JP7348469B2 - Abnormal shadow identification program, abnormal shadow identification device, and abnormal shadow identification method - Google Patents

Description

The present invention relates to an abnormal shadow identification program, an abnormal shadow identification apparatus, and an abnormal shadow identification method.

In recent years, in image diagnosis work performed by doctors, images of cases similar to the case being diagnosed (hereinafter also referred to as similar case images) may be referred to. Therefore, in recent years, technology related to similar case image retrieval (hereinafter also simply referred to as image retrieval) for automatically retrieving such similar case images has been developed. Specifically, in this kind of image search, for example, images that can be judged to be similar to the case being diagnosed are searched from a database that has accumulated images corresponding to each case acquired in the past, and then the search is performed. Display the results on an output device, etc.

This makes it possible for doctors who perform image diagnosis work to efficiently refer to search results obtained by image searches (for example, see Patent Documents 1 to 3).

Japanese Patent Application Publication No. 2007-275216 JP2007-286945A Japanese Patent Application Publication No. 2015-191285

When the image search described above is performed, the system that performs the image search (hereinafter also referred to as an information processing system), for example, extracts abnormal shadows from the image corresponding to the case being diagnosed, and Identify the features that indicate the distribution. The information processing system then identifies images that can be determined to have similar feature amounts to images corresponding to the case being diagnosed from similar case images acquired in the past, and outputs the identified images as search results to a device. etc.

However, when extracting abnormal shadows from images corresponding to cases in which it is difficult to extract abnormal shadows, such as diffuse lung disease, the information processing system may detect abnormal shadows (e.g., blood vessels or parts of the chest wall). There is a possibility that the image may be mistakenly extracted as a honeycomb pattern (one of the types of abnormal shadows), and this may be a major cause of lowering the search accuracy of image search.

Therefore, in the image search described above, even when extracting abnormal shadows from images corresponding to cases such as diffuse lung disease, it is necessary to extract abnormal shadows with high precision. ing.

Therefore, in one aspect, an object of the present invention is to provide an abnormal shadow identification program, an abnormal shadow identification device, and an abnormal shadow identification method that enable highly accurate extraction of abnormal shadows included in an image. do.

In one aspect of the embodiment, abnormal shadow candidates are extracted from an image inside the body, and among the extracted candidates, a candidate whose specific value based on a detected position in the body satisfies a predetermined condition is determined as the abnormal shadow. Among the identified and extracted candidates, a candidate detected at a position corresponding to the identified abnormal shadow and having a predetermined similarity relationship with the identified abnormal shadow is further identified as the abnormal shadow. , cause the computer to execute the process.

According to one aspect, it is possible to extract abnormal shadows included in an image with high precision.

FIG. 1 is a diagram illustrating the configuration of an information processing system 10. FIG. 2 is a diagram illustrating a specific example of the process of specifying the feature amount of the image data 131, and both (A) and (B) are diagrams showing an axial cross section of the lung. FIG. 3 is a diagram illustrating a specific example of the process of identifying the feature amount of the image data 131. FIG. 4 is a diagram illustrating a specific example of the abnormal shadow SH extracted from the image data 131, and (A), (B), and (C) are diagrams each showing an axial cross section of the lung. FIG. 5 is a diagram illustrating the hardware configuration of the information processing device 1. FIG. 6 is a block diagram of the functions of the information processing device 1. FIG. 7 is a flowchart illustrating an outline of abnormal shadow identification processing in the first embodiment. FIG. 8 is a diagram illustrating an outline of abnormal shadow identification processing in the first embodiment. FIG. 9 is a diagram illustrating an outline of abnormal shadow identification processing in the first embodiment. FIG. 10 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 11 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 12 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 13 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 14 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 15 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 16 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 17 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 18 is a flowchart illustrating details of abnormal shadow identification processing in the first embodiment. FIG. 19 is a diagram illustrating a specific example of probability information 132. FIG. 20 is a diagram illustrating a specific example of the coefficient information 133. FIG. 21 is a diagram illustrating a specific example of the coefficient information 133. FIG. 22 is a diagram illustrating a specific example of the coefficient information 133. FIG. 23 is a diagram illustrating a specific example of the threshold information 134. FIG. 24 is a diagram illustrating a specific example of information output in the process of S145, and (A), (B), and (C) are diagrams each showing a coronal cross section of the lung. FIG. 25 is a diagram illustrating the effect of the abnormal shadow identification process in the first embodiment, and is a diagram showing an axial cross section of the lung.

[Information processing system configuration]
First, the configuration of the information processing system 10 will be explained. FIG. 1 is a diagram illustrating the configuration of an information processing system 10.

As shown in FIG. 1, the information processing system 10 includes an information processing device 1, an operating terminal 2 operated by a doctor, and a storage unit 130 provided inside or outside the information processing device 1. The information processing device 1 and the operating terminal 2 access each other via the network NW.

The operating terminal 2 receives, for example, input of image data 131 (hereinafter also referred to as new image data 131) regarding a case under diagnosis. Further, the operation terminal 2 receives, for example, an input to perform an image search for image data 131 (similar case images) similar to the new image data 131 that has been input. Then, the operation terminal 2 transmits, for example, an instruction to the information processing device 1 to perform an image search for image data 131 similar to the new image data 131 that has been input.

For example, when receiving an instruction to perform an image search from the operating terminal 2, the information processing device 1 performs an image search of the image data 131 corresponding to the received instruction. Then, the information processing device 1 transmits the image data 131 identified by performing the image search to the operation terminal 2. After that, the operation terminal 2 outputs the image data 131 received from the information processing device 1 to the output device, for example.

When performing such an image search, the information processing device 1, for example, extracts abnormal shadows from new image data 131 corresponding to the case being diagnosed, and identifies feature amounts indicating the distribution of the extracted abnormal shadows. . Then, the information processing device 1 identifies image data 131 that can be determined to have similar feature amounts to the image corresponding to the case under diagnosis from the image data 131 corresponding to each case acquired in the past, and identifies The resulting image data 131 is output to an output device. A specific example of the process of identifying the feature amount of new image data 131 will be described below. Note that the following description will be made assuming that the image data 131 corresponding to each case includes a plurality of slice image data 131a.

[Processing to identify features of new image data]
FIG. 2 is a diagram illustrating a specific example of the process of identifying the feature amount of new image data 131, and both (A) and (B) are diagrams showing axial cross sections of the lung. FIG. 2(A) is a specific example of one slice image data 131a included in new image data 131 regarding the lung LU under diagnosis, and FIG. This is a specific example of an abnormal shadow SH extracted from the image data 131a.

Further, FIG. 3 is a diagram illustrating a specific example of the process of identifying the feature amount of the new image data 131. FIG. 3(A) is a histogram showing the occurrence status of abnormal shadow SH extracted from the slice image data 131a shown in FIG. 2(A), and FIG. This is a histogram showing the occurrence status of all abnormal shadows SH included in the image data 131.

Specifically, the information processing device 1 uses, for example, a classifier (for example, a support vector machine) generated in advance by machine learning to determine the grid-like shape included in the slice image data 131a shown in FIG. 2(A). It is determined whether each of the partial images is classified as an abnormal shadow SH. Then, as shown in FIG. 2(B), the information processing device 1 selects a partial image classified as an abnormal shadow SH from among the grid-shaped partial images included in the slice image data 131a shown in FIG. 2(A). Identify.

Furthermore, as shown in FIG. 3A, the information processing device 1 detects abnormal shadows SH in each of the peripheral part of the left lung, the central part of the left lung, the central part of the right lung, and the peripheral part of the right lung, for example. The number of partial images classified as .

Specifically, in the histogram shown in FIG. 3(A), the horizontal axis corresponds to each region (central and peripheral) in the lung LU, and the vertical axis corresponds to the number of partial images classified as abnormal shadows SH. ing. Therefore, the histogram shown in FIG. 3(A) shows that the region with the largest number of partial images classified as abnormal shadow SH is the peripheral region of the left lung.

Thereafter, the information processing apparatus 1 performs the same processing as the aggregation described in FIG. 3A for each piece of slice image data 131a included in the new image data 131 for the lung LU under diagnosis. Then, as shown in FIG. 3B, the information processing device 1 classifies each of the peripheral part of the left lung, the central part of the left lung, the central part of the right lung, and the peripheral part of the right lung as an abnormal shadow SH. Collect the number of partial images.

Specifically, in each histogram corresponding to each region of the lung shown in FIG. 3(B), the horizontal axis corresponds to the number of partial images classified as abnormal shadows SH in each slice image data 131a, and the vertical axis corresponds to the number of partial images classified as abnormal shadows SH. corresponds to each piece of slice image data 131a included in the new image data 131 for the lung LU under diagnosis (that is, each part of the lung LU in the vertical direction). Therefore, the histogram corresponding to the peripheral part of the left lung (the first histogram from the left) and the histogram corresponding to the peripheral part of the right lung (the fourth histogram from the left) shown in FIG. 3(B) each show an abnormal shadow. It is shown that the region with the largest number of partial images classified as SH is the lower lung field. On the other hand, the histogram corresponding to the central part of the left lung (second histogram from the left) and the histogram corresponding to the central part of the right lung (third histogram from the left) shown in FIG. 3(B) each show an abnormal shadow. This shows that the region with the largest number of partial images classified as SH is the middle lung field.

This makes it possible for the information processing device 1 to specify, for example, the shape of each histogram shown in FIG. 3(B) as the feature amount of the image data 131 regarding the lung LU under diagnosis. Therefore, the information processing device 1 compares the feature amount specified for the new image data 131 with the feature amount specified for the image data 131 (image data 131 corresponding to a past case) stored in the storage unit 130. This makes it possible to extract image data 131 (similar case images) that is similar to the image data 131 regarding the case under diagnosis.

However, when extracting an abnormal shadow SH from the image data 131 corresponding to a case in which it is difficult to extract an abnormal shadow SH, such as a diffuse lung disease, the information processing device 1 may detect blood vessels, a part of the chest wall, etc. There is a possibility that it may be mistakenly extracted as an abnormal shadow SH, and this may be a major cause of reducing the search accuracy of image search. A specific example of the abnormal shadow SH extracted from the image data 131 will be described below.

[Specific examples of abnormal shadows extracted from image data]
FIG. 4 is a diagram illustrating a specific example of the abnormal shadow SH extracted from the image data 131, and (A), (B), and (C) are diagrams each showing an axial cross section of the lung. Note that the lung LUa and the lung LUb shown in FIG. 4 correspond to two different cross sections of the lung LU explained in FIG. 2.

For example, as shown in FIG. 4A, the slice image data 131a corresponding to the lung LUa includes an abnormal shadow SHa, and the slice image data 131a corresponding to the lung LUb includes an abnormal shadow SHb. If so, it is preferable that the information processing device 1 extracts only the abnormal shadow SHa and the abnormal shadow SHb.

However, depending on the criteria for determining the abnormal shadow SH, the information processing device 1 may extract not only the abnormal shadow SHa and the abnormal shadow SHb but also a blood vessel, a part of the chest wall, etc. as the abnormal shadow SHc, as shown in FIG. 4(B). There are cases where

On the other hand, for example, if the criteria for determining the abnormal shadow SH is set more strictly than in the case of FIG. 4(B), the information processing device 1 extracts only the abnormal shadow SHb, as shown in FIG. 4(C). , the abnormal shadow SHa may not be extracted.

Therefore, in image retrieval by the information processing device 1, even when extracting abnormal shadows SH from images corresponding to cases such as diffuse lung disease, it is required to extract abnormal shadows SH with high accuracy. It is being

Therefore, the information processing device 1 in this embodiment extracts candidates for abnormal shadows SH from the image data 131. Then, the information processing apparatus 1 identifies, as the abnormal shadow SH, a candidate whose specific value based on the detected position in the body satisfies a predetermined condition among the extracted candidates for the abnormal shadow SH.

Thereafter, the information processing device 1 selects candidates detected at a position corresponding to the identified abnormal shadow SH from among the extracted candidates for the abnormal shadow SH and that have a predetermined similarity relationship with the identified abnormal shadow SH. , further specified as an abnormal shadow SH.

That is, the information processing device 1 in the present embodiment not only identifies a candidate for an abnormal shadow SH that is determined to have a high probability of being an abnormal shadow SH as an abnormal shadow SH, but also identifies a candidate as an abnormal shadow SH. A candidate detected at a position corresponding to the determined candidate and similar to the candidate determined to be the abnormal shadow SH is also specified as the abnormal shadow SH.

As a result, the information processing device 1 can highly identify the abnormal shadow SH included in the image data 131 even when extracting the abnormal shadow SH from the image data 131 corresponding to a case such as diffuse lung disease. It becomes possible to do it with precision. Therefore, the information processing device 1 can prevent a decrease in search accuracy of image search.

[Hardware configuration of information processing system]
Next, the hardware configuration of the information processing system 10 will be explained. FIG. 5 is a diagram illustrating the hardware configuration of the information processing device 1.

As shown in FIG. 5, the information processing device 1 includes a CPU 101 that is a processor, a memory 102, an external interface (I/O unit) 103, and a storage medium 104. Each part is connected to each other via a bus 105.

The storage medium 104 has, for example, a program storage area (not shown) that stores a program 110 for performing processing for identifying the abnormal shadow SH (hereinafter also referred to as abnormal shadow identification processing). Furthermore, the storage medium 104 includes a storage unit 130 (hereinafter also referred to as an information storage area 130) that stores information used when performing abnormal shadow identification processing, for example. Note that the storage medium 104 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD).

The CPU 101 executes the program 110 loaded into the memory 102 from the storage medium 104 to perform abnormal shadow identification processing.

Further, the external interface 103 communicates with the operating terminal 2, for example, via the network NW.

[Functions of information processing system]
Next, the functions of the information processing system 10 will be explained. FIG. 6 is a block diagram of the functions of the information processing device 1.

As shown in FIG. 6, the information processing device 1 has an information management section 111 and a candidate extraction section 112 through organic cooperation between hardware such as the CPU 101 and memory 102 of the information processing device 1 and a program 110. Various functions including a first identifying section 113, a second identifying section 114, and an information output section 115 are realized.

Further, the information processing device 1 stores image data 131, probability information 132, coefficient information 133, and threshold information 134 in the information storage area 130, as shown in FIG.

For example, when various information such as new image data 131 is input via the operating terminal 2, the information management unit 111 stores the input various information in the information storage area 130.

The candidate extraction unit 112 extracts candidates for the abnormal shadow SH from new image data 131 input via the operation terminal 2, for example.

Specifically, the candidate extraction unit 112 generates, for example, by learning a plurality of learning data for the image data 131 including the abnormal shadow SH and a plurality of learning data for the image data 131 not including the abnormal shadow SH. By using a classifier (not shown), candidates for the abnormal shadow SH are extracted from the new image data 131.

The first specifying unit 113 specifies, as the abnormal shadow SH, a candidate whose specific value based on the position detected in the body satisfies a predetermined condition among the candidates for the abnormal shadow SH extracted by the candidate extracting unit 112.

Specifically, the first specifying unit 113 determines, for each candidate for the abnormal shadow SH extracted by the candidate extracting unit 112, a value indicating the likelihood of occurrence of the abnormal shadow SH at the position where each candidate is detected in the body. , and a value indicating the probability that each candidate is an abnormal shadow SH. Then, for example, the first specifying unit 113 specifies, as the abnormal shadow SH, a candidate whose calculated specific value is larger than the first threshold value, among the candidates for the abnormal shadow SH extracted by the candidate extracting unit 112.

The second specifying unit 114 selects a candidate detected at a position corresponding to the abnormal shadow SH specified by the first specifying unit 113 from among the candidates for the abnormal shadow SH extracted by the candidate extracting unit 112. A candidate having a predetermined similarity relationship with the abnormal shadow SH identified in step 113 is further specified as the abnormal shadow SH.

Specifically, the second specifying unit 114 selects, for each abnormal shadow SH candidate extracted by the candidate extracting unit 112, one of the abnormal shadow SH candidates extracted by the candidate extracting unit 112 that is located adjacent to the abnormal shadow SH in the body. Identify other candidates. Then, for example, for each candidate extracted by the candidate extraction unit 112 and for each other candidate, the second identification unit 114 distinguishes between information included in the candidate extracted by the candidate extraction unit 112 and information included in the other candidates. A value indicating the degree of similarity is calculated. Thereafter, the second specifying unit 114 selects each candidate, for example, among the combinations of the abnormal shadow SH candidate extracted by the candidate extracting unit 112 and other candidates, whose calculated similarity exceeds the second threshold. Concatenate. Then, the second identifying unit 114 identifies, as the abnormal shadow SH, a candidate that is connected to the abnormal shadow SH candidate identified by the first identifying unit 113, for example, among the abnormal shadow SH candidates extracted by the candidate extracting unit 112. do.

The information output unit 115 outputs information indicating the abnormal shadow SH identified by the first identifying unit 113 and the second identifying unit 114 to an output device of the operation terminal 2 or the like.

[Outline of first embodiment]
Next, an outline of the first embodiment will be described. FIG. 7 is a flowchart illustrating an outline of abnormal shadow identification processing in the first embodiment. Further, FIGS. 8 and 9 are diagrams illustrating an outline of abnormal shadow identification processing in the first embodiment.

As shown in FIG. 7, the information processing device 1 waits until the abnormal shadow identification timing (NO in S1). The abnormal shadow identification timing may be, for example, the timing at which the doctor inputs an instruction to identify the abnormal shadow SH via the operation terminal 2.

Then, when the abnormal shadow identification timing has come (YES in S1), the information processing device 1 extracts candidates for the abnormal shadow SH from the image data 131 about the inside of the body, as shown in FIGS. 7 and 8 (S2). .

Next, as shown in FIGS. 7 and 9, the information processing device 1 selects candidates for which a specific value based on the position detected in the body satisfies a predetermined condition among the candidates for the abnormal shadow SH extracted in the process of S2. is identified as an abnormal shadow SH (S3).

Thereafter, as shown in FIGS. 7 and 9, the information processing apparatus 1 selects a candidate detected at a position corresponding to the abnormal shadow SH identified in the process of S3, among the candidates for the abnormal shadow SH extracted in the process of S2. A candidate having a predetermined similarity relationship with the abnormal shadow SH identified in the process of S3 is further identified as the abnormal shadow SH (S4).

This allows the information processing device 1 to identify the abnormal shadow SH included in the image data 131 with high accuracy. Therefore, the information processing device 1 can prevent a decrease in search accuracy of image search.

[Details of the first embodiment]
Next, details of the first embodiment will be described. 10 to 18 are flowcharts illustrating details of the abnormal shadow identification process in the first embodiment. Further, FIGS. 19 to 25 are diagrams illustrating details of the abnormal shadow identification process in the first embodiment. Note that the following description will be made assuming that the abnormal shadow SH to be identified in the abnormal shadow identification process is a honeycomb shadow SH.

The candidate extraction unit 112 of the information processing device 1 waits until the abnormal shadow identification timing, as shown in FIG. 10 (NO in S11).

Then, when the abnormal shadow identification timing has come (YES in S11), the candidate extracting unit 112 uses, for example, a pre-generated classifier to extract the information contained in the new image data 131 input via the operation terminal 2. A candidate for the honeycomb shadow SH is identified (S12). The details of the process in S12 will be explained below.

[Details of processing in S12]
FIGS. 12 and 13 are diagrams illustrating details of the process of S12.

The candidate extraction unit 112 acquires new image data 131 regarding the case being diagnosed (S101). Specifically, the candidate extracting unit 112 acquires, for example, new image data 131 regarding a case under diagnosis, which includes a plurality of slice image data 131a.

Then, the candidate extraction unit 112 acquires one of the slice image data 131a included in the image data 131 acquired in the process of S101 (S102).

After that, the candidate extraction unit 112 divides the slice image data 131a acquired in the process of S102 into grid-like partial images (S103).

Subsequently, the candidate extraction unit 112 identifies one of the partial images divided in the process of S103 (S104).

Then, the candidate extracting unit 112 acquires feature amounts from the partial image identified in the process of S104 (S105). Specifically, the candidate extraction unit 112 acquires feature amounts indicating the contents of various information included in the partial image identified in the process of S104.

Next, as shown in FIG. 13, the candidate extraction unit 112 acquires information output by inputting the feature amount acquired in the process of S105 to a classifier generated in advance (S111).

Specifically, the classifier here classifies the partial image obtained in the process of S104 as "normal" or "honeycomb" based on the feature amount obtained in the process of S105 (the feature amount input to the classifier). The abnormal shadow is classified into one of "an abnormal shadow", "another abnormal shadow A", and "another abnormal shadow B". Then, in the process of S111, the candidate extraction unit 112 obtains, for example, the classification result output from the classifier and the identification value that is a value indicating the certainty of the classification result.

Then, the candidate extraction unit 112 associates the information obtained in the process of S111 with the partial image obtained in the process of S104 (S112). Specifically, the candidate extraction unit 112 associates the partial image obtained in the process of S104 with the classification result and identification value obtained in the process of S111.

After that, if all partial images have been identified in the process of S104 and all the slice image data 131a have been acquired in the process of S102 (YES in S113, YES in S114), candidate extraction The unit 112 ends the process of S12.

On the other hand, if all partial images have not been identified in the process of S104 (NO in S113), the candidate extracting unit 112 performs the processes from S104 onwards again. Furthermore, if acquisition of all the slice image data 131a is not completed in the process of S102 (NO in S114), the candidate extracting unit 112 performs the process from S102 onwards again.

This makes it possible for the candidate extraction unit 112 to associate classification results and identification values with all partial images included in the image data 131. Therefore, the candidate extracting unit 112 can identify a partial image that may be a honeycomb shadow SH (a partial image that is a candidate for a honeycomb shadow SH).

Returning to FIG. 10, the first identifying unit 113 of the information processing device 1 determines the likelihood of occurrence of the honeycomb shadow SH at the position where each candidate is detected in the body, for each candidate for the honeycomb shadow SH identified in the process of S12. A specific value is calculated from the value indicating the probability and the value indicating the probability of each candidate as the honeycomb shadow SH (S13).

Then, the first identifying unit 113 identifies, as the honeycomb shadow SH, a candidate for which the specific value calculated in the process of S13 is larger than the first threshold value, among the candidates for the honeycomb shadow SH identified in the process of S12. (S14). The first threshold may be, for example, a threshold determined in advance and stored in the information storage area 130. The details of the processing in S13 and S14 will be described below.

[Details of processing in S13 and S14]
FIGS. 14 and 15 are diagrams illustrating details of the processing in S13 and S14.

As shown in FIG. 14, the first specifying unit 113 specifies one of the partial images divided in the process of S103 (S121).

Then, the first identifying unit 113 determines whether the classification result associated with the partial image identified in the process of S121 is a honeycomb shadow SH (S122). Specifically, the first identifying unit 113 determines whether or not the classification result associated with the partial image identified in the process of S121 in the process of S112 is a honeycomb shadow SH.

As a result, if the classification result is a honeycomb shadow SH (YES in S122), the first identifying unit 113 refers to the probability information 132 stored in the information storage area 130, and identifies the partial image identified in the process of S121. A first probability of occurrence of the honeycomb shadow SH based on the position in the vertical direction is specified (S123). The probability information 132 is information indicating the probability of occurrence of a honeycomb shadow SH at each position in the lung.

Further, in this case, the first specifying unit 113 refers to the probability information 132 stored in the information storage area 130 and determines the second occurrence of the honeycomb shadow SH based on the horizontal position of the partial image specified in the process of S121. The probability is specified (S124). A specific example of the probability information 132 will be described below.

[Specific example of probability information]
FIG. 19 is a diagram illustrating a specific example of probability information 132.

The probability information 132 shown in FIG. 19 includes an "item number" in which information for identifying each piece of information included in the probability information 132 is stored, a "position" in which the name of each position in the lung is stored, and each position in the lung. It has an item "probability of occurrence" in which the probability of occurrence of honeycomb shadow SH in is stored.

Specifically, in the example shown in FIG. 19, for the information whose "item number" is "1", "upper lung field" is stored as "position" and "0.1" is stored as "probability of occurrence". has been done. Furthermore, in the example shown in FIG. 19, for the information whose "item number" is "2", "middle lung field" is stored as "position" and "0.3" is stored as "probability of occurrence". There is. Furthermore, in the example shown in FIG. 19, in the information whose "item number" is "3", "lower lung field" is stored as "position" and "1.0" is stored as "probability of occurrence". There is.

Furthermore, in the example shown in FIG. 19, in the information whose "item number" is "4", "center" is stored as the "position" and "0.1" is stored as the "probability of occurrence". Further, in the example shown in FIG. 19, in the information whose "item number" is "5", "middle" is stored as the "position" and "0.5" is stored as the "probability of occurrence". Further, in the example shown in FIG. 19, in the information whose "item number" is "6", "peripheral" is stored as the "position" and "1.0" is stored as the "probability of occurrence".

Therefore, for example, if the vertical position of the partial image identified in the process of S121 is the upper lung field, the first identifying unit 113 refers to the probability information 132 shown in FIG. The information "0.1" stored in the "probability of occurrence" of the information in which "field" is stored (information whose "item number" is "1") is specified as the first probability of occurrence.

Similarly, for example, if the horizontal position of the partial image identified in the process of S121 is intermediate, the first identifying unit 113 refers to the probability information 132 shown in FIG. 19 and stores "intermediate" in the "position". "0.5", which is the information stored in the "probability of occurrence" of the information (information whose "item number" is "5"), is specified as the second probability of occurrence.

Returning to FIG. 14, the first identifying unit 113 identifies the identification value associated with the partial image identified in the process of S121 (S125). Specifically, the first identifying unit 113 identifies the associated identification value in the process of S112 for the partial image identified in the process of S121.

Then, as shown in FIG. 15, the first specifying unit 113 determines the first probability of occurrence specified in the process of S123, the second probability of occurrence specified in the process of S124, and the identification value specified in the process of S125. A specific value is calculated based on this (S131). Specifically, the first specifying unit 113 calculates, for example, a value obtained by multiplying the first probability of occurrence specified in the process of S123, the second probability of occurrence specified in the process of S124, and the identification value specified in the process of S125. Calculate as a specific value.

Subsequently, the first specifying unit 113 compares the specific value calculated in the process of S131 and the first threshold value (S132).

As a result, if the specific value calculated in the process of S131 is larger than the first threshold value (YES in S132), the first identifying unit 113 determines that the partial image acquired in the process of S121 has a honeycomb shadow SH. Information indicating that the probability is high is associated (S133).

On the other hand, if the specific value calculated in the process of S131 is not larger than the first threshold (NO in S132), the first identifying unit 113 does not perform the process of S133.

If all partial images have been identified in the process of S121 (YES in S134), the first identifying unit 113 ends the processes of S13 and S14.

On the other hand, if the identification of all partial images is not completed in the process of S121 (NO in S134), the first identifying unit 113 performs the processes from S121 onwards again.

Note that in the process of S122, even if the classification result is not a honeycomb shadow SH (NO in S122), the first identifying unit 113 similarly performs the process of S134.

As a result, the first identifying unit 113 determines, among the partial images divided in the process of S103, that the partial image can be determined to be a honeycomb shadow SH with a high probability (information indicating that there is a high probability that the honeycomb shadow SH is a honeycomb shadow SH). This makes it possible to specify the partial image (with which the partial image is associated).

Returning to FIG. 10, the second identifying unit 114 of the information processing device 1 selects, for each candidate for the honeycomb shadow SH identified in the process of S12, among the candidates for the honeycomb shadow SH identified in the process of S12, which are adjacent to each other in the body. Then, other candidates are identified (S15).

Then, as shown in FIG. 11, the second identifying unit 114 performs the extraction in the process of S12 for each candidate for the honeycomb shadow SH extracted in the process of S12 and for each other candidate identified in the process of S15. A value indicating the degree of similarity between the information included in the candidate for the honeycomb shadow SH and the information included in the other candidates identified in the process of S15 is calculated (S21).

Next, the second identifying unit 114 determines the positional relationship in the body of each candidate and the positional relationship of each candidate in the body for each candidate for the honeycomb SH identified in the process of S12 and for each other candidate identified in the process of S15. A second threshold value is determined based on the comparison result of the probability as a honeycomb shadow SH (S22).

Furthermore, the second specifying unit 114 determines which of the combinations of the honeycomb shadow SH candidate identified in the process of S12 and the other candidates specified in S15, the similarity calculated in the process of S21 corresponds to each combination. The candidates for the honeycomb shadow SH included in the combinations exceeding the threshold of 2 are connected (S23).

Then, the second specifying unit 114 further specifies, as the honeycomb shadow SH, a candidate that is connected to the candidate identified as the honeycomb shadow SH in the process of S14, among the candidates for the honeycomb shadow SH identified in the process of S12. (S24). The details of the processing from S15 to S24 will be explained below.

[Details of processing from S15 to S24]
FIGS. 16 to 18 are diagrams illustrating details of the processing from S15 to S24.

As shown in FIG. 16, the second specifying unit 114 specifies one of the partial images divided in the process of S103 (S141).

Then, the second identifying unit 114 determines whether the classification result associated with the partial image identified in the process of S141 is a honeycomb shadow SH (S142). Specifically, the second specifying unit 114 determines whether or not the classification result associated with the partial image in the process of S112 is a honeycomb shadow SH.

As a result, if the classification result is a honeycomb shadow SH (YES in S142), the second specifying unit 114 performs a partial image connection determination process for each partial image adjacent to the partial image identified in the process in S141. (S143).

On the other hand, if the classification result is not a honeycomb shadow SH (NO in S142), the second identifying unit 114 does not perform the process in S143. The details of the process of S143 (connection determination process) will be explained below.

[Details of processing in S143]
FIGS. 17 and 18 are diagrams illustrating details of the process of S143.

The second identifying unit 114 identifies one of the partial images adjacent to the partial image identified in the process of S141 (S151). Specifically, the second identifying unit 114 selects a partial image that is horizontally or vertically adjacent to the partial image identified in the process of S141 (a partial image included in the slice image data 131a that is different from the partial image identified in the process of S141). ).

Then, the second identifying unit 114 determines whether the classification result associated with the partial image identified in the process of S151 is a honeycomb shadow SH (S152). Specifically, the second specifying unit 114 determines whether the classification result associated with the partial image in the process of S112 is a honeycomb shadow SH or not.

As a result, if the classification result associated with the partial image identified in the process of S151 is a honeycomb shadow SH (YES in S152), the second identifying unit 114 determines whether the partial image included in the partial image identified in the process of S141 is The degree of similarity between the information and the information included in the partial image identified in the process of S151 is calculated (S153).

Next, the second specifying unit 114 refers to the coefficient information 133 stored in the information storage area 130 and determines the vertical positional relationship between the partial image specified in the process of S141 and the partial image specified in the process of S151. The first coefficient is specified based on the first coefficient (S154). The coefficient information 133 is information including each coefficient used when determining the second threshold value to be compared with the similarity calculated in the process of S153.

Further, the second specifying unit 114 refers to the coefficient information 133 stored in the information storage area 130, and based on the horizontal positional relationship between the partial image specified in the process of S141 and the partial image specified in the process of S151. The second coefficient is specified (S155).

Furthermore, as shown in FIG. 18, the second specifying unit 114 refers to the coefficient information 133 stored in the information storage area 130, and selects the identification value associated with the partial image specified in the process of S141 and the process of S151. A third coefficient is specified based on the comparison result with the identification value associated with the partial image specified in (S161). A specific example of the coefficient information 133 will be described below. Note that hereinafter, the coefficient information 133 referred to in the process of S154 is also referred to as first coefficient information 133a, and the coefficient information 133 referred to in the process of S155 is also referred to as second coefficient information 133b, and the coefficient information 133 referred to in the process of S161 is also referred to as first coefficient information 133a. The information 133 is also called third coefficient information 133c.

[Specific example of coefficient information]
20 to 22 are diagrams illustrating specific examples of the coefficient information 133. Specifically, FIG. 20 is a diagram for explaining a specific example of the first coefficient information 133a, FIG. 21 is a diagram for explaining a specific example of the second coefficient information 133b, and FIG. 22 is a diagram for explaining a specific example of the second coefficient information 133b. 133c is a diagram illustrating a specific example.

[Specific example of first coefficient information]
First, a specific example of the first coefficient information 133a will be explained.

The first coefficient information 133a shown in FIG. 20 includes an "item number" in which information identifying each piece of information included in the first coefficient information 133a is stored, and the position of the partial image identified in the process of S141 is determined in the process of S151. The items include a "relational expression" that stores an equation indicating whether the partial image is located below the position of the specified partial image, and a "coefficient" that stores a coefficient used in the process of S154. The "relational expression" includes "a>b" indicating that the position of the partial image identified in the process of S141 is located below the position of the partial image identified in the process of S151, and "a=b" indicating that the vertical position of the partial image identified in S151 is the same as the vertical position of the partial image identified in S151, or the position of the partial image identified in S141 is S151 "a<b" is set indicating that the partial image is located above the position of the partial image identified in the process.

Specifically, in the first coefficient information 133a shown in FIG. 20, for information whose "item number" is "1", "a>b" is stored as a "relational expression", and "1" is stored as a "coefficient". is memorized. Furthermore, in the first coefficient information 133a shown in FIG. 20, for information whose "item number" is "2", "a=b" is stored as a "relational expression" and "0" is stored as a "coefficient". has been done. Furthermore, in the first coefficient information 133a shown in FIG. 20, in the information whose "item number" is "3", "a<b" is stored as a "relational expression", and "-1" is stored as a "coefficient". remembered.

[Specific example of second coefficient information]
Next, a specific example of the second coefficient information 133b will be explained.

The second coefficient information 133b shown in FIG. 21 includes an "item number" in which information identifying each piece of information included in the second coefficient information 133b is stored, and the position of the partial image identified in the process of S141 is determined in the process of S151. The items include a "relational equation" in which an equation indicating whether the specified partial image is located on the peripheral side of the position is stored, and a "coefficient" in which a coefficient used in the process of S155 is stored. The "relational expression" includes "a>b", which indicates that the position of the partial image identified in the process of S141 is located on the more distal side than the position of the partial image identified in the process of S151; "a=b" indicating that the horizontal position of the specified partial image is the same as the horizontal position of the partial image specified in the process of S151, or the position of the partial image specified in the process of S141 is "a<b" indicating that the partial image is located closer to the central region than the position of the partial image identified in the process of S151 is set.

Specifically, in the second coefficient information 133b shown in FIG. 21, for information whose "item number" is "1", "a>b" is stored as a "relational expression", and "1" is stored as a "coefficient". is memorized. Furthermore, in the second coefficient information 133b shown in FIG. 21, for information whose "item number" is "2", "a=b" is stored as a "relational expression" and "0" is stored as a "coefficient". has been done. Furthermore, in the second coefficient information 133b shown in FIG. 21, in the information whose "item number" is "3", "a<b" is stored as a "relational expression", and "-1" is stored as a "coefficient". remembered.

[Specific example of third coefficient information]
Next, a specific example of the third coefficient information 133c will be explained.

The third coefficient information 133c shown in FIG. 21 includes an "item number" in which information for identifying each piece of information included in the third coefficient information 133c is stored, and an identification value of the partial image identified in the process of S141. The items include a "relational expression" that stores an equation indicating whether the identification value is greater than the identification value of the partial image identified in step S161, and a "coefficient" that stores a coefficient used in the process of S161. The "relational expression" includes "a>b" indicating that the identifier of the partial image specified in the process of S141 is larger than the identifier of the partial image specified in the process of S151, and the identifier of the partial image specified in the process of S141. "a=b" indicating that the identifier of the partial image specified in the process of S151 is equal to the identifier of the partial image specified in the process of S141, or that the identifier of the partial image specified in the process of S151 is smaller than the identifier of the partial image specified in the process of S151. ``a<b'' indicating that ``a<b'' is set.

Specifically, in the third coefficient information 133c shown in FIG. 22, for information whose "item number" is "1", "a>b" is stored as a "relational expression", and "1" is stored as a "coefficient". is memorized. In addition, in the third coefficient information 133c shown in FIG. 22, for information whose "item number" is "2", "a=b" is stored as a "relational expression" and "0" is stored as a "coefficient". has been done. Furthermore, in the third coefficient information 133c shown in FIG. 22, in the information whose "item number" is "3", "a<b" is stored as a "relational expression", and "-1" is stored as a "coefficient". remembered.

Therefore, for example, if the position of the partial image identified in the process of S141 is on the lower lung field side and central side than the position of the partial image identified in the process of S151, the second identifying unit 114 performs the process of S154. In the process of S155, "1" is specified as the first coefficient, and in the process of S155, "-1" is specified as the second coefficient.

Further, for example, the identification value associated with the partial image identified in the process of S141 is "0.2", and the identification value associated with the partial image identified in the process of S151 is "0.6". If so, the second specifying unit 114 specifies "-1" as the third coefficient in the process of S161.

Returning to FIG. 18, the second specifying unit 114 calculates a comparison value based on the first coefficient specified in the process of S154, the second coefficient specified in the process of S155, and the third coefficient specified in the process of S161. Calculate (S162). Specifically, the second specifying unit 114 calculates, for example, a value obtained by multiplying the sum of the first coefficient and the second coefficient by the third coefficient as the comparison value.

Then, the second specifying unit 114 refers to the threshold information 134 stored in the information storage area 130 and specifies the second threshold corresponding to the comparison value calculated in the process of S162 (S163). A specific example of the threshold information 134 will be described below.

[Specific example of threshold information]
FIG. 23 is a diagram illustrating a specific example of the threshold information 134.

The threshold information 134 shown in FIG. 23 includes an "item number" in which information identifying each piece of information included in the threshold information 134 is stored, and a "comparison value" in which the comparison value calculated in the process of S162 is stored. It has an item "threshold value" in which each threshold value is stored. Note that "d" included in the threshold values stored in "Threshold Values" is, for example, a value determined in advance through experiments or the like. Moreover, "T" included in the threshold value stored in "threshold value" is, for example, a negative value for adjusting the threshold value.

Specifically, in the threshold information 134 shown in FIG. 23, for information whose "item number" is "1", "2" is stored as a "comparison value", and "d+2T" is stored as a "threshold value". There is. Further, in the threshold information 134 shown in FIG. 23, in the information whose "item number" is "2", "1" is stored as the "comparison value" and "d+T" is stored as the "threshold value". Further, in the threshold information 134 shown in FIG. 23, in the information whose "item number" is "3", "0" is stored as the "comparison value" and "d" is stored as the "threshold value".

Furthermore, in the threshold information 134 shown in FIG. 23, for information whose "item number" is "4", "-1" is stored as a "comparison value" and "dT" is stored as a "threshold". ing. In addition, in the threshold information 134 shown in FIG. 23, for information whose "item number" is "5", "-2" is stored as a "comparison value" and "d-2T" is stored as a "threshold value". ing.

That is, in the threshold information 134 shown in FIG. 23, each value is set such that the larger the value stored in the "comparison value" is, the more relaxed (becomes) the second threshold used in the processing described below. There is.

Returning to FIG. 18, the second specifying unit 114 compares the degree of similarity calculated in the process of S153 with the second threshold value specified in the process of S163 (S164).

As a result, if the similarity calculated in the process of S153 is larger than the second threshold value specified in the process of S163 (YES in S164), the second specifying unit 114 combines the partial image specified in the process of S141 with the second threshold value specified in the process of S163. The partial images identified in the process are connected (S165).

That is, honeycomb opacities SH in diffuse lung disease have a property that they have a high probability of occurring in the lower lung fields and periphery. In addition, honeycomb opacities SH in diffuse lung disease may occur in the lower lung fields or periphery and spread toward the upper lung fields or the central region, whereas they may occur singly in the upper lung fields or central region of the lung. It has the property that the probability that it will happen is extremely low.

Therefore, when the position of the partial image specified in the process of S151 is above the position of the partial image specified in the process of S141 or in the case that the position of the partial image specified in the process of S141 is on the central side, the partial image specified in the process of S141 has a honeycomb shadow SH. In this case, it can be determined that the partial image identified in the process of S151 is likely to have become a honeycomb shadow SH due to expansion from the honeycomb shadow SH corresponding to the partial image identified in the process of S141. It is possible.

On the other hand, if the position of the partial image identified in the process of S151 is below or on the peripheral side of the position of the partial image specified in the process of S141, the partial image identified in the process of S151 is Even if the identified partial image is a honeycomb shadow SH, it is determined that there is a low possibility that the honeycomb shadow SH was created by enlarging the honeycomb shadow SH corresponding to the partial image identified in the process of S141. It is possible to do so.

Therefore, if the position of the partial image identified in the process of S151 is above the position of the partial image specified in the process of S141 or is on the central side, the second identifying unit 114 determines that the position of the partial image identified in the process of S141 is The process in S164 is performed using a threshold value that makes it easier to determine whether to connect the image and the partial image identified in the process in S151. As a result, if the position of the partial image identified in the process of S151 is above the position of the partial image identified in the process of S141 or is on the central side, the second identifying unit 114 determines the position of the partial image in S151 as described later. It becomes possible to increase the possibility that the partial image identified by the process will be determined as a honeycomb shadow SH.

On the other hand, if the position of the partial image identified in the process of S151 is below or on the peripheral side of the position of the partial image identified in the process of S141, the second identifying unit 114 determines that the position of the partial image identified in the process of S141 is The process in S164 is performed using a threshold value that makes it more difficult to determine whether to connect the image and the partial image identified in the process in S151. As a result, if the position of the partial image identified in the process of S151 is lower than the position of the partial image identified in the process of S141, or is on the peripheral side, the second identifying unit 114 determines the position of the partial image in S151 as described later. It becomes possible to reduce the possibility that the partial image identified by the process described above will be determined as a honeycomb shadow SH.

Returning to FIG. 18, if all partial images have been identified in the process of S151 (YES in S166), the second identifying unit 114 ends the processes of S15 to S24.

On the other hand, if the identification of all partial images is not completed in the process of S151 (NO in S166), the second identifying unit 114 performs the processes from S151 onwards again.

Thereby, when a certain partial image is determined to be a honeycomb shadow SH, the second identifying unit 114 identifies another partial image that can be determined to be a honeycomb shadow SH. becomes possible. Therefore, the information processing device 1 can extract the honeycomb shadow SH from the new image data 131 input via the operation terminal 2 with high precision, for example.

Returning to FIG. 16, if the process of S143 is performed, or if the classification result is a honeycomb shadow SH in the process of S142, the second identifying unit 114 completes identifying all the partial images in the process of S141. It is determined whether or not (S144).

As a result, if it is determined that all the partial images have been identified in the process of S141 (YES in S144), the second specifying unit 114 selects the partial images of S133 from among the partial images divided in the process of S103. Information indicating that the partial image is a honeycomb shadow SH is associated with the partial image that has been associated with information indicating that the probability of the shadow SH being a honeycomb shadow SH is high (S145).

On the other hand, if it is determined in the process of S141 that the identification of all partial images is not completed (NO in S144), the second identifying unit 114 performs the processes from S141 onwards again.

Returning to FIG. 11, after the process in S24, the information output unit 115 of the information processing device 1 outputs, for example, information indicating that the candidate identified as the honeycomb shadow SH in the processes in S14 and S24 is the honeycomb shadow SH. Output (S25).

Specifically, the information output unit 115 determines that among the partial images divided in the process of S103, the partial image associated with the information indicating that the probability of being a honeycomb shadow SH is high in the process of S133 is a honeycomb shadow SH. Information indicating that it is SH is output to the output device of the operating terminal 2, etc. Furthermore, the information output unit 115 determines that among the partial images divided in the process of S103, the partial image associated with the information indicating that it is a honeycomb shadow SH in the process of S145 is a honeycomb shadow SH. The information shown is output to the output device of the operating terminal 2, etc. A specific example of the information output in the process of S145 will be described below.

[Specific example of information output in the process of S145]
FIG. 24 is a diagram illustrating a specific example of information output in the process of S145, and (A), (B), and (C) are diagrams each showing a coronal cross section of the lung. FIG. 24(A) is information indicating candidates for the honeycomb shadow SH identified in the process of S12, and FIG. 24(B) is information indicating the honeycomb shadow SH identified in the process of S14. FIG. 24C shows information indicating the honeycomb shadow SH identified in the process of S24.

Specifically, FIG. 24(A) includes information indicating the candidate SH11 identified in the lung LU. Further, FIG. 24(B) includes information indicating the honeycomb shadow SH12 identified in the process of S14 among the candidates SH11 identified in the lung LU. Further, FIG. 24C includes information indicating the honeycomb shadow SH13 identified in the process of S24 among the candidates SH11 identified in the lung LU.

In this way, the information processing device 1 in this embodiment extracts candidates for the abnormal shadow SH from the image data 131. Then, the information processing apparatus 1 identifies, as the abnormal shadow SH, a candidate whose specific value based on the detected position in the body satisfies a predetermined condition among the extracted candidates for the abnormal shadow SH.

Thereafter, the information processing device 1 selects candidates detected at a position corresponding to the identified abnormal shadow SH from among the extracted candidates for the abnormal shadow SH and that have a predetermined similarity relationship with the identified abnormal shadow SH. , further specified as an abnormal shadow SH.

That is, the information processing device 1 in the present embodiment not only identifies a candidate for an abnormal shadow SH that is determined to have a high probability of being an abnormal shadow SH as an abnormal shadow SH, but also identifies a candidate as an abnormal shadow SH. A candidate detected at a position corresponding to the determined candidate and similar to the candidate determined to be the abnormal shadow SH is also specified as the abnormal shadow SH.

This allows the information processing device 1 to identify the abnormal shadow SH included in the image data 131 with high accuracy. Specifically, as shown in FIG. 25, the information processing device 1 identifies with high accuracy the abnormal shadow SH that needs to be identified (the abnormal shadow SHa included in the lung LUa and the abnormal shadow SHb included in the lung LUb). becomes possible. Therefore, the information processing device 1 can prevent a decrease in search accuracy of image search.

The above embodiments can be summarized as follows.

(Additional note 1)
Extract candidates for abnormal shadows from images inside the body,
Among the extracted candidates, a candidate whose specific value based on the position detected in the body satisfies a predetermined condition is identified as the abnormal shadow;
further specifying a candidate detected at a position corresponding to the identified abnormal shadow among the extracted candidates and having a predetermined similarity relationship with the identified abnormal shadow as the abnormal shadow;
An abnormal shadow identification program characterized by causing a computer to execute processing.

(Additional note 2)
In Appendix 1,
In the extraction process, the candidates are extracted by using a classifier generated by learning a plurality of learning data for images containing abnormal shadows and a plurality of learning data for images not containing abnormal shadows. I do,
An abnormal shadow identification program.

(Additional note 3)
In Appendix 2,
In the extraction process,
For each partial image included in the in-body image, obtain information output from the classifier in response to inputting each partial image,
extracting, as the candidate, a partial image that includes information indicating that the acquired information is the candidate from among the partial images included in the in-body image;
An abnormal shadow identification program.

(Additional note 4)
In Appendix 1,
In the process of identifying candidates that meet the predetermined conditions,
For each of the extracted candidates, the specific value is determined from a value indicating the likelihood of occurrence of an abnormal shadow at the position where each candidate is detected in the body, and a value indicating the probability of each candidate as the abnormal shadow. Calculate,
Among the extracted candidates, a candidate for which the calculated specific value is larger than a first threshold is identified as the abnormal shadow;
An abnormal shadow identification program.

(Appendix 5)
In Appendix 1,
In the process of identifying candidates having a predetermined similarity relationship with the abnormal shadow,
For each of the extracted candidates, identify other candidates located adjacent to each other in the body among the extracted candidates;
For each of the extracted candidates and for each of the other candidates, calculate a value indicating the degree of similarity between the information included in the extracted candidate and the information included in the other candidate,
Among the combinations of the extracted candidate and the other candidate, each candidate included in the combination in which the calculated similarity exceeds a second threshold is connected;
Among the extracted candidates, a candidate that is connected to the candidate identified in the process of identifying a candidate satisfying the predetermined condition is identified as the abnormal shadow;
An abnormal shadow identification program.

(Appendix 6)
In Appendix 5,
In the process of identifying candidates having a predetermined similarity relationship with the abnormal shadow,
For each of the extracted candidates and for each of the other candidates, the second threshold is determined based on the positional relationship of each candidate in the body and the comparison result of the probability of each candidate as the abnormal shadow. ,
connecting each candidate included in a combination among the combinations of the extracted candidate and the other candidate for which the calculated degree of similarity exceeds the second threshold corresponding to each combination;
An abnormal shadow identification program.

(Appendix 7)
In Appendix 1,
The image inside the body is a slice image of the lungs.
An abnormal shadow identification program.

(Appendix 8)
In Appendix 1,
The abnormal shadow is a honeycomb shadow.
An abnormal shadow identification program.

(Appendix 9)
a candidate extraction unit that extracts candidates for abnormal shadows from internal images of the body;
A first identifying unit that identifies, as the abnormal shadow, a candidate whose specific value based on the position detected in the body satisfies a predetermined condition among the extracted candidates;
A second identification of further identifying a candidate detected at a position corresponding to the specified abnormal shadow among the extracted candidates and having a predetermined similarity relationship with the specified abnormal shadow as the abnormal shadow. having a part and
An abnormal shadow identification device characterized by:

(Appendix 10)
In Appendix 9,
The first specific part is
For each of the extracted candidates, the specific value is determined from a value indicating the likelihood of occurrence of an abnormal shadow at the position where each candidate is detected in the body, and a value indicating the probability of each candidate as the abnormal shadow. Calculate,
Among the extracted candidates, a candidate for which the calculated specific value is larger than a first threshold is identified as the abnormal shadow;
An abnormal shadow identification device characterized by:

(Appendix 11)
In Appendix 9,
The second specific part is
For each of the extracted candidates, identify other candidates located adjacent to each other in the body among the extracted candidates;
For each of the extracted candidates and for each of the other candidates, calculate a value indicating the degree of similarity between the information included in the extracted candidate and the information included in the other candidate,
Among the combinations of the extracted candidate and the other candidate, each candidate included in the combination in which the calculated similarity exceeds a second threshold is connected;
Among the extracted candidates, a candidate that is connected to the candidate identified in the process of identifying a candidate satisfying the predetermined condition is identified as the abnormal shadow;
An abnormal shadow identification device characterized by:

(Appendix 12)
In Appendix 11,
The second specific part is
For each of the extracted candidates and for each of the other candidates, the second threshold is determined based on the positional relationship of each candidate in the body and the comparison result of the probability of each candidate as the abnormal shadow. ,
connecting each candidate included in a combination among the combinations of the extracted candidate and the other candidate for which the calculated degree of similarity exceeds the second threshold corresponding to each combination;
An abnormal shadow identification device characterized by:

(Appendix 13)
Extract candidates for abnormal shadows from images inside the body,
Among the extracted candidates, a candidate whose specific value based on the position detected in the body satisfies a predetermined condition is identified as the abnormal shadow;
further specifying a candidate detected at a position corresponding to the identified abnormal shadow among the extracted candidates and having a predetermined similarity relationship with the identified abnormal shadow as the abnormal shadow;
A method for identifying abnormal shadows characterized by the following.

(Appendix 14)
In Appendix 13,
In the step of identifying candidates satisfying the predetermined conditions,
For each of the extracted candidates, the specific value is determined from a value indicating the likelihood of occurrence of an abnormal shadow at the position where each candidate is detected in the body, and a value indicating the probability of each candidate as the abnormal shadow. Calculate,
Among the extracted candidates, a candidate for which the calculated specific value is larger than a first threshold is identified as the abnormal shadow;
A method for identifying abnormal shadows characterized by the following.

(Appendix 15)
In Appendix 13,
In the step of identifying a candidate having a predetermined similarity relationship with the abnormal shadow,
For each of the extracted candidates, identify other candidates located adjacent to each other in the body among the extracted candidates;
For each of the extracted candidates and for each of the other candidates, calculate a value indicating the degree of similarity between the information included in the extracted candidate and the information included in the other candidate,
Among the combinations of the extracted candidate and the other candidate, each candidate included in the combination in which the calculated similarity exceeds a second threshold is connected;
Among the extracted candidates, a candidate that is connected to the candidate identified in the process of identifying a candidate satisfying the predetermined condition is identified as the abnormal shadow;
A method for identifying abnormal shadows characterized by the following.

(Appendix 16)
In Appendix 14,
In the step of identifying a candidate having a predetermined similarity relationship with the abnormal shadow,
For each of the extracted candidates and for each of the other candidates, the second threshold is determined based on the positional relationship of each candidate in the body and the comparison result of the probability of each candidate as the abnormal shadow. ,
connecting each candidate included in a combination among the combinations of the extracted candidate and the other candidate for which the calculated degree of similarity exceeds the second threshold corresponding to each combination;
A method for identifying abnormal shadows characterized by the following.

1: Information processing device 2: Operation terminal 130: Information storage area NW: Network

Claims (6)

Among the abnormal shadow candidates included in the in-body image, a candidate whose specific value based on the detected position in the body satisfies a predetermined condition is identified as the abnormal shadow;
For each abnormal shadow candidate, identifying other candidates located adjacent to each other in the body among the abnormal shadow candidates;
For each of the abnormal shadow candidates and for each of the other candidates, calculate a value indicating the degree of similarity between information included in the abnormal shadow candidate and information included in the other candidates;
A second threshold is determined for each candidate of the abnormal shadow and for each of the other candidates based on the positional relationship of each candidate in the body and the comparison result of the probability of each candidate as the abnormal shadow. ,
out of the combinations of the abnormal shadow candidate and the other candidate, identifying a first combination in which the calculated degree of similarity exceeds the second threshold corresponding to each combination;
Connecting each candidate included in the combination including the candidate identified as the abnormal shadow in the process of identifying the abnormal shadow among the identified first combination;
further specifying, as the abnormal shadow, a candidate that is connected to the abnormal shadow identified in the process of identifying the abnormal shadow among the candidates included in the first combination;
An abnormal shadow identification program characterized by causing a computer to execute processing. In claim 1,
In the process of identifying candidates that meet the predetermined conditions,
For each abnormal shadow candidate included in the in-body image, a value indicating the likelihood of occurrence of an abnormal shadow at the position where each candidate is detected in the body, and a value indicating the probability of each candidate as the abnormal shadow. Calculate the specific value from
Among the abnormal shadow candidates included in the in-vivo image, a candidate for which the calculated specific value is larger than a first threshold is identified as the abnormal shadow;
An abnormal shadow identification program. In claim 1,
The image inside the body is a slice image of the lungs.
An abnormal shadow identification program. In claim 1,
The abnormal shadow is a honeycomb shadow.
An abnormal shadow identification program. a first identifying unit that identifies, as the abnormal shadow, a candidate whose specific value based on the detected position in the body satisfies a predetermined condition among the abnormal shadow candidates included in the internal body image;
For each abnormal shadow candidate, identify another candidate located adjacent to the abnormal shadow candidate in the body, and identify the abnormal shadow for each abnormal shadow candidate and for each other candidate. A value indicating the degree of similarity between the information included in the candidate and the information included in the other candidate is calculated, and for each candidate for the abnormal shadow and for each other candidate, the positional relationship of each candidate in the body is calculated. A second threshold is determined based on the comparison result of the probability of each candidate as the abnormal shadow, and among the combinations of the abnormal shadow candidate and the other candidates, the calculated similarity is A first combination exceeding the second threshold value corresponding to the combination is identified, and among the identified first combinations, the combination includes the candidate identified as the abnormal shadow in the process of identifying the abnormal shadow. a second specifying unit that connects each of the candidates included in the first combination and further specifies, as the abnormal shadow, a candidate that is connected to the abnormal shadow identified in the process of identifying the abnormal shadow; , has
An abnormal shadow identification device characterized by: a first identifying unit identifies, as the abnormal shadow, a candidate whose specific value based on the position detected in the body satisfies a predetermined condition among the abnormal shadow candidates included in the in-body image;
A second identifying unit identifies, for each abnormal shadow candidate, another candidate located adjacent to the abnormal shadow candidate in the body, and for each abnormal shadow candidate, identifies another candidate located adjacent to the abnormal shadow candidate in the body; For each abnormal shadow candidate, a value indicating the degree of similarity between the information included in the abnormal shadow candidate and the information included in the other candidates is calculated, and for each abnormal shadow candidate and for each other candidate, a value is calculated for each candidate. A second threshold is determined based on the positional relationship in the body and the comparison result of the probability of each candidate as the abnormal shadow, and the calculation identifying a first combination in which the degree of similarity exceeds the second threshold corresponding to each combination, and among the identified first combinations, the abnormal shadow is identified as the abnormal shadow in the process of identifying the abnormal shadow. Connecting each candidate included in a combination including candidates, and further identifying a candidate included in the first combination that is connected to the abnormal shadow identified in the process of identifying the abnormal shadow as the abnormal shadow. do,
A method for identifying abnormal shadows characterized by the following.

Images