System and method for detecting abnormalities in medical images

Abstract

This abnormality detection system (50) is a system for detecting, on the basis of a medical image, presence or absence of an abnormality in one preliminarily specified anatomic structure included in the medical image. The abnormality detection system comprises: an acquisition unit (52) that acquires block information showing a block, which is a region in which pixels showing the anatomic structure in the medical image are connected; and a determination unit (53) that determines whether the anatomic structure is abnormal or not on the basis of the number of blocks shown by the block information acquired by the acquisition unit (52), and outputs the same. The determination unit (53) determines that the anatomic structure is not abnormal when the number of blocks shown by the block information is 1 and determines that the anatomic structure is abnormal when the number of blocks shown by the block information is 2 or more and the blocks shown by the block information meet a predetermined condition.

Description

[Technical field]

[0001] The present disclosure relates to an anomaly detection system and method for medical images, and in particular to a system for detecting the presence or absence of an abnormality in an anatomical structure included in a medical image. [Background technology]

[0002] Conventionally, a technique has been proposed for detecting the presence or absence of abnormalities in anatomical structures such as organs based on medical images (see, for example, Patent Document 1). In Patent Document 1, image segmentation is performed on the medical image to separate a region of interest from the medical image, and the presence or absence of abnormalities is determined based on the centroid position, volume, shape, intensity, density, transparency, etc. of the separated region of interest. In this specification, "abnormal" / "normal" refers to "abnormal" / "normal" from a medical perspective (i.e., diagnostically). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2019-530488 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in a method that utilizes the characteristics of the region of interest separated by image segmentation, such as that described in Patent Document 1, it may not be possible to identify abnormal cases, for example, in cases where other abnormalities such as tumors or water overlap part of the normal structure of an organ in a medical image.

[0005] Therefore, an object of the present disclosure is to provide an abnormality detection system and method for medical images that can detect the presence or absence of an abnormality in an anatomical structure with higher accuracy. [Means for solving the problem]

[0006] In order to achieve the above object, a medical image anomaly detection system according to an embodiment of the present disclosure detects an anomaly in a medical image by detecting an anomaly in a medical image, the anomaly being detected as a single block, the anomaly being detected as a single block based on the medical image. That An anomaly detection system for detecting the presence or absence of an anomaly, the anomaly including a tumor and other anomalies including water superimposed on a part of a normal structure of an organ in a medical image, the anomaly detection system including an acquisition unit that acquires block information indicating a block that is an area in which pixels indicating the anatomical structure are connected in the medical image, and a number of blocks indicated by the block information acquired by the acquisition unit. Depends on the case a determination unit that determines whether or not the block is abnormal and outputs the result, and when the number of blocks indicated by the block information is 1, Depends on the case When it is determined that there is no abnormality, and the number of blocks indicated by the block information is two or more, and the block indicated by the block information satisfies a predetermined condition, Depends on the case It is determined to be abnormal.

[0007] In order to achieve the above object, a method for detecting an abnormality in a medical image according to an embodiment of the present disclosure includes detecting an abnormality in a medical image based on the medical image, the abnormality being detected as a single block of an anatomical structure that is identified in advance and is included in the medical image. That An anomaly detection method using an anomaly detection system for detecting the presence or absence of an anomaly, the anomaly including a tumor and other abnormalities including water superimposed on a part of a normal structure of an organ in a medical image, the anomaly detection method including the steps of: acquiring block information indicating a block in the medical image, the block being an area in which pixels indicating the anatomical structure are connected; and detecting the number of blocks indicated by the block information acquired in the acquiring step. Depends on the case and a determination step of determining whether or not the block is abnormal and outputting the result, wherein in the determination step, when the number of blocks indicated by the block information is 1, Depends on the caseWhen it is determined that there is no abnormality, and the number of blocks indicated by the block information is two or more, and the block indicated by the block information satisfies a predetermined condition, Depends on the case It is determined to be abnormal. Effect of the Invention

[0008] The present disclosure provides an abnormality detection system and method for medical images that can detect the presence or absence of an abnormality in an anatomical structure with higher accuracy. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing a configuration of a diagnosis support system according to an embodiment. [Figure 2A] FIG. 2A shows an example of an anatomical structure in a medical image. [Figure 2B] FIG. 2B shows another example of an anatomical structure in a medical image. [Figure 2C] FIG. 2C shows another example of an anatomical structure in a medical image. [Figure 2D] FIG. 2D shows another example of an anatomical structure in a medical image. [Diagram 3] FIG. 3 is a flowchart showing the main operations of the anomaly detection system in FIG. [Figure 4] FIG. 4 is a flow chart showing a specific example of step S23 in FIG. [Diagram 5] FIG. 5 is a schematic diagram showing examples of anatomical structures in a medical image for explaining the example results of the determinations shown in FIGS. [Figure 6] FIG. 6 is a diagram showing the results of an experiment using cases for evaluating the judgment shown in FIG. [Figure 7] FIG. 7 is a flow chart showing another specific example of step S23 in FIG. [Figure 8A]FIG. 8A is a schematic diagram showing an example of an anatomical structure in a medical image for explaining the example determination results shown in FIG. 3 and FIG. [Figure 8B] FIG. 8B is a diagram showing an example of imaging of an anatomical structure in a medical image for explaining the example result of the determination shown in FIG. [Figure 9] FIG. 9 is a flow chart showing yet another specific example of step S23 in FIG. [Figure 10A] FIG. 10A is a flowchart showing a method of determining the first threshold value used in the determination using the area shown in FIG. [Figure 10B] FIG. 10B is a flowchart showing a method of determining the second threshold value used in the determination using the distance shown in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the embodiments of the present disclosure will be described in detail with reference to the drawings. Each of the embodiments described below shows a specific example of the present disclosure. The numerical values, shapes, materials, components, the arrangement and connection of the components, steps, and the order of steps shown in the following embodiments are merely examples and are not intended to limit the present disclosure. In addition, each figure is not necessarily a strict illustration. In each figure, the same reference numerals are used for substantially the same configurations, and duplicated explanations are omitted or simplified.

[0011] 1 is a block diagram showing the configuration of a diagnosis support system 10 according to an embodiment. The diagnosis support system 10 is a system that supports doctors in efficiently diagnosing anatomical structures based on medical images, and includes an imaging device 20, an image server 30, a viewer device 40, and an abnormality detection system 50, all of which are connected via a communication path such as the Internet.

[0012] The imaging device 20 is a device that generates medical images 31a by imaging human anatomical structures, and in this embodiment, is an X-ray imaging device that generates chest X-ray images as medical images and stores them in the image server 30. Note that the medical images are not limited to chest X-ray images, and may be X-ray images of other parts of the body, or images provided by CT scans, PET / CT scans, SPECT scans, MRIs, ultrasound, X-rays, mammography, angiography, fluorescence imaging, micrographs, or combinations of these.

[0013] The image server 30 is a data server that holds the medical images 31a generated by the imaging device 20 and the post-detection medical images 32a generated by the abnormality detection system 50, and provides the held medical images 32b to the viewer device 40, and is, for example, a computer device equipped with a storage device such as a hard disk. Note that data exchanged between the image server 30 and other devices may include not only medical images but also various information associated with the medical images (photographing date and time, information on the patient, detection results, etc.).

[0014] The viewer device 40 is a device that displays the medical images 32b stored in the image server 30 and various information associated with the medical images in various formats in accordance with instructions from a user obtained via a GUI (Graphical User Interface), and is, for example, a computer device equipped with peripheral devices such as a display, an input device such as a mouse, and a storage device such as a hard disk.

[0015] The abnormality detection system 50 is a system that detects the presence or absence of a medical abnormality in a pre-specified anatomical structure contained in a medical image based on a medical image 31b of the detection target obtained from the image server 30, and includes a block information generation unit 51, an acquisition unit 52, a judgment unit 53, and an image processing unit 54.

[0016] The block information generating unit 51 generates block information indicating blocks, which are regions in which pixels indicating anatomical structures included in the medical image 31b are connected, from the medical image 31b to be detected, acquired from the image server 30; that is, it performs a labeling process on the medical image 31b to separate it into image chunks. The block information is not limited to a specific format as long as it is information that identifies each block in the medical image 31b. For example, the block information may be information that distinguishes between pixels that belong to each block in the medical image 31b and other pixels (for example, a binary image in which pixels that make up a block are set to "1" and other pixels are set to "0"), or may be information that indicates the characteristics of each block (for example, area and contour information of each block obtained by analyzing the binary image).

[0017] The acquisition unit 52 acquires the block information generated by the block information generation unit 51 .

[0018] The determination unit 53 determines whether or not an anatomical structure included in the medical image 31b is abnormal based on the number of blocks indicated by the block information acquired by the acquisition unit 52, and outputs the result. At this time, when the number of blocks indicated by the block information is 1, the determination unit 53 recognizes that an anatomical structure that should be detected as one block has been detected as one block, and determines that the anatomical structure included in the medical image 31b is not abnormal. On the other hand, when the number of blocks indicated by the block information is 2 or more, and the block indicated by the block information satisfies a predetermined condition, the determination unit 53 recognizes that an anatomical structure that should be detected as one block is not noise on the image or in image processing but is separated into two or more, and determines that the anatomical structure included in the medical image 31b is abnormal.

[0019] Here, the predetermined condition is, for example, a condition regarding the block having the second largest area in the medical image 31b among the two or more blocks indicated by the block information. More specifically, the predetermined condition is that the block having the second largest area in the medical image 31b among the two or more blocks indicated by the block information has an area equal to or larger than a first threshold value.

[0020] The predetermined condition is, for example, that the minimum value of the distance between the block having the largest area in the medical image 31b and each of the other blocks among the two or more blocks indicated by the block information is equal to or less than a second threshold value. In this case, the predetermined condition may further include that the block having the second largest area in the medical image 31b among the two or more blocks indicated by the block information has an area equal to or greater than a first threshold value.

[0021] The image processing unit 54 reflects the result of the determination by the determination unit 53 in the medical image 31b of the detection target, thereby generating a post-detection medical image 32a and storing it in the image server 30.

[0022] The block information generation unit 51, acquisition unit 52, judgment unit 53, and image processing unit 54 constituting the anomaly detection system 50 are specifically realized by at least one computer having a non-volatile memory for storing programs, a volatile memory as a temporary working area, a processor for executing the programs, and input / output circuits including a communication interface and a communication port.

[0023] In addition, the block information generation unit 51, acquisition unit 52, judgment unit 53, and image processing unit 54 that constitute the anomaly detection system 50 may be realized by one computer or image processing device, or may be realized in a distributed manner by multiple computers or image processing devices connected by a communication path.

[0024] In addition, the diagnostic support system 10 is composed of an imaging device 20, an image server 30, a viewer device 40, and an abnormality detection system 50, which are connected by a communication path, but all or part of these components may be assembled into a single device.

[0025] Furthermore, the abnormality detection system 50 does not necessarily have to include the block information generation unit 51 and the image processing unit 54, but may include at least the acquisition unit 52 and the determination unit 53. Even with such a configuration, the presence or absence of an abnormality in an anatomical structure included in a medical image can be detected. Conversely, the abnormality detection system 50 may include at least one function of the imaging device 20, the image server 30, and the viewer device 40.

[0026] 2A to 2D are diagrams showing examples of anatomical structures (seven types of anatomical structures in total) in medical images (i.e., included in medical images). More specifically, FIG. 2A to FIG. 2D show, as examples of anatomical structures, right atrial shadow and left ventricular shadow, descending aorta shadow, right diaphragm dome shadow and left diaphragm dome shadow, right pulmonary outer shadow and left pulmonary outer shadow, respectively. In each table of FIG. 2A to FIG. 2D, from the left, a chest X-ray image, an anatomical structure name, an area of ​​the anatomical structure, and an example of superimposition of the chest X-ray image and the anatomical structure are shown.

[0027] Anatomical structures on chest X-ray images include a right atrial shadow, a left ventricular shadow, a descending aorta shadow, a diaphragm dome shadow, and the like, as shown in Figs. 2A to 2C. The right atrial shadow is a boundary line depicted on a chest X-ray image between the right atrium and its surroundings. The left ventricular shadow is a boundary line depicted on a chest X-ray image between the left ventricle and its surroundings. The descending aorta shadow is a boundary line depicted on a chest X-ray image between the descending aorta and its surroundings. The diaphragm dome shadow is a boundary line depicted on a chest X-ray image between the diaphragm and its surroundings.

[0028] The anatomical structures on the chest X-ray image are not limited to these. For example, as shown in FIG. 2D, the anatomical structures may be the shadow of the outer edge of the lung, which is the boundary line depicted on the border between the outer part of the lung and its surroundings on the chest X-ray image, or the shadow of other structures or organs. Furthermore, the anatomical structures are not limited to structures included in the chest X-ray image, but may be structures included in medical images captured by other imaging devices, such as CT images and MRI images.

[0029] Next, the operation of the diagnosis support system 10 according to this embodiment configured as above will be described, focusing on the operation of the characteristic abnormality detection system 50 constituting the diagnosis support system 10.

[0030] Fig. 3 is a flowchart showing the main operation (i.e., an anomaly detection method) of anomaly detection system 50 in Fig. 1. Fig. 3(a) shows the flow of the main operation of anomaly detection system 50, and Fig. 3(b) shows details of step S2 in Fig. 3(a).

[0031] First, the acquisition unit 52 acquires the block information generated by the block information generation unit 51 (acquisition step S1).

[0032] Next, the determination unit 53 determines whether or not the anatomical structure included in the medical image is abnormal based on the number of blocks indicated by the block information acquired by the acquisition unit 52, and outputs the result (determination step S2). More specifically, the determination unit 53 determines the number of blocks indicated by the block information (S21), and if the number of blocks is 1 (Yes in S21), it recognizes that the anatomical structure that should be detected as one block has been detected as one block, and determines that the anatomical structure included in the medical image is not abnormal (S22).

[0033] On the other hand, when the number of blocks indicated by the block information is two or more, the judgment unit 53 judges whether the blocks indicated by the block information satisfy a predetermined condition (S23). If the predetermined condition is satisfied (Yes in S23), it recognizes that an anatomical structure that should be detected as a single block and is not due to noise on the image or in image processing has been separated into two or more blocks, and judges that the anatomical structure included in the medical image is abnormal (S24). If the predetermined condition is not satisfied (No in S23), it recognizes that an anatomical structure that should be detected as a single block has been separated into two or more blocks due to noise on the image or in image processing, and judges that the anatomical structure included in the medical image is not abnormal (S22).

[0034] FIG. 4 is a flow chart showing a specific example of step S23 in FIG. 3 (determining whether or not a predetermined condition is satisfied).

[0035] In the judgment example shown in this figure, the judgment unit 53 judges whether or not a specified condition is satisfied by judging whether or not the block with the second largest area in the medical image among two or more blocks indicated by the block information has an area equal to or greater than a first threshold value (S23a).

[0036] As a result, if the second largest area block in the medical image has an area equal to or greater than the first threshold (Yes in S23a), the judgment unit 53 recognizes that the anatomical structure that should be detected as a single block is not noise in the image or image processing, but is instead separated into two or more blocks, and judges that the anatomical structure contained in the medical image is abnormal (go to S24 in Figure 3).

[0037] On the other hand, if the second largest area block in the medical image does not have an area equal to or greater than the first threshold (No in S23a), the judgment unit 53 recognizes that an anatomical structure that should be detected as a single block has been separated into two or more blocks due to noise in the image or image processing, and judges that the anatomical structure contained in the medical image is not abnormal (go to S22 in Figure 3).

[0038] FIG. 5 is a schematic diagram showing an example of an anatomical structure (here, the descending aorta shadow) in a medical image for explaining the example of the judgment results shown in FIG. 3 and FIG. 4. More specifically, FIG. 5(a) shows an example of the descending aorta shadow that is detected as one block 60 (Yes in step S21 in FIG. 3) and is judged to be "not abnormal". FIG. 5(b) shows an example of the descending aorta shadow that is detected as two blocks 60a and 60b (No in step S21 in FIG. 3) and is judged to be "abnormal" because the second largest block 60a has an area equal to or greater than the first threshold (Yes in step S23a in FIG. 4). FIG. 5(c) shows an example of the descending aorta shadow that is detected as two blocks 60c and 60d (No in step S21 in FIG. 3) and is judged to be "not abnormal" because the second largest block 60c does not have an area equal to or greater than the first threshold (No in step S23a in FIG. 4).

[0039] Fig. 6 is a diagram showing the results of an experiment using a case for evaluating the judgment shown in Fig. 4. More specifically, Fig. 6(a) shows an ROC curve (receiver operating characteristic curve) when the presence or absence of abnormality is judged with the right atrial shadow as an anatomical structure, Fig. 6(b) shows an ROC curve when the presence or absence of abnormality is judged with the left ventricular shadow as an anatomical structure, Fig. 6(c) shows an ROC curve when the presence or absence of abnormality is judged with the descending aorta shadow as an anatomical structure, and Fig. 6(d) shows the threshold selection results and AUC (Area Under the Curve) of the ROC curves shown in Fig. 6(a) to (c).

[0040] 6(a) to (c), the horizontal axis indicates the false positive rate (i.e., the probability (1-specificity) that the anomaly detection system 50 judges a case known to be normal to be abnormal), and the vertical axis indicates the positive rate (i.e., the probability (sensitivity) that the anomaly detection system 50 judges a case known to be abnormal to be abnormal). These ROC curves are plots of the judgment results by the anomaly detection system 50 obtained by changing the first threshold value used for judgment shown in FIG. 4 for multiple cases known to be normal and abnormal.

[0041] The closer the ROC curve is to the upper left corner, the more accurate the case is. According to the inventors' comparison, it has been found that the accuracy of the judgment for the cases related to the left ventricular shadow and the descending aorta shadow shown in (b) and (c) of FIG. 6 is improved compared to the judgment accuracy based on the conventional image processing. This is because the method of judging abnormality / normality based on the number of blocks extracted from the medical image and the size of the second largest area block for one anatomical structure makes it possible to make a correct judgment even in cases where it is difficult to judge by the conventional method (for example, a case where a normal structure of an organ is partially overlapped with other abnormalities such as a tumor or water in the medical image). It is to be noted that the first threshold value corresponding to the point closest to the upper left corner of the ROC curve (the black point in (a) to (c) of FIG. 6) is the most suitable threshold value to be used for the judgment shown in FIG. 4.

[0042] There are several methods for selecting the threshold value used in judgment on an ROC curve. For example, one method is to select the point on the ROC curve where the positive rate + (1 - false positive rate) value is the largest (the black points in (a) to (c) of Figure 6). Positive rate + (1 - false positive rate) is called the Youden Index, and this method selects the point where it is maximum. Other methods include selecting a point where the positive rate is a target value (e.g., 0.9) or where the false positive rate is a target value (e.g., 0.1).

[0043] FIG. 7 is a flowchart showing another specific example of step S23 in FIG. 3 (determining whether or not a predetermined condition is satisfied).

[0044] In the judgment example shown in this figure, the judgment unit 53 judges whether or not a predetermined condition is satisfied by judging whether or not the minimum value of the distance in the medical image between the block with the largest area in the medical image and each of the other blocks among two or more blocks indicated by the block information is less than or equal to a second threshold value (S23b).

[0045] As a result, if the minimum value of the distance in the medical image between the block with the largest area in the medical image and each of the other blocks is equal to or less than the second threshold value (Yes in S23b), the judgment unit 53 recognizes that the anatomical structure that should be detected as a single block is not noise in the image or image processing but is separated into two or more blocks, and judges that the anatomical structure included in the medical image is abnormal (go to S24 in Figure 3).

[0046] On the other hand, if the minimum value of the distance in the medical image between the block with the largest area in the medical image and each of the other blocks is not less than the second threshold value (No in S23b), the judgment unit 53 recognizes that an anatomical structure that should be detected as a single block has been separated into two or more blocks due to noise in the image or image processing, and judges that the anatomical structure contained in the medical image is not abnormal (go to S22 in Figure 3).

[0047] FIG. 8A is a schematic diagram showing an example of an anatomical structure (here, the descending aorta shadow) in a medical image for explaining the example of the judgment results shown in FIG. 3 and FIG. 7. More specifically, FIG. 8A (a) shows an example of the descending aorta shadow that is detected as one block 60 (Yes in step S21 in FIG. 3) and is judged to be "not abnormal". FIG. 8A (b) shows an example of the descending aorta shadow that is detected as two blocks 60c and 60d (No in step S21 in FIG. 3) and is judged to be "abnormal" because the minimum value (here, the distance between block 60d and block 60c) among the distances in the medical image between block 60d, which has the largest area in the medical image, and each of the other blocks (here, block 60c) is equal to or less than a second threshold (Yes in step S23b in FIG. 7). FIG. 8A (c) shows an example of a descending aorta shadow that is detected as two blocks 60e and 60f (No in step S21 of FIG. 3), but is determined to be “not abnormal” because the minimum distance in the medical image between block 60f, which has the largest area in the medical image, and each of the other blocks (here, block 60e) (here, the distance between block 60f and block 60e) is not less than the second threshold (No in step S23b of FIG. 7).

[0048] FIG. 8B is a diagram showing an actual imaging example of an anatomical structure (here, the descending aorta shadow) in a medical image for explaining the result example of the determination shown in FIG. 7. FIG. 8B shows an image of the descending aorta shadow actually extracted as a block. (a) to (c) of FIG. 8B show an example of the descending aorta shadow that is correctly determined to be "not abnormal" because the minimum value of the distance between the block with the largest area in the medical image (here, 70b, 70d, 70f) and each of the other blocks (here, 70a, 70c, 70e) in the medical image is not equal to or less than the second threshold (No in step S23b of FIG. 7), and (d) of FIG. 8B shows an example of the descending aorta shadow that is correctly determined to be "abnormal" because the minimum value of the distance between the block with the largest area in the medical image (here, 70h) and each of the other blocks (here, 70g, 70i) in the medical image is equal to or less than the second threshold (No in step S23b of FIG. 7).

[0049] Specifically, blocks 70a and 70b shown in FIG. 8B(a) are detection results of the descending aorta shadow, and show an example in which a discontinuity occurs but the shadow is determined to be "not abnormal". Blocks 70c and 70d shown in FIG. 8B(b) are detection results of the right diaphragm dome shadow, and show an example in which a discontinuity occurs but the shadow is determined to be "not abnormal". Blocks 70e and 70f shown in FIG. 8B(c) are detection results of the left ventricular shadow, and show an example in which a discontinuity occurs but the shadow is determined to be "not abnormal". Blocks 70g, 70h, and 70i shown in FIG. 8B(d) are detection results of another left ventricular shadow, and show an example in which a small discontinuity occurs and the shadow is determined to be "not abnormal".

[0050] Note that specific examples of step S23 in FIG. 3 (determination of whether or not a predetermined condition is satisfied) are not limited to the determination using area shown in FIG. 4 and the determination using distance shown in FIG. 7. Determination using both the determination using area and the determination using distance may also be used. FIG. 9 is a flowchart showing an example of a determination using both the determination using area and the determination using distance. In other words, FIG. 9 is a flowchart showing yet another specific example of step S23 in FIG. 3 (determination of whether or not a predetermined condition is satisfied).

[0051] In the judgment example shown in this figure, the judgment unit 53 first judges whether or not a predetermined condition is satisfied by determining whether or not the block with the second largest area in the medical image among two or more blocks indicated by the block information has an area equal to or greater than a first threshold value (S23a).

[0052] As a result, if the second largest area block in the medical image does not have an area equal to or greater than the first threshold (No in S23a), the judgment unit 53 recognizes that an anatomical structure that should be detected as a single block has been separated into two or more blocks due to noise in the image or image processing, and judges that the anatomical structure contained in the medical image is not abnormal (go to S22 in Figure 3).

[0053] On the other hand, if the block with the second largest area in the medical image has an area equal to or greater than the first threshold (Yes in S23a), the judgment unit 53 then judges whether the minimum value of the distance in the medical image between the block with the largest area in the medical image and each of the other blocks among the two or more blocks indicated by the block information is equal to or less than the second threshold (S23b).

[0054] As a result, if the minimum value of the distance in the medical image between the block with the largest area in the medical image and each of the other blocks is equal to or less than the second threshold value (Yes in S23b), the judgment unit 53 recognizes that the anatomical structure that should be detected as a single block is not noise in the image or image processing but is separated into two or more blocks, and judges that the anatomical structure included in the medical image is abnormal (go to S24 in Figure 3).

[0055] On the other hand, if the minimum value of the distance in the medical image between the block with the largest area in the medical image and each of the other blocks is not less than the second threshold value (No in S23b), the judgment unit 53 recognizes that an anatomical structure that should be detected as a single block has been separated into two or more blocks due to noise in the image or image processing, and judges that the anatomical structure contained in the medical image is not abnormal (go to S22 in Figure 3).

[0056] In the judgment shown in FIG. 9, an anatomical structure included in a medical image is judged to be abnormal if the condition regarding area is met (Yes in S23a) and the condition regarding distance is met (Yes in S23b), and is judged to be not abnormal in other cases. Alternatively, an anatomical structure included in a medical image may be judged to be abnormal if at least one of the condition regarding area (S23a) and the condition regarding distance (S23b) is met, and is judged to be not abnormal in other cases.

[0057] Furthermore, as a specific example of step S23 in FIG. 3 (determining whether or not a predetermined condition is satisfied), which of the determination examples shown in FIG. 4, FIG. 7, FIG. 9, etc. is to be adopted depends on the type of anatomical structure to be determined, and experiments may be performed on cases in which the abnormality / normality is known in advance, to determine the determination method with the highest accuracy rate for abnormality / normality determination.

[0058] Fig. 10A is a flowchart showing a method of determining a first threshold value used in the determination using the area shown in Fig. 4. In the determination using the area shown in Fig. 4, the determination unit 53 determines whether or not a predetermined condition is satisfied by determining whether or not the block with the second largest area in the medical image, among two or more blocks indicated by the block information, has an area equal to or larger than the first threshold value, and the method of determining the first threshold value used at that time is as follows.

[0059] First, for a plurality of cases that are known to be normal and abnormal, the first threshold value used for the judgment shown in Fig. 4 is changed and the abnormality detection system 50 is made to make a judgment, thereby calculating an ROC curve as shown in Fig. 6(a) to (c) (S30). Next, the first threshold value corresponding to the point closest to the upper left corner in the calculated ROC curve is determined as the threshold value used for the judgment shown in Fig. 4 (S31).

[0060] In other words, various first thresholds are used in advance to make provisional judgments, and the first threshold with the highest accuracy is adopted as the judgment threshold for an anatomical structure whose abnormality / normality is unknown. This allows the presence or absence of an abnormality in an anatomical structure to be detected with high accuracy.

[0061] Fig. 10B is a flowchart showing a method of determining the second threshold value used in the determination using the distance shown in Fig. 7. In the determination using the distance shown in Fig. 7, the determination unit 53 determines whether or not a predetermined condition is satisfied by determining whether or not the minimum value of the distances in the medical image between the block having the largest area in the medical image and each of the other blocks among two or more blocks indicated by the block information is equal to or less than the second threshold value, and the method of determining the second threshold value used at that time is as follows.

[0062] First, for a plurality of cases that are known to be normal and abnormal, the second threshold value used for the judgment shown in Fig. 7 is changed and the abnormality detection system 50 performs judgment to calculate an ROC curve as shown in Fig. 6(a) to (c) (S40). Next, the second threshold value corresponding to the point closest to the upper left corner on the calculated ROC curve is determined as the threshold value used for the judgment shown in Fig. 7 (S41).

[0063] In other words, various second thresholds are used in advance to make provisional judgments, and the second threshold with the highest accuracy is adopted as the judgment threshold for anatomical structures whose abnormality / normality is unknown. This allows the presence or absence of abnormality in anatomical structures to be detected with high accuracy.

[0064] The method of determining the first threshold shown in FIG. 10A and the method of determining the second threshold shown in FIG. 10B can be considered as an example of a determination step of acquiring an ROC curve for distinguishing between abnormalities and normalities in an anatomical structure, and determining predetermined conditions (here, the first threshold and the second threshold) based on the acquired ROC curve.

[0065] As described above, the abnormality detection system 50 constituting the diagnostic support system 10 according to this embodiment is a system that detects the presence or absence of an abnormality in a pre-specified anatomical structure contained in a medical image based on the medical image, and includes an acquisition unit 52 that acquires block information indicating blocks which are areas in the medical image in which pixels indicating an anatomical structure are connected, and a determination unit 53 that determines whether or not the anatomical structure is abnormal based on the number of blocks indicated by the block information acquired by the acquisition unit 52, and outputs the result. The determination unit 53 determines that the anatomical structure is not abnormal if the number of blocks indicated by the block information is 1, and determines that the anatomical structure is abnormal if the number of blocks indicated by the block information is 2 or more and the block indicated by the block information satisfies a predetermined condition.

[0066] This allows the presence or absence of an abnormality to be determined based on whether an anatomical structure that should be detected as a single block has been detected as a single block, making it possible to make a correct determination even in cases where it was difficult to make a determination using conventional methods (for example, a medical image in which a normal organ structure is partially overlapped with other abnormalities such as a tumor or water).As a result, the presence or absence of an abnormality in an anatomical structure can be detected with higher accuracy than conventional methods.

[0067] Also, as an example, the predetermined condition is a condition regarding the block having the second largest area in the medical image among the two or more blocks indicated by the block information. More specifically, the predetermined condition is that the block having the second largest area in the medical image among the two or more blocks indicated by the block information has an area equal to or greater than a first threshold.

[0068] As a result, as shown in the experimental results in (b) and (c) of FIG. 6, the determination accuracy of the determining unit 53 is improved compared to the determination accuracy based on conventional image processing.

[0069] The predetermined condition may be that the minimum value of the distances in the medical image between the block with the largest area in the medical image and each of the other blocks among the two or more blocks indicated by the block information is equal to or less than a second threshold value. Here, the predetermined condition may further be that the block with the second largest area in the medical image among the two or more blocks indicated by the block information has an area equal to or greater than a first threshold value.

[0070] This allows the abnormality / normality of an anatomical structure to be determined based on information about how an anatomical structure that should normally be detected as a single block is detected as multiple blocks, making it possible to make a correct judgment even in cases where this would have been difficult to do using conventional methods.

[0071] The first threshold value may be a value determined depending on ROC curves for distinguishing between abnormal and normal anatomical structures, which are obtained in advance for various areas, and the second threshold value may be a value determined depending on ROC curves for distinguishing between abnormal and normal anatomical structures, which are obtained in advance for various distances.

[0072] As a result, the first and second thresholds are determined from an ROC curve obtained for a case in which abnormality / normality has been determined in advance, so that the presence or absence of abnormality in an anatomical structure can be detected with high accuracy.

[0073] Furthermore, the abnormality detection method according to this embodiment is an abnormality detection method by an abnormality detection system 50 that detects the presence or absence of an abnormality in a pre-specified anatomical structure contained in a medical image based on the medical image, and includes an acquisition step S1 of acquiring block information indicating blocks which are areas in the medical image in which pixels indicating an anatomical structure are connected, and a determination step S2 of determining whether or not the anatomical structure is abnormal based on the number of blocks indicated by the block information acquired in the acquisition step S1, and outputting the result. In the determination step S2, if the number of blocks indicated by the block information is 1, it is determined that the anatomical structure is not abnormal, and if the number of blocks indicated by the block information is 2 or more, it is determined that the anatomical structure is abnormal when the block indicated by the block information satisfies a predetermined condition.

[0074] This allows the presence or absence of an abnormality to be determined based on whether an anatomical structure that should be detected as a single block has been detected as a single block, making it possible to make a correct determination even in cases where it was difficult to make a determination using conventional methods (for example, a medical image in which a normal organ structure is partially overlapped with other abnormalities such as a tumor or water).As a result, the presence or absence of an abnormality in an anatomical structure can be detected with higher accuracy than conventional methods.

[0075] Furthermore, the method may further include a determination step (S30 to S31, S40 to S41) of acquiring an ROC curve for distinguishing between abnormality and normality of an anatomical structure, and determining the predetermined condition based on the acquired ROC curve. In this way, the predetermined condition including the first threshold and the second threshold is determined from the ROC curve acquired for a case in which abnormality / normality has been known in advance, so that the presence or absence of abnormality in an anatomical structure can be detected with high accuracy.

[0076] The diagnosis support system, medical image anomaly detection system, and anomaly detection method of the present disclosure have been described above based on the embodiments, but the present disclosure is not limited to these embodiments. As long as they do not deviate from the gist of the present disclosure, various modifications conceived by a person skilled in the art to the present embodiments and other forms constructed by combining some of the components of the embodiments are also included within the scope of the present disclosure.

[0077] For example, the present disclosure may be realized as a program for executing the steps included in the anomaly detection method in the above-described embodiment, or as a computer-readable non-transitory recording medium such as a DVD on which the program is recorded. The program may be stored in one storage device, or may be distributed and stored in multiple storage devices. The program may be executed by one computer, or may be distributed and executed by multiple computers connected by a communication path.

[0078] In addition, the anomaly detection method in the above embodiment includes steps by the acquisition unit 52 and the judgment unit 53, but is not limited to these steps and may include at least one of processing by the block information generation unit 51, the image processing unit 54, and the shooting device 20, and processing by the viewer device 40.

[0079] In addition, in the above embodiment, multiple judgment examples are shown as specified conditions, but these multiple judgment examples may be determined by the abnormality detection system 50 based on instructions from a user, or the abnormality detection system 50 may automatically determine and update the judgment example that has had the highest accuracy rate in the past depending on the type of medical image or anatomical structure.

[0080] In addition, in the above embodiment, the first and second thresholds used for the specified condition are determined to be values ​​corresponding to the point closest to the upper left corner on the ROC curve, but they may be determined to be another point close to that point, or may be selected by the user from multiple points close to the upper left corner. [Industrial Applicability]

[0081] The medical image anomaly detection system and method disclosed herein can be used as a system for detecting the presence or absence of abnormalities in anatomical structures contained in medical images, for example, as a system to assist doctors in efficiently diagnosing diseases. [Explanation of symbols]

[0082] 10 Diagnostic Support System 20 Imaging Equipment 30 Image Server 31a, 31b, 32a, 32b Medical Imaging 40 Viewer device 50 Anomaly Detection System 51 Block information generation unit 52 Acquisition Department 53 Judgment section 54 Image processing section 60, 60a-60f, 70a-70f blocks

Claims

1. 1. An anomaly detection system for detecting the presence or absence of an abnormality in a pre-specified anatomical structure included in a medical image, the anatomical structure being essentially to be detected as a single block, comprising: The abnormalities include cases in which normal structures of organs are overlaid with tumors and other abnormalities, including water, in medical images; The anomaly detection system includes: an acquisition unit that acquires block information indicating a block that is a region in which pixels indicating the anatomical structure are connected in the medical image; a determination unit that determines whether or not an abnormality occurs due to the case based on the number of blocks indicated by the block information acquired by the acquisition unit, and outputs the determination result; the determination unit determines that the abnormality is not due to the case when the number of blocks indicated by the block information is 1, and determines that the abnormality is due to the case when the number of blocks indicated by the block information is 2 or more and the block indicated by the block information satisfies a predetermined condition. Anomaly detection systems.

2. the predetermined condition is a condition regarding a block having a second largest area in the medical image among two or more blocks indicated by the block information; The anomaly detection system of claim 1 .

3. The predetermined condition is that the block having the second largest area in the medical image among the two or more blocks indicated by the block information has an area equal to or larger than a first threshold value. The anomaly detection system according to claim 2 .

4. The predetermined condition is that the minimum value of the distances between the block having the largest area in the medical image and each of the other blocks among the two or more blocks indicated by the block information is equal to or less than a second threshold value. The anomaly detection system of claim 1 .

5. The predetermined condition is further that the block having the second largest area in the medical image among the two or more blocks indicated by the block information has an area equal to or larger than a first threshold value. The anomaly detection system according to claim 4 .

6. The first threshold value is determined based on a receiver operating characteristic curve (ROC curve) for distinguishing between abnormal and normal anatomical structures, the receiver operating characteristic curve being previously obtained for various areas. It is a value that is determined by The anomaly detection system according to claim 3 or 5.

7. The second threshold value is a value determined depending on a receiver operating characteristic (ROC) curve for distinguishing between abnormal and normal anatomical structures, the receiver operating characteristic being previously obtained for various distances. The anomaly detection system according to claim 4 .

8. 1. An anomaly detection method for detecting the presence or absence of an abnormality in a pre-specified anatomical structure included in a medical image, the anatomical structure being essentially detected as a single block, comprising: The abnormalities include cases in which normal structures of organs are overlaid with tumors and other abnormalities, including water, in medical images; The abnormality detection method includes: an acquisition step of acquiring block information indicating a block which is a region in which pixels indicating the anatomical structure are connected in the medical image; a determination step of determining whether or not the case is abnormal based on the number of blocks indicated by the block information acquired in the acquisition step, and outputting the result; In the determination step, when the number of blocks indicated by the block information is 1, it is determined that the abnormality is not due to the case, and when the number of blocks indicated by the block information is 2 or more, it is determined that the abnormality is due to the case when the block indicated by the block information satisfies a predetermined condition. Anomaly detection methods.

9. The method further includes a determination step of acquiring an ROC curve for distinguishing between an abnormality and a normality of the anatomical structure, and determining the predetermined condition based on the acquired ROC curve. The anomaly detection method according to claim 8.

10. The anatomical structure is the left ventricle, the right atrium, the descending aorta, or the diaphragm. The anomaly detection system according to any one of claims 1 to 7.

Images