective diameter and degrees of circularity and irregularity, average gradient, standard deviation of gradient orientation, contrast and net contrast (e.g., as taught in Patent Application No. 08/562,087).
• a rule-based method e.g., as taught in Patent Nos. 5,463,548 and 5,622,171 and Patent Application Nos.
• 08/562,087; 08/562,188; 08/758,438; 08/900,361; and 09/027,685) is then applied which removes some candidates as false positives, when their features are matched to those of normal anatomic structures such as end-on vessels, rib-rib crossings, rib-vessel crossings, aggregates of vessels, and rib-clavicle crossings.
• ANN trained artificial neural network
• Figure 1(c) is an overall scheme for the removal of false positives from difference images using a contralateral subtraction technique in computerized detection of lung nodules in single chest radiographs.
• the contralateral subtraction images were obtained for all of ten chest images (steps S200-S600). All of the 24 locations on original chest images, which included 10 nodules and 14 false positives, indicating the computer output of the nodule detection scheme were then converted to the corresponding locations on the contralateral subtraction images (step S900). This conversion was necessary because the contralateral subtraction images were obtained after the original chest images were rotated and shifted for correction of the lateral inclination (steps S200-S600), as described earlier.
• ribs in one lung match with those in another lung, ribs are generally eliminated and thus have uniform density which is the same as the adjacent background, and nodules appear as dark round patterns. Therefore, false positive computer output due to ribs can be identified, if image features associated with a computer detected candidate would match those of ribs.
• the contrast of the candidate as an image feature to distinguish between nodules and false positives due to ribs was employed (step SI 000). The contrast is defined here as the difference in the average pixel values between the central area of the candidate and the immediately adjacent background area.
• the average pixel value in the central area of the candidate was empirically determined over a circle with a diameter d, whereas the average pixel value in the background area was determined over a doughnut shape area with the inner diameter of d and the outer diameter of 2d.
• the contrasts of all candidates including nodules and non-nodules (false positives) were determined, for example, over a wide range of the diameter d from 2 mm to 40 mm. It was found that the contrasts of nodules in contralateral subtraction images tend to be generally greater than those of non-nodules. For example, Fig.
• Fig. 11 shows the relative standard deviations in the central and the background area. The relative standard deviation was determined by the ratio of the standard deviation of pixel value variation to the average pixel value in the central area and/or the background area. It is apparent in Fig.
• Fig. 12 Another method for elimination of false positives is illustrated in Fig. 12, where the correlation between the contrast and the relative standard deviation of the central areas of the nodules is demonstrated. It is apparent in Fig. 12 that nodules tend to have large contrasts and small standard deviations, whereas non-nodules tend to have small contrasts or large contrasts with large relative standard deviations. Therefore, when a contrast versus standard deviation threshold as indicated by the dotted line is employed, 13 out of 14 false positives can be eliminated (step S1200-S1300). This result indicates that a large number of false positives can be eliminated and thus can improve substantially the performance of the computerized scheme for detection of lung nodules in chest images by employing contralateral subtraction images.
• FIG. 13 is a schematic illustration of a computer system for detection of asymmetric abnormalities in a single chest radiograph.
• a computer 100 implements the method of the present invention, wherein the computer includes, for example, a display device 102, such as a touch screen monitor with a touch-screen interface, a keyboard 104, a pointing device 106, a mouse pad or digitizing pad 108, a hard disk 110, or other fixed, high density media drives, connected using an appropriate device bus (e.g., a SCSI bus, an Enhanced IDE bus, an Ultra DMA bus, a PCI bus, etc.), a floppy drive 112, a tape or CD ROM drive 114 with tape or CD media 116, or other removable media devices, such as magneto-optical media, etc., and a mother board 118.
• a display device 102 such as a touch screen monitor with a touch-screen interface
• a keyboard 104 such as a touch screen monitor with a touch-screen interface
• the mother board 118 includes, for example, a processor 120, a RAM 122, and a ROM 124 (e.g., DRAM, ROM, EPROM, EEPROM, SRAM, SDRAM, and Flash RAM, etc.), I/O ports 126 which may be used to couple to an image acquisition device (not shown), and optional special purpose logic devices (e.g., ASICs) or configurable logic devices (e.g., GAL and reprogrammable FPGA) 128 for performing specialized hardware/software functions, such as sound processing, image processing, signal processing, neural network processing, etc., a microphone 130, and a speaker or speakers 132.
• a processor 120 for example, a processor 120, a RAM 122, and a ROM 124 (e.g., DRAM, ROM, EPROM, EEPROM, SRAM, SDRAM, and Flash RAM, etc.), I/O ports 126 which may be used to couple to an image acquisition device (not shown), and optional special purpose logic devices (e.g., A
• the system includes at least one computer readable medium.
• Examples of computer readable media are compact discs, hard disks, floppy disks, tape, magneto-optical disks, PROMs (EPROM, EEPROM, Flash EPROM), DRAM, SRAM, SDRAM, etc.
• the present invention includes software for controlling both the hardware of the computer 100 and for enabling the computer 100 to interact with a human user.
• Such software may include, but is not limited to, device drivers, operating systems and user applications, such as development tools.
• Such computer readable media further includes the computer program product of the present invention for performing any of the processes according to the present invention, described above (see, e.g., Figs. l(a)-(c)).
• the computer code devices of the present invention can be any interpreted or executable code mechanism, including but not limited to scripts, interpreters, dynamic link libraries, Java classes, and complete executable programs, etc.
• the programming of general purpose computer 100 may include a software module for digitizing and storing images obtained from an image acquisition device (not shown).
• the present invention can also be implemented to process digital data derived from images obtained by other means, such as a picture archive communication system (PACS).
• PACS picture archive communication system
• the invention may also be implemented by the preparation of application specific integrated circuits or by interconnecting an appropriate network of conventional component circuits, as will be readily apparent to those skilled in the art.
• the present invention is defined in terms of eliminating false positives in candidate nodules, based on contrast, relative standard deviation, and/or correlation between the contrast and the relative standard deviation of the central and/or background areas of the nodules, other image features and/or other areas of the nodules can be used, as will be readily apparent to those skilled in the art.
• a novel contralateral subtraction technique for the detection of asymmetric opacities on single chest radiograph has been developed, and the three techniques (i.e., lateral inclination correction, exclusion of cardiac region, and separate fitting of shift values for left and right lungs) to improve the contralateral subtraction image have been applied.
• the contralateral subtraction techmque can remove most of peripheral ribs, and thus enhance low-contrast peripheral lesions on chest radiographs. It is believed that this technique can assist radiologists in the detection of subtle lung opacities.

Applications Claiming Priority (3)

Publications (2)

Family

ID=22314805

Family Applications (1)

Country Status (4)

Families Citing this family (3)

Citations (2)

Family Cites Families (9)

• 1999
  • 1999-11-05 AU AU17060/00A patent/AU1706000A/en not_active Abandoned
  • 1999-11-05 WO PCT/US1999/024007 patent/WO2000028466A1/en active Application Filing
  • 1999-11-05 EP EP99960126A patent/EP1127330A4/en not_active Withdrawn
  • 1999-11-05 JP JP2000581581A patent/JP4638042B2/ja not_active Expired - Lifetime

Patent Citations (2)

Non-Patent Citations (4)

Also Published As

Similar Documents

Legal Events