JP2005074080A - Method and apparatus for detecting abnormal shadow candidate, and program therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and an apparatus for correctly detecting an abnormal shadow candidate in a simple way without anatomical inconsistency by using two or more images which indicate a same subject and are different in direction for viewing the subject, and to provide a program therefor. <P>SOLUTION: A detection control means 50 compares the numbers of fine calcified abnormal shadow candidates N<SB>ML</SB>and N<SB>CC</SB>which are detected by a fine calcified abnormal shadow candidate detecting means 10 from a side image P<SB>ML</SB>and a front image P<SB>CC</SB>of the same subject. Till the difference ΔN becomes not more than a prescribed reference value ΔN<SB>max</SB>, a threshold changing means 30 repeatedly changes the detection threshold values of the fine calcified abnormal shadow candidates with respect to at least one of the images P<SB>ML</SB>and P<SB>CC</SB>, so as to allow the difference ΔN of the detection result to come close to the prescribed reference value ΔN<SB>max</SB>. The fine calcified abnormal shadow candidate detecting means 10 also repeatedly detects the fine calcified abnormal shadow candidates, based on the changed detection threshold. Thus, each detection result when the detection result difference ΔN becomes not more than the prescribed reference value ΔN<SB>max</SB>is outputted as a final result. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a method, an apparatus, and a program for detecting an abnormal shadow candidate in an image representing a subject, and more specifically, an abnormality using two or more images representing the same subject and having different viewing directions. The present invention relates to a method, an apparatus, and a program for detecting a shadow candidate.

  In the medical field, a computer-aided diagnosis system (CAD: Computer Aided Diagnosis) that automatically detects abnormal shadow candidates in an image and highlights the detected abnormal shadow candidates is known.

  As a method for detecting an abnormal shadow candidate, for example, image processing using an iris filter is performed on a radiation image (mammography) of a breast, and an output value thereof is subjected to threshold processing, whereby a tumor shadow that is one form of breast cancer or the like ( One form of abnormal shadow) and image processing using a morphological filter are performed automatically, and the output value is thresholded to form a microcalcification shadow that is another form such as breast cancer. A method of automatically detecting a candidate (one form of abnormal shadow) is known (for example, Patent Document 1).

  These abnormal shadow candidate detection methods have significantly higher detection accuracy than the conventional methods, but currently cannot detect only abnormal shadows with 100% accuracy. That is, radiation noise that occurs randomly or other noise during reading cannot be correctly determined as noise and cannot be completely eliminated, and false detection of these as abnormal shadow candidates is completely zero. Has not been realized. In addition, regarding normal tissues and normal structures such as mammary glands, erroneous detection as abnormal shadow candidates cannot be completely reduced to zero. On the other hand, abnormal shadows may not be correctly determined and may be missed. For this reason, further improvement in diagnostic performance (detection accuracy) is strongly desired in the medical field.

Therefore, an abnormal shadow candidate is detected from each of a plurality of images having different shooting directions, the identity of the position of the abnormal shadow candidate detected from each image is determined, and the abnormal shadow candidate detected at substantially the same position is further determined. A method of detecting as a definite candidate has been proposed (for example, Patent Documents 2 and 3). This method is based on the fact that the detection results of abnormal shadow candidates should be the same in principle even if the shooting direction of the detection target image is different for the same subject. By comprehensively judging the detection results, the diagnostic performance (detection accuracy) is to be improved.
JP-A-8-294479 Japanese Patent Laid-Open No. 10-108859 JP 2001-346786 A

  However, in practice, when the imaging directions are different, the overlapping state of anatomical structures such as mammary glands on the image is different. In addition, since a plurality of images having different shooting directions are shot separately, shooting conditions such as a tube voltage, a tube current, and an irradiation time of the radiation source vary between the images, and the S / N ratio on the image Contrast may be different between images. On the other hand, if only the abnormal shadow candidate detection method described in Patent Document 1 is used, detection is performed on any image at a constant detection level. As a result, the detection result of the abnormal shadow candidate may greatly differ depending on the shooting direction, for example, erroneous detection occurs only in the image in one shooting direction. On the other hand, if it is possible to strictly determine the identity of the three-dimensional position of the abnormal shadow candidate by the methods of the inventions described in Patent Documents 2 and 3, it is assumed that a false detection has occurred in the processing of one image. However, since it is unlikely that the erroneously detected candidate and the candidate detected in the processing of the other image are at the same three-dimensional position, this can be an effective technique. However, in reality, it is very difficult to strictly determine the identity of the three-dimensional position of abnormal shadow candidates detected from a plurality of images having different shooting directions. In the inventions described in Patent Documents 2 and 3, However, only a rough determination is made that the identity of the position is determined based on the distance from the end of the subject such as the nipple. In such a rough determination, when an erroneous detection occurs in the processing of one image, the erroneously detected candidate has the same three-dimensional position as the candidate detected in the processing of the other image. May be erroneously determined as such, and a decrease in detection accuracy becomes a problem.

  The present invention has been made in view of the above circumstances, and by using two or more images representing the same subject and in different directions of viewing the subject, abnormal shadow candidates can be more accurately and more accurately obtained. An object of the present invention is to provide a method, an apparatus, and a program for detecting an anatomical contradiction.

  The abnormal shadow candidate detection method according to the present invention detects an abnormal shadow candidate from each of two or more images that represent the same subject and have different viewing directions, and detects an abnormal shadow candidate for each of the two or more images. Compare the results, and change the detection level of the abnormal shadow candidate for at least one of these two or more images so that the difference in the detection results approaches the predetermined standard until the difference in the detection results satisfies the predetermined standard And detecting an abnormal shadow candidate based on the changed detection level, and each of the detection results when the difference in detection results satisfies the predetermined criterion is set as a final result. And

  An abnormal shadow candidate detection apparatus according to the present invention is an apparatus for realizing this method, and detects an abnormal shadow candidate from each of two or more images representing the same subject and having different viewing directions. The detection results of the candidate detection means, the detection level changing means for changing the detection level of the abnormal shadow candidate, and the abnormal shadow candidate detection means for each of these two or more images are compared, and the difference between the detection results satisfies a predetermined reference. Based on the changed detection level, the detection level changing means changes the detection level of the abnormal shadow candidate for at least one of the two or more images so that the difference between the detection results approaches the predetermined reference until satisfying the condition. The abnormal shadow candidate detection means to detect abnormal shadow candidates, and each of the detection results when the difference in detection results satisfies the predetermined criterion is maximized. Characterized in that a detection control means for outputting as a result.

  Furthermore, the abnormal shadow candidate detection program according to the present invention is a program for causing a computer to perform this method. The abnormal shadow candidate is detected from each of two or more images representing the same subject and having different viewing directions. Detection of abnormal shadow candidates for each of these two or more images, and detection of abnormal shadow candidates for at least one of these two or more images until the difference between the detection results satisfies a predetermined criterion. The level is changed so that the difference between the detection results approaches the predetermined reference, and the abnormal shadow candidate is detected based on the changed detection level, and the difference between the detection results exceeds the predetermined reference. Each of the detection results when satisfied is output as a final result.

  Next, details of the abnormal shadow candidate detection method, apparatus, and program according to the present invention will be described.

  As an example of “two or more images representing the same subject and having different viewing directions”, two or more images obtained by photographing the same subject from different directions can be considered. Specifically, a side image (Medio-lateral view or Medio-lateral oblique view) obtained by sandwiching the breast in the lateral direction in breast cancer screening and photographing from the lateral direction, and vertically sandwiching the breast vertically Two images of a front image (Cranio-caudal view) obtained by photographing from the direction, a front image of the chest (this front image means an image obtained from the front-rear direction of the body) and a side image One image etc. are mentioned. Further, it may be two or more projected images obtained by projecting pixels in a region of interest in a three-dimensional image representing a subject onto different projection planes. Here, the three-dimensional image refers to an image based on image data in which the position of each pixel is defined in a three-dimensional space. Specific examples include volume data, CT (Computerized Tomography), and MRI (Computerized Tomography). An image based on image data taken and acquired by Magnetic Resonance Imaging) is conceivable. As a specific example of the process of projecting the pixels in the attention area onto the projection plane, a process for obtaining the average value of the pixel values of the pixels in the attention area in the direction perpendicular to the projection plane can be considered.

  As a specific example of the process of “detecting an abnormal shadow candidate”, an image process using an iris filter is performed, and a threshold value process is performed on the output value of the tumor shadow (a form of abnormal shadow) that is a form of breast cancer. By performing processing that automatically detects candidate regions and image processing using a morphological filter and thresholding the output value, microcalcification shadows that are other forms such as breast cancer (a form of abnormal shadows) For example, a process for automatically detecting candidate areas).

  As a specific example of the “abnormal shadow candidate detection result”, the number of detected abnormal shadow candidates can be considered. In this case, the “predetermined criterion” is that the difference in the number of abnormal shadow candidates for each of these two or more images is not more than a predetermined value, and the abnormal shadow candidates detected from each of these two or more images Until the difference in the number of abnormal shadow candidates becomes equal to or smaller than the predetermined value, the detection level of the abnormal shadow candidate for at least one of these two or more images is reduced. Each of the detection results when the difference in the number of abnormal shadow candidates is less than or equal to the predetermined value is repeated. It can be considered as a result.

  As a specific example of the “detection level”, a detection threshold and a detection rank in the abnormal shadow candidate detection process can be considered. The detection rank refers to the number of candidates detected by the abnormal shadow candidate detection process described above, and the number of abnormalities that are set as detection ranks in descending order of the output value (the possibility of abnormal shadows is high). Detect as a shadow candidate.

  The detection level may be changed so that the difference between detection results approaches a predetermined reference, and each detection result approaches a predetermined second reference. Here, as a specific example of the “predetermined second reference”, the number of detected abnormal shadow candidates is found empirically as the actual number of abnormal shadows that can be detected from an image representing the diagnostic region. A criterion of being within the range of values determined is conceivable. For example, when four abnormal shadow candidates are detected from the first image and one abnormal shadow candidate is detected from the second image, the difference in the number of abnormal shadow candidates between the images is 1 or less. If the detection threshold is changed based on the first criterion that there is a second criterion that the number of abnormal shadow candidates in each image is two, the abnormal shadow candidate detected from both images Since the difference in the number of the first and second images is larger than 2, the detection threshold is changed so that the difference is reduced for at least one of the first and second images according to the first criterion, and the abnormal shadow candidate is detected again. In addition, since the number of abnormal shadow candidates detected from the second image is larger than 2, the second criterion is such that the number of abnormal shadow candidates detected for the second image is reduced, that is, Number of abnormal shadow candidates for the second criterion ( To approach 2), it will detect again the abnormal shadow candidate by increasing the detection threshold. Further, for the second image, the detection threshold is lowered again so as to increase the number of detected abnormal shadow candidates, that is, to approach the number of the second reference abnormal shadow candidates (= 2). An abnormal shadow candidate may be detected.

  When the detection level is changed so that the number of detected abnormal shadow candidates decreases, and abnormal shadow candidates are detected based on the changed detection level, instead of doing this, the detection level is already changed based on the detection level before the change. Detect abnormal shadow candidates based on a new detection level by selecting a predetermined number of abnormal shadow candidates from the detected abnormal shadow candidates in descending order of output value (high possibility of abnormal shadow candidates) Similar results may be obtained. Here, the predetermined number is a value determined so as to satisfy a predetermined standard for a difference between detection results or a predetermined second standard for each detection result, or to approach these standards. is there.

  The abnormal shadow candidate detection method, apparatus, and program of the present invention are implemented by comparing a detection result obtained by detecting an abnormal shadow candidate from each of two or more images representing the same subject and having different viewing directions. Until the detection result difference satisfies a predetermined criterion, the detection level of the abnormal shadow candidate is changed, and the abnormal shadow candidate is detected based on the changed detection level. Each detection result when a predetermined criterion is satisfied is defined as a final result. Therefore, since the identity of the three-dimensional positions of the abnormal shadow candidates detected from these two or more images is not judged, the abnormal shadow candidates are more accurately and anatomically consistent with a simpler method. Therefore, the detection accuracy can be improved. In addition, the diagnostician can make a diagnosis based on an appropriate final result without a large difference between the respective images to be detected, so that the diagnostic ability is improved.

  When the detection level is changed so that the difference between the detection results approaches a predetermined reference and the detection result approaches a predetermined second reference, the detection result between two or more images Since not only the difference but also the detection result of each image becomes more appropriate, the detection accuracy is further improved and further contributes to the improvement of the diagnostic ability of the diagnostician.

  In addition, when the detection level is changed so that the number of detected abnormal shadow candidates is reduced, and abnormal shadow candidates are detected based on the changed detection level, instead of doing this, based on the detection level before the change If a predetermined number of abnormal shadow candidates are selected in descending order of the output value (high possibility of abnormal shadow candidates) from the already detected abnormal shadow candidates, the abnormal shadow candidate detection process is performed again. Since there is no need to do this, the processing efficiency is improved.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

The abnormal shadow candidate detection apparatus A according to the first embodiment of the present invention is a part of the computer-aided image diagnostic system 101 (see FIG. 1). As shown in FIG. 1, the computer-aided diagnostic imaging system 101 includes image data P _ML (hereinafter referred to as a side image) of a side image obtained by sandwiching the breast of the same subject in the side direction and photographing from the side direction. and that the data P _ML), a storage unit 91 for the breast image data P _CC of the front image obtained by photographing a vertically sandwiched in the vertical direction (hereinafter, stores and) that front image data P _CC, A minute calcification shadow candidate is detected from each of the side image image _PML image (hereinafter referred to as side image _PML ) and the front image data image _CC (hereinafter referred to as front image _PCC ). These images _PML and P _CC are detected by the calcified shadow candidate detecting means 10, the threshold changing means 30 for changing the detection thresholds T1 and T2 of the microcalcified shadow candidates, and the microcalcified shadow candidate detecting means 10. The number of microcalcification shadow candidates (hereinafter referred to as the number of candidates) detected from each of _NML and N _CC is compared, and the threshold changing means 30 is informed until the difference ΔN becomes equal to or less than a predetermined reference value ΔN _max. and the detection result of the difference .DELTA.N the detection threshold T1, T2 microcalcification shadow candidate for at least one of these images P _ML and P _CC are changing so as to approach to the predetermined reference value .DELTA.N _max, altered detected The detection result when the difference ΔN in the detection result is less than or equal to the predetermined reference value ΔN _max is repeated by causing the micro calcification shadow candidate detecting means 10 to detect the micro calcification shadow candidate based on the threshold values T1 and T2. a detection control unit 50 for outputting each of the end result, on the basis of the final result, the local image processing means 93 for enhancement of microcalcification shadow candidate region in each of the side image P _ML and the front image P _CC, also Images with Gradation processing with respect to the side surface image data P _ML and the front image data P _CC is data, the entire image processing unit 94 for performing image processing of the frequency processing, etc., sides after being image processed by the entire image processing unit 94 image data P _ML 'and the front image data P _CC' and the entire image represented with a side partial image data P1 _ML 'the front partial image data P1 _CC' and the minute represented after being image processed by the local image processing means 93 Display means 92 for displaying the calcified shadow candidate as a visible image. Microcalcification shadow candidate detecting section 10 includes a morphology filter processing unit 11 that performs processing by morphology filter with respect to the side surface image data P _ML and the front image data P _CC, the output value Mo by morphology filter, the Mgrad, initialization The threshold processing unit 12 performs threshold processing based on the detection thresholds T1 and T2 that have been changed or have been changed by the threshold changing means 30, and detects candidates for microcalcification shadows.

  Here, the microcalcification shadow candidate detection means 10, the threshold value changing means 30, the detection control means 50, the local image processing means 93, and the whole image processing means 94 are a server computer (hereinafter referred to as an image processing server) that performs image processing. ) Is implemented.

Storage means 91, and a side image data P _ML and the front image data P _CC is the original image data, and stored in association with the information about the imaging conditions of the subject and the image which is the subject of It is implemented in a server computer (hereinafter referred to as an image management server) having a large-capacity external storage device that includes a database that can be searched using information as a search key.

  The display means 92 is mounted on a personal computer (hereinafter referred to as client PC) having a high-definition liquid crystal display.

  An image processing server, an image management server, a client PC, and an image capturing / reading apparatus that captures and acquires a radiation image of a breast can communicate via a network.

Next, processing performed by the computer-aided image diagnostic system 101 will be described with reference to the flowchart of FIG. Incidentally, in the storage unit 91 is acquired by the imaging and reading apparatus via the network and a side image data P _ML was sent to the image management server and the front image data P _CC are assumed to be stored . In addition, it is assumed that all image data in the present embodiment is data with high luminance and high signal level.

First, morphology filter processing unit 11 of the microcalcification shadow candidate detecting section 10 performs calculation according to the following equation (1) to the side face image data P _ML, acquires the output value Mo (# 1).

  Here, Bi is M linear structural elements B (i = 1, 2, 3, and 4 in FIG. 3), and is set larger than the microcalcification shadow to be detected.

In Expression (1), first, a process of searching for the minimum value in a predetermined width centered on the target pixel, determined according to the structural element B (erosion process; see FIG. 4B) Then, a process for searching for the maximum value in the predetermined width (dilation process; see FIG. 4A) is performed (opening process; see FIG. 4C). By this opening process, the minute calcified image that is a convex data change portion (image portion that fluctuates in a spatially narrow range) finer than the structural element B is removed. On the other hand, the elongated non-calcified shadow has a length longer than that of the structural element B, and a portion whose inclination (extending direction) coincides with any of the M structural elements Bi remains as it is (formula (1)). Calculation of the second term). Therefore, by subtracting the obtained smoothed image by opening processing (image calcified pattern has been removed) from the side image data P _ML, images that contain only candidates for microcalcification image is obtained.

In addition, even in this case, a part of the non-calcified shadow having the same size as the micro-calcified shadow may remain. In such a case, the differential information based on the morphological operation of the following equation (2) is used. The non-calcification image contained in Mo of Formula (1) is further removed.

  In equation (2), the greater the output value Mgrad, the greater the possibility of a microcalcification shadow.

  Note that these output values Mo and Mgrad are stored in the memory of the image processing server or output to a file and can be used in subsequent processing.

  Next, the threshold processing unit 12 determines whether the output value Mo of the morphology filter processing unit 11 is based on the detection thresholds T1 and T2 that are initially set in a memory in which the detection thresholds T1 and T2 are stored (hereinafter referred to as threshold memory). If the output value Mgrad is larger than the detection threshold T2 and the output value Mgrad is larger than the detection threshold T2, the region having the output values Mo and Mgrad is detected as a candidate for the microcalcification shadow, and in other cases, A region having the output values Mo and Mgrad is not detected as a candidate for a minute calcified shadow (# 2).

Performs a process according to the morphology filter processing unit 11 and the threshold processing unit 12 in the same manner as the side image data P _ML also the front image data _{P CC (# 1, # 2} ), for detecting a microcalcification shadow candidate.

Next, the detection control means 50 counts the number of microcalcification shadow candidates N _ML and N _CC detected from each of the images P _ML and P _CC, and the difference ΔN between these two candidates and a predetermined number The reference value ΔN _max is compared (# 3). Since the side image _PML and the front image _PCC have the same breast as the subject, the reference value ΔN _max of the difference between these two candidates is ideally 0, but in both directions Any value may be found empirically as a realistic difference in the number of microcalcification shadows that can be detected from

When this difference .DELTA.N is larger than the reference value ΔN _max (# 3), such that the number of candidates N _ML or N _CC approaches the predetermined reference value N _max, the threshold value changing means changes the detection threshold T1 and T2 30 In addition, the threshold processing unit 12 performs threshold processing based on the detection thresholds T1 and T2 after the change. Here, the predetermined reference value N _max is a value empirically found as a realistic maximum value of the number of microcalcification shadows that can be detected from the image representing the breast.

Specifically, the larger one of the number of candidates N _ML and N _CC (hereinafter described as an example where N _ML > N _CC ) is compared with the reference value N _max (# 4), and N _ML When> N _max , the threshold changing unit 30 is requested to increase the detection thresholds T1 and T2 in order to make it difficult to detect the microcalcification shadow candidate from the side image _PML . In response to this request, the detection threshold values T1 and T2 stored in the threshold memory are updated to larger values (# 5). Here, “Y” is set to the threshold UP flag for the side image _PML (# 6). The meaning of this flag will be described later. After completing the change of the detection thresholds T1 and T2, the detection control means 50 sets “Y” in the reprocessing flag (# 8), and the output value of the morphology filter processing unit 11 having the side image data _PML as input. A request to reprocess Mo and Mgrad based on the detection threshold values T1 and T2 after the change is made to the threshold processing unit 12 (# 13). In response to this request, the threshold processing unit 12 performs threshold processing based on the detection thresholds T1 and T2 after change set in the threshold memory (# 2). On the other hand, in the case of N _ML ≦ N _max (# 4), N _CC <N _ML ≦ N _max is satisfied, so that a microcalcification shadow candidate is detected from the front image P _CC farther from the reference value N _max. For the sake of simplicity, a request to lower the detection thresholds T1 and T2 is made to the threshold changing unit 30, and the threshold changing unit 30 makes the detection thresholds T1 and T2 stored in the threshold memory smaller in response to this request. Update to a value (# 7). After completing the change of the detection thresholds T1 and T2, the detection control means 50 sets “Y” in the reprocessing flag (# 8), and the output value of the morphology filter processing unit 11 that receives the front image data _PCC as an input. A request to reprocess Mo and Mgrad based on the detection threshold values T1 and T2 after the change is made to the threshold processing unit 12 (# 13). In response to this request, the threshold processing unit 12 performs threshold processing based on the detection thresholds T1 and T2 after change set in the threshold memory (# 2). In the case of N _ML <N _CC , the processing is performed by replacing the side surface (ML) and the front surface (CC) described above with each other.

On the other hand, if the difference ΔN in the number of candidates is less than or equal to the reference value ΔN _max (# 3), after setting “N” in the reprocessing flag once (# 9), whether or not the number of candidates is 0 for each candidate number And, it is determined whether or not the threshold UP flag for an image whose candidate number is 0 is “Y” (# 10). For an image with the number of candidates being 0 and the threshold UP flag being “Y”, that is, an image in which the number of candidates has become 0 as a result of increasing the detection thresholds T 1 and T 2 by the threshold changing means 30, In order to facilitate detection of candidates, a request for lowering the detection threshold is made to the threshold changing unit 30, and the threshold changing unit 30 uses the detection thresholds T1 and T2 stored in the threshold memory in response to the request. Update to a smaller value (# 11). After completing the change of the detection thresholds T1 and T2, the detection control means 50 sets “Y” in the reprocessing flag (# 12), and the output value of the morphology filter processing unit 11 that receives the image data of the image as input. A request to reprocess Mo and Mgrad based on the detection threshold values T1 and T2 after the change is made to the threshold processing unit 12 (# 13). In response to this request, the threshold processing unit 12 performs threshold processing based on the detection thresholds T1 and T2 after change set in the threshold memory (# 2).

The above processing (# 3 to # 13) is repeated every time the threshold processing unit performs processing (# 2). When the number of candidates satisfies both of the following conditions (a) and (b), the value of the reprocessing flag is “N” (# 13), so that the detection control means 50 The microcalcification shadow candidate is output as the final result (# 14).
(A) The difference ΔN in the number of candidates is equal to or less than the reference value ΔN _max .
(B) The number of candidates is not 0, or remains 0 from the beginning.

Local image processing means 93, the final result and the side image data P _ML, as input front image data P _CC, emphasizing image processing microcalcification shadow candidate region in each of the side image P _ML and the front image P _CC Then, side local image data _P1ML 'and front local image data _P1CC ' are output.

Further, the whole image processing unit 94, as input and side image data P _ML and the front image data P _CC is the original image data, tone processing, performs image processing of the frequency processing, etc., side view image after processing Data P _ML ′ and front image data P _CC ′ are output.

Side local image data P1 _ML ′ and front local image data P1 _CC ′ obtained by processing by the local image processing means 93 and side image data P _ML ′ obtained by processing by the whole image processing means 94 and the front image The data P _CC ′ is input to the display unit 92. The display unit 92 displays an image based on the side image data P _ML ′ and the front image data P _CC ′, while only the image portion of the microcalcification shadow candidate is a local image. The image is processed by the processing means 93 and replaced with the image by the side local image data _P1ML 'and the front local image data _P1CC '. As a result, the display unit 92 displays an image of the micro calcified shadow candidate that is emphasized more than the entire image together with the entire image representing the entire subject, and is provided for diagnosis by a doctor or other diagnostician.

As described above, according to the abnormal shadow candidate detection apparatus A according to the first embodiment of the present invention, the detection control means 50 uses the micro calcified shadow candidate detection means 10 to each of the side image _PML and the front image _PCC . The candidate numbers N _ML and N _{CC of} the microcalcification shadow candidates detected from the above are compared, and the threshold value changing means 30 makes these images P _ML and P _CC until the difference ΔN becomes a predetermined reference value ΔN _max or less. The detection thresholds T1 and T2 of the microcalcification shadow candidate for at least one of the above are changed so that the difference ΔN in the detection results approaches the predetermined reference value ΔN _max , and the detection thresholds T1 and T2 are small The calcified shadow candidate detection means 10 is repeatedly made to detect the minute calcified shadow candidates, and each of the detection results when the difference ΔN in the detection results becomes equal to or smaller than the predetermined reference value ΔN _max is output as the final result. . Therefore, since the identity of the three-dimensional positions of the abnormal shadow candidates detected from these two or more images is not judged, the abnormal shadow candidates are more accurately and anatomically consistent with a simpler method. Therefore, the detection accuracy can be improved. Further, the diagnostician appropriately performs image processing by the local image processing unit 93 and the whole image processing unit 94 based on an appropriate final result that does not greatly differ between the respective images to be detected, and then displays the image by the display unit 95. Since it becomes possible to make a diagnosis using the displayed side and front images, the diagnostic ability is improved.

Further, the detection control means 50 is a micro lime that can detect the difference between detection results ΔN and the predetermined reference value ΔN _max and the individual candidate numbers _NML and _NCC from the image representing the breast. Since the detection thresholds T1 and T2 are changed so as to approach the value N _max empirically found as the realistic maximum value of the number of shaded shadows, not only the difference in the detection results between the two images, Since the detection results of the individual images are also more appropriate, the detection accuracy is further improved and further contributes to the improvement of the diagnostic ability of the diagnostician.

The abnormal shadow candidate detection apparatus B according to the second embodiment of the present invention is a part of the computer-aided image diagnostic system 102 (see FIG. 5). As shown in FIG. 5, the computer-aided image diagnosis system 102 includes side image data _PML obtained by sandwiching the breast of the same subject in the lateral direction and photographing from the lateral direction, and the breast in the vertical direction. sandwiched in a storage unit 91 for storing the front image data P _CC obtained by photographing from the vertical direction, the side image P _ML and the front image P _CC detect tumor shadow candidate detecting means from tumor shadow candidate each in 20, tumor and detecting rank changing means 40 for changing the detection rank R of shadow candidate, tumor by shadow candidate detecting section 20 of the image P _ML and P _CC number each detected tumor pattern from the candidates of the N _ML and N comparing the _CC, until the difference .DELTA.N falls below a predetermined reference value .DELTA.N _max, the detection rank R of tumor shadow candidate for at least one of these images P _ML and P _CC in detecting rank change unit 40 Out and that the results of the difference .DELTA.N causes changes so as to approach to the predetermined reference value .DELTA.N _max, repeatedly and thereby detecting the masses shadow candidate tumor shadow candidate detecting section 20 based on the changed detection rank R, detected A detection control means 50 for outputting each detection result as a final result when the result difference ΔN is equal to or smaller than the predetermined reference value ΔN _max , and a rectangular region of interest composed of this tumor shadow candidate region and its neighboring region a region of interest setting means 95 for setting a W, an image representing the region of interest W (hereinafter, referred to as ROI image W) and displays together with side images P _ML and the front image P _CC, for the detected tumor shadow candidate Display means 92 for displaying the index value. The mass shadow candidate detecting means 20 is initialized based on the iris filter processing unit 21 that performs processing by the iris filter on the side image data _PML and the front image data _PCC, and the output value I by the iris filter. Alternatively, the first candidate detection unit 22 that detects primary candidates that can be tumor shadow candidates by the number of detection ranks R that have been changed by the detection rank changing means 40, and the first candidate for each of the detected first candidates. An index value calculation unit 23 that calculates an index value indicating the characteristics of the next candidate, and a second candidate detection unit 24 that detects a secondary candidate for a tumor shadow based on the calculated index value. .

  Here, the tumor shadow candidate detection means 20, the detection rank change means 40, the detection control means 50, and the region-of-interest setting means 95 are mounted on a server computer (hereinafter referred to as an image processing server) that performs image processing.

Storage means 91, and a side image data P _ML and the front image data P _CC is the original image data, and stored in association with the information about the imaging conditions of the subject and the image which is the subject of It is implemented in a server computer (hereinafter referred to as an image management server) having a large-capacity external storage device that includes a database that can be searched using information as a search key.

  The display means 92 is mounted on a personal computer (hereinafter referred to as client PC) having a high-definition liquid crystal display.

  An image processing server, an image management server, a client PC, and an image capturing / reading apparatus that captures and acquires a radiation image of a breast can communicate via a network.

Next, processing performed by the computer-aided image diagnostic system 102 will be described with reference to the flowchart of FIG. Incidentally, in the storage unit 91 is acquired by the imaging and reading apparatus via the network and a side image data P _ML was sent to the image management server and the front image data P _CC are assumed to be stored . In addition, it is assumed that all image data in the present embodiment is data of high density and high signal level.

The iris filter processing unit 21 of the tumor shadow candidate detecting means 20 performs the following processing on the side image data _PML (# 21).

First, with respect to all the pixels constituting the target image, the orientation θ of the gradient vector of the pixel value of the image data is obtained for each pixel m based on the calculation formula shown in the following formula (3).

  Here, as shown in FIG. 7, f1 to f16 are pixel values (image data) corresponding to the pixels on the outer periphery of the mask of 5 vertical pixels × 5 horizontal pixels around the pixel j.

  Next, as shown in FIG. 8, consider radial lines in M types of directions every 2π / M degrees (in FIG. 8, 32 directions every 11.25 degrees are illustrated) with the pixel of interest at the center. For each line, the output value up to the pixel at which the maximum gradient vector concentration is obtained among the pixels on the line is defined as the concentration Cimax in that direction, and the concentration Cimax is averaged in all directions. Then, the degree of concentration C of the gradient vector group for the target pixel is calculated.

Specifically, first, the degree of concentration Ci (n) obtained from the target pixel to the nth pixel on the i-th radial line is obtained by the following equation (4).

  Here, θij is an angle formed by a line segment connecting the j-th pixel on the i-th radial line and the target pixel, and the gradient vector calculated by Equation (3) in the j-th pixel.

  Therefore, for each of the natural numbers n between Rmin and Rmax, the expression (4) is obtained by setting the target pixel as the starting point and the end point as the nth pixel from the target pixel on the i-th radial line. The average value of the components in the pixel direction of the gradient vector in each pixel is calculated as the degree of concentration Ci (n). Here, Rmin and Rmax are the minimum value and the maximum value of the radius of the tumor shadow to be extracted, respectively.

Next, the degree of concentration C of the gradient vector group is calculated by the following equations (5) and (6).

Here, Cimax in Expression (5) is the maximum value of the concentration degree Ci (n) for each radial line obtained in Expression (4), and therefore the concentration degree Ci (n) from the target pixel is the maximum value. The region up to the pixel becomes the candidate region for the tumor shadow in the direction of the line. Then, by calculating equation (5) for all radial lines to determine the area of the tumor shadow on each line, by connecting the area of the tumor shadow on each line with a straight line or a non-linear curve between adjacent lines, The shape of the outer peripheral edge of the region that can be a candidate for the shadow of the tumor can be specified. Note that the coordinates ([x], [y]) on the i-th line and the n-th pixel from the target pixel are expressed by the following formula (7), where the coordinate of the target pixel is (k, l). ).

However, [x] and [y] are maximum integers that do not exceed x and y.

  Further, in Expression (6), the maximum concentration value Cimax given in Expression (5) in this region is averaged in all directions of the radial line (in Expression (6), the case of 32 directions is exemplified). Ask for. This obtained value is the output value I of the iris filter process. The output value I is stored in the memory of the image processing server or output to a file, and can be used in subsequent processing.

The primary candidate detection unit 22 rearranges the iris filter output values I for the side image data _{PML in} descending order, and stores the detection rank R in descending order of the output value I (hereinafter, detection rank memory). The number of primary candidates for tumor shadows is detected by the number initially set (# 22).

Next, the index value calculation unit 23 calculates the following feature amount for each detected primary candidate (# 23). First, as the first feature amount, since the contour of the tumor shadow is a shape close to a circle, the circularity Sp (Spreadness) is used. As shown in FIG. 9, the area A and the center of gravity AO of the detected tumor shadow (solid line portion) as the first candidate are obtained, and the radius R having an area equivalent to the area A around the center of gravity AO is obtained. Assuming a virtual circle (broken line portion in the figure), the degree of circularity is calculated as an occupancy ratio of the primary candidates included in the virtual circle with respect to the area A. That is, assuming that the area of the portion where the virtual circle and the candidate region overlap is A ′, the circularity is calculated by the following equation (8).

Next, the following three feature values are used as the feature values inside the candidate area. That is, a histogram of the density value S of the primary candidate is created, and the frequency of the density value S is P (S). Based on this density value S and frequency P (S), The second feature amount (9), the third feature amount (10) representing the contrast con, and the fourth feature amount (11) representing the angular moment asm are calculated.

  In addition, edge information such as dispersion, bias, correlation value, moment, and entropy for a simultaneous generation matrix generated based on an IFED image (Iris Filter EDge) can also be used as a feature amount (Japanese Patent Laid-Open No. 2003-115041, etc.) reference). Hereinafter, a case where nine feature amounts including these are calculated will be described as an example.

  The secondary candidate detection unit 24 calculates the Mahalanobis distance based on these feature amounts calculated by the index value calculation unit 23, performs a determination using the likelihood ratio, and detects a secondary candidate for the mass shadow. (# 24).

First, the Mahalanobis distance Dm1 from the pattern class w1 indicating the non-malignant shadow and the Mahalanobis distance Dm2 from the pattern class w2 indicating the malignant shadow, which are experimentally obtained in advance, are calculated by the following equation (12).

  The feature values calculated by the index value calculation unit 23 correspond to x1 to x9, respectively, and represent a nine-dimensional space (x1, x2, x3,..., X9). The Mahalanobis distance between the primary candidate pattern expressed in the 9-dimensional pattern space and the non-malignant shadow pattern is Dm1, and similarly, the Mahalanobis distance between the malignant shadow pattern is Dm2.

  The non-malignant shadow pattern and the malignant shadow pattern are pattern spaces defined by vectors x for each non-malignant shadow and each malignant shadow, which are set based on the results of an experimental investigation of a large number of tumor shadow candidates in advance. Means. For example, the pattern class w1 formed by the average of the vector x for the non-malignant shadow and the pattern class w2 formed by the average of the vector x for the malignant shadow are shown.

  Here, the secondary candidate detection unit 24 calculates the Mahalanobis distance using a non-malignant shadow pattern and a malignant shadow pattern in a preset subject area.

  Next, when the candidate area is a malignant shadow, the Mahalanobis distance from the pattern class of the malignant shadow is close (Dm2 indicates a low value), and the Mahalanobis distance from the pattern class of the non-malignant shadow tends to vary. When the candidate area is a non-malignant shadow, the Mahalanobis distance from the pattern class of the non-malignant shadow is close (Dm1 indicates a low value), and the Mahalanobis distance from the pattern class of the malignant shadow tends to vary. Therefore, a likelihood ratio that can significantly distinguish a malignant shadow and a non-malignant shadow according to this tendency is calculated for each candidate region.

  The likelihood ratio is expressed by Dm1 / Dm2, and indicates the inclination on the coordinate plane of FIG. That is, it can be determined that the likelihood ratio is higher as the likelihood ratio is higher, and the possibility that the non-malignant shadow is higher as the likelihood ratio is smaller. For example, when the likelihood ratio is 2 or more, the threshold is set to 2. When it is less than 2, it is determined to be non-malignant.

  As described above, the secondary candidate detection unit 24 evaluates the feature amount of each primary candidate, and detects those determined to be malignant as secondary candidates for tumor shadows.

Similarly to the side image data _PML , the front image data _PCC is processed by the iris filter processing unit 21, the primary candidate detection unit 22, the index value calculation unit 23, and the secondary candidate detection unit 24 (# 21 ˜ # 24), detecting a secondary candidate for a mass shadow.

Next, the detection control means 50 counts the number of secondary candidates (hereinafter referred to as the number of candidates) N _ML and N _CC of the mass shadow detected from each of the images P _ML and P _CC, and these two The difference ΔN in the number of candidates is compared with a predetermined reference value ΔN _max (# 25). Since the side image _PML and the front image _PCC have the same breast as the subject, the reference value ΔN _max of the difference between these two candidates is ideally 0, but in both directions Any value that can be found empirically as a realistic difference in the number of mass shadows that can be detected from the above.

When the difference ΔN is larger than the reference value ΔN _max (# 25), the predetermined reference value N _max is compared with the number of candidates N _ML and N _CC (# 26). Here, the predetermined reference value N _max is a value empirically found as a realistic maximum value of the number of mass shadows that can be detected from an image representing the breast.

When the reference value N _max is the largest among these three values (# 26), in order to make it easier to detect the secondary candidate from the image with the smaller number of candidates, that is, the image farther from the reference value N _max. A request to increase the detection rank R is made to the detection rank changing means 40, and the detection rank changing means 40 updates the detection rank R stored in the detection rank memory to a larger value in response to this request (# 27). After completing the change of the detection rank R, the detection control means 50 sets “Y” for the re-detection flag and “N” for the re-comparison flag (# 28), and the image data of the image with the smaller number of candidates. The first candidate detection unit 22 is requested to perform processing based on the detection rank R after the change for the output value I of the iris filter processing unit 21 that receives the input (# 35). In response to this request, the primary candidate detection unit 22 rearranges the iris filter output values I for the image data in descending order, and detects the changed detection values set in the detection rank memory in descending order of the output value I. As many as the number of ranks R, primary candidates for tumor shadows are detected (# 22), and in the same manner as described above, index values are calculated (# 23) and secondary candidates are detected (# 24). Again, the difference ΔN in the number of candidates is compared with a predetermined reference value ΔN _max (# 25).

If the reference value _Nmax is the second largest among these three values (# 26), the candidate number of the image with the larger candidate number is replaced with the value of the reference value _Nmax (# 29). As a result, the detection rank R is changed so that the number of candidates of the image with the larger number of candidates matches the value of the reference value N _max , the detection of the primary candidate (# 22), and the calculation of the index value (# 23), the same result as the detection of the secondary candidate (# 24) is obtained. Then, “N” is set in the re-detection flag and “Y” is set in the re-comparison flag (# 30), and the difference ΔN between the number of candidates and the predetermined reference value ΔN _max are compared again (# 34, # twenty five).

When the reference value N _max is the smallest among these three values (# 26), the candidate with the larger number of candidates, that is, the image candidate farther from the reference value N _max is set as the candidate with the smaller number of candidates, that is, Replace with the value of the number of candidates for an image closer to the reference value N _max (# 31). Thus, by changing the detection rank R to match the number of candidates of the farther image from the reference value N _max the number of candidates values towards the image close to the reference value N _max, of the first primary candidate of The same results are obtained as when detection (# 22), index value calculation (# 23), and secondary candidate detection (# 24) are performed. Then, “N” is set in the re-detection flag and “N” is set in the re-comparison flag (# 32).

If the difference ΔN in the number of candidates is less than or equal to the reference value ΔN _max (# 25), “N” is set in the redetection flag and “N” is set in the recomparison flag (# 33).

The above process is repeated, and when the difference ΔN in the number of candidates becomes equal to or smaller than the reference value ΔN _max (including the case where the number of candidates N _ML and N _CC match by the process of # 32 described above), re-detection is performed. flag, the value is both "N" and the re-comparison flag, the detection control unit 50, the detected tumor pattern candidates from the respective image P _ML and P _CC, the number of candidates of each of the time N _ML, N _CC Are output as the final result (# 34, 35, # 36).

Region of interest setting means 95, based on the output the final result, the local area of each of the image P _ML, a predetermined shape including a tumor shadow detected from P _CC candidate region and its neighboring region (e.g., rectangle or a circle) Set as the region-of-interest image W. When there are a plurality of detected tumor shadow candidates, a plurality of region-of-interest images W are also set. The region-of-interest setting means 95 outputs image data W representing the set region-of-interest image W.

Display means 92, and a side image data P _ML and the front image data P _CC, the index value for each of the final candidate of the detection controlling section 50 outputs mass shadow (features) K and the output region of interest setting means 95 Image data W representing the region-of-interest image W obtained, and the original side image _PML , front image P _CC , region-of-interest image W and index value K are simultaneously displayed on the screen for diagnosis by a diagnostician such as a doctor. Provided.

The abnormal shadow candidate detecting device B according to the second embodiment of the present invention as described above, the detection control unit 50, the tumor shadow candidate detecting section 20 is detected from each of the side image P _ML and the front image P _CC The number of candidate tumor shadows N _ML and N _CC are compared, and until the difference ΔN is equal to or smaller than a predetermined reference value ΔN _max , the detection rank changing means 40 makes a tumor for at least one of these images P _ML and P _CC. The shadow candidate detection rank R is changed so that the difference ΔN in the detection results approaches the predetermined reference value ΔN _max , and the tumor shadow candidate detection means 20 detects the tumor shadow candidate based on the changed detection rank R. Each of the detection results when the difference ΔN between the detection results becomes equal to or smaller than the predetermined reference value ΔN _max is output as a final result. Therefore, since the identity of the three-dimensional positions of the abnormal shadow candidates detected from these two or more images is not judged, the abnormal shadow candidates are more accurately and anatomically consistent with a simpler method. Therefore, the detection accuracy can be improved. Further, the diagnostician displays an index value K related to the determination as to whether or not to detect as a tumor shadow candidate, which is displayed by the display unit 92 based on an appropriate final result that does not greatly differ between the respective images to be detected. , the detected tumor shadow candidate, the original image P _ML, it becomes possible to diagnose by using P _CC, diagnostic performance is improved.

The detection control means 50, detected as the result of the difference .DELTA.N approaches the predetermined reference value .DELTA.N _max, and, and the number of individual candidate N _ML and N _CC, detected may tumor pattern from an image representing the breast Since the detection rank R is changed so as to approach the value N _max empirically found as a realistic maximum value of the number of images, not only the difference in the detection results between the two images, but also the individual images Since the detection result is also more appropriate, the detection accuracy is further improved, which further contributes to the improvement of the diagnostic ability of the diagnostician.

  Further, the detection control means 50 changes the detection rank R so that the number of tumor shadow candidates detected by the detection rank changing means 40 decreases, and the tumor shadow candidate is selected as a tumor shadow candidate based on the changed detection rank R. When detecting the detection means 20, instead of doing this, a high output value and index value are selected from the tumor shadow candidates already detected based on the detection rank R before the change (possibility of a tumor shadow candidate). Since a predetermined number of tumor shadow candidates are selected in descending order (# 29, # 31, # 36), it is not necessary to perform the detection process of the tumor shadow candidates again, and the processing efficiency is improved.

In the above description, when the difference ΔN in the number of candidates is larger than the reference value ΔN _max (# 25) and the predetermined reference value N _max is larger than the number of candidates N _ML and N _CC (# 26), the number of candidates The detection rank R is changed so that the secondary candidate is easily detected from the image with the smaller number (# 27), the primary candidate is detected (# 22), the index value is calculated (# 23), the first The secondary candidate detection (# 24) is performed again, but the primary candidate detection unit 22 stores the detection rank before the change in the memory so that the difference from the detection rank after the change can be calculated. Then, only the primary candidate corresponding to the detection rank increment can be additionally detected, so that the calculation of the subsequent index value (# 23) and the detection of the secondary candidate (# 24) It is not necessary to perform each of these processes again on the tumor shadow candidate that has already been detected as the second candidate, and the processing efficiency is further improved.

  In these two embodiments, each means is distributed and implemented in a plurality of computers, but some or all of these may be implemented in one computer. This is useful for small systems.

1 is a block diagram showing a configuration of a computer-aided image diagnostic system including an abnormal shadow candidate detection apparatus A according to a first embodiment of the present invention. The flowchart which shows the flow of a process of the abnormal shadow candidate detection apparatus A used as the 1st Embodiment of this invention. A diagram showing the concept of structural elements used in morphological operation processing A graph explaining the basic operation of morphological operation processing The block diagram which shows the structure of the computer-aided image diagnostic system containing the abnormal shadow candidate detection apparatus B used as the 2nd Embodiment of this invention. The flowchart which shows the flow of a process of the abnormal shadow candidate detection apparatus B used as the 2nd Embodiment of this invention. The figure which shows the mask which calculates the gradient vector in an iris filter process Conceptual diagram showing an iris filter set so that its contour shape changes adaptively The figure which shows the virtual circle which has an area equivalent to the area A of a candidate area | region Diagram explaining likelihood ratio

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Micro calcification shadow candidate detection means 11 Morphology filter processing part 12 Threshold processing part 20 Mass shadow candidate detection means 21 Iris filter processing part 22 Primary candidate detection part 23 Index value calculation part 24 Secondary candidate detection part 30 Threshold change Means 40 Detection rank changing means 50 Detection control means 91 Storage means 92 Display means 93 Local image processing means 94 Whole image processing means 95 Region of interest setting means 101 Computer-aided image diagnostic system 102 including the first embodiment of the present invention Computer-aided image diagnosis system A including the second embodiment of the present invention The abnormal shadow candidate detection device B according to the first embodiment of the present invention The abnormal shadow candidate detection device of the second embodiment of the present invention

Claims (6)

An abnormal shadow candidate is detected from each of two or more images representing the same subject and having different viewing directions;
Comparing detection results of the abnormal shadow candidates for each of the two or more images;
Changing the detection level of the abnormal shadow candidate for at least one of the two or more images so that the difference between the detection results approaches the predetermined reference until the difference between the detection results satisfies a predetermined reference; Repeatedly detecting the abnormal shadow candidate based on the changed detection level;
A method for detecting an abnormal shadow candidate, characterized in that each of the detection results when the difference between the detection results satisfies the predetermined criterion is a final result.   Until the difference between the detection results satisfies the predetermined criterion, the detection level of the abnormal shadow candidate for at least one of the two or more images is set so that the difference between the detection results approaches the predetermined criterion, and The abnormality according to claim 1, wherein the detection result is repeatedly changed so as to approach a predetermined second reference, and the abnormal shadow candidate is detected based on the changed detection level. Shadow candidate detection method. An abnormal shadow candidate detecting means for detecting an abnormal shadow candidate from each of two or more images representing the same subject and having different viewing directions;
Detection level changing means for changing the detection level of the abnormal shadow candidate;
The detection result by the abnormal shadow candidate detection unit for each of the two or more images is compared, and at least one of the two or more images is sent to the detection level changing unit until a difference between the detection results satisfies a predetermined criterion. The detection level of the abnormal shadow candidate is changed so that the difference between the detection results approaches the predetermined reference, and the abnormal shadow candidate detection unit detects the abnormal shadow candidate based on the changed detection level. And a detection control means for outputting each of the detection results when the difference between the detection results satisfies the predetermined criterion as a final result.   The detection control means compares the detection results by the abnormal shadow candidate detection means for each of the two or more images, and until the difference between the detection results satisfies the predetermined criterion, the detection level changing means The detection level of the abnormal shadow candidate for at least one of two or more images is changed so that the difference between the detection results approaches the predetermined reference and the detection result approaches the predetermined second reference And causing the abnormal shadow candidate detection means to detect the abnormal shadow candidate based on the changed detection level, and the detection result when the difference between the detection results satisfies the predetermined criterion The abnormal shadow candidate detection apparatus according to claim 3, wherein each of the output is output as a final result. On the computer,
An abnormal shadow candidate is detected from each of two or more images that represent the same subject and have different directions for viewing the subject,
Comparing the detection results of the abnormal shadow candidates for each of the two or more images;
Changing the detection level of the abnormal shadow candidate for at least one of the two or more images so that the difference between the detection results approaches the predetermined reference until the difference between the detection results satisfies a predetermined reference; Detecting the abnormal shadow candidate based on the changed detection level,
An abnormal shadow candidate detection program that outputs each of the detection results when the difference between the detection results satisfies the predetermined criterion as a final result.   Until the difference between the detection results satisfies the predetermined criterion, the detection level of the abnormal shadow candidate for at least one of the two or more images is set so that the difference between the detection results approaches the predetermined criterion, and 6. The method according to claim 5, wherein the detection result is changed so as to approach a predetermined second reference, and the abnormal shadow candidate is detected based on the changed detection level. Abnormal shadow candidate detection program.

Images