JP2014030548A - Image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processor capable of identifying a lesion and comparing the lesion with another image with high accuracy regardless of a distribution shape of feature quantities.SOLUTION: An image processor 3 has a hard disk 27 for holding a feature quantity distribution information group being a set of feature quantity distribution information in a predetermined biological mucous membrane state as a feature quantity distribution information group for reference, an image input part 21 for inputting a medical image 41, a feature quantity distribution calculation part 31 for calculating at least one feature quantity from the medical image 41 inputted by the image input part 21, and deriving distribution information of the calculated feature quantity, an inter-distribution distance calculation part 32 for calculating a plurality of inter-distribution distances being a scale representing a distribution similarity between the feature quantity distribution information group for reference held by the hard disk 27 and the feature quantity distribution information calculated by the feature quantity distribution calculation part 31, a representative value calculation part 33 for calculating a representative value of the plurality of inter-distribution distances from the plurality of inter-distribution distances, and a discrimination part 34 for discriminating a biological mucous membrane state of a mucous membrane on the basis of the representative value.

Description

  The present invention relates to an image processing apparatus, and more particularly to an image processing apparatus used for diagnosis of a living mucous membrane.

  Conventionally, for example, a plurality of feature quantities such as the length of a blood vessel are calculated from a medical image obtained by imaging a biological mucous membrane by inserting an endoscope into a body cavity of a subject, and an average value of the calculated feature quantities or 2. Description of the Related Art Image processing apparatuses that identify lesions and compare with other images based on the distribution and reference feature values stored in a database or the like are widely used.

  For example, in Japanese Patent Application Laid-Open No. 2008-146539, a plurality of similarities between feature amounts of the feature amount data of the input face image data and the feature amount data stored in the feature amount data storage unit are calculated. Disclosed is a face authentication apparatus that calculates an average value and a variance value of similarities between feature quantities of the two.

  This face authentication device calculates an average value and variance from the similarity between a plurality of feature amounts, assuming that the similarity between the plurality of feature amounts has a normal distribution, and is registered in the face image data registration unit. It is determined whether the existing face image data and the input face image data belong to the same person.

JP 2008-146539 A

Yossi Rubner, Carlo Tomasi, Leonidas J. et al. Guibas, (1998): “A Metric for Distributions with Applications to Image Databases”, Processed ICCV 1998: 59-66. Kullback, S.M. , And Leibler, R .; A. , (1951): “On information and safety”, Anals of Mathematical Statistics 22: 79-86.

  However, if the distribution of feature values does not follow the normal distribution when identifying lesions or comparing with other images based on the average value or variance of multiple feature values calculated from the input image, the average value or variance Is not a correct summary of the distribution, and there is a problem that the reliability of the identification result of the lesion and the comparison result with other images becomes low.

  SUMMARY An advantage of some aspects of the invention is that it provides an image processing apparatus capable of accurately identifying a lesion and comparing it with another image regardless of the distribution of feature amounts.

  An image processing apparatus according to an aspect of the present invention includes a reference feature quantity distribution information group holding unit that holds a feature quantity distribution information group that is a set of feature quantity distribution information in a predetermined biological mucosa state as a reference feature quantity distribution information group An image input unit that inputs a medical image obtained by imaging the biological mucous membrane, a feature amount calculation unit that calculates at least one feature amount related to the biological mucous membrane from the medical image input to the image input unit, A distribution information deriving unit for deriving distribution information of the feature amount calculated by the feature amount calculating unit; the reference feature amount distribution information group held by the reference feature amount distribution information group holding unit; and the distribution information deriving unit. The inter-distribution distance calculation unit that calculates a plurality of inter-distribution distances, which is a measure representing the similarity to the distribution information derived by the above, and the plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit Comprising a representative value calculation unit that calculates a representative value of the distribution distance, and a determination unit to determine the state of the living body mucous membrane of the inputted medical image based on the representative value.

  An image processing apparatus according to another aspect of the present invention includes a representative feature quantity distribution information holding unit that holds representative feature quantity distribution information calculated from a plurality of feature quantity distribution information in a predetermined biological mucosa state, and a biological mucosa. A feature amount calculation unit that calculates at least one feature amount from the medical image, and a distribution information derivation unit that derives distribution information of the feature amount calculated by the feature amount calculation unit. And the inter-distribution distance, which is a measure representing the similarity between distributions, based on the representative feature quantity distribution information held by the representative feature quantity distribution information holding unit and the distribution information derived by the distribution information deriving unit. An inter-distribution distance calculation unit to be calculated; and a determination unit that determines a state of the biological mucosa of the biological mucosa based on the inter-distribution distance calculated by the inter-distribution distance calculation unit.

  According to the image processing apparatus of the present invention, it is possible to accurately identify a lesion and compare it with another image regardless of the feature shape distribution shape.

It is a figure which shows the structure of the medical system which comprises the image processing apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the main processing functions which an arithmetic processing part comprises. It is a figure for demonstrating the process in a feature-value distribution information calculation part. It is a figure for demonstrating the process in the distance calculation part between distributions. 3 is a flowchart illustrating an example of a processing flow of the image processing apparatus according to the first embodiment. It is a flowchart for demonstrating the example of the flow of the representative value calculation process of step S6. It is a figure which shows the main processing functions which the arithmetic processing part of the modification 1 comprises. 12 is a flowchart for explaining an example of a flow of a representative value calculation process in step S6 of Modification 1. It is a figure which shows the main processing functions which the arithmetic processing part of the modification 2 comprises. 14 is a flowchart for explaining an example of the flow of a representative value calculation process in step S6 of Modification 2. It is a figure which shows the main processing functions which the arithmetic processing part of the modification 3 comprises. 14 is a flowchart for explaining an example of a flow of a representative value calculation process in step S6 of Modification 3. It is a figure for demonstrating the process in the distance calculation part between distribution of the modification 4. FIG. 10 is a flowchart illustrating an example of a processing flow of an image processing apparatus according to a fourth modification. It is a figure which shows the main processing functions which the arithmetic processing part of 2nd Embodiment comprises. It is a figure for demonstrating the process in the distance calculation part between distribution of 2nd Embodiment. 12 is a flowchart illustrating an example of a processing flow of the image processing apparatus according to the second embodiment. It is a figure for demonstrating the example of the process which derives representative feature-value distribution information. It is a figure for demonstrating the example of the other process which derives representative feature-value distribution information.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
(First embodiment)

  First, based on FIG. 1, the structure of the medical system which comprises the image processing apparatus which concerns on the 1st form of this invention is demonstrated.

  FIG. 1 is a diagram illustrating a configuration of a medical system including an image processing apparatus according to a first embodiment of the present invention.

  As shown in FIG. 1, the medical system 1 includes a medical observation device 2 that captures an image of a subject such as a biological mucous membrane surface in a body cavity and outputs a video signal, a personal computer, and the like, and is output from the medical observation device 2. The image processing device 3 performs image processing on the image signal, outputs the image signal after the image processing as an image signal, and displays an image based on the image signal output from the image processing device 3. And a monitor 4.

  The medical observation device 2 is inserted into the body cavity, and images the subject in the body cavity and outputs it as an imaging signal, and the subject to be imaged by the endoscope 6 is illuminated. Various controls are performed on the light source device 7 that supplies illumination light (for example, RGB light) and the endoscope 6, and a video signal is generated by performing signal processing on the imaging signal output from the endoscope 6. And a camera control unit (hereinafter abbreviated as CCU) 8 and a monitor 9 for displaying an image of a subject imaged by the endoscope 6 based on a video signal output from the CCU 8. It is configured.

  The endoscope 6 as a medical imaging apparatus includes an insertion unit 11 that is inserted into a body cavity and an operation unit 12 that is provided on the proximal end side of the insertion unit 11. A light guide 13 for transmitting illumination light supplied from the light source device 7 is inserted into the insertion portion 11 from the proximal end side to the distal end portion 14 on the distal end side.

  The light guide 13 is configured such that the distal end side is disposed at the distal end portion 14 of the endoscope 6 and the rear end side is connectable to the light source device 7. And according to such a structure, after the illumination light supplied from the light source device 7 is transmitted by the light guide 13, the illumination window (not shown) provided in the front-end | tip surface of the front-end | tip part 14 of the insertion part 11 is shown. ). The biological mucous membrane or the like as the subject is illuminated by the illumination light emitted from the illumination window.

  At the distal end portion 14 of the endoscope 6, an objective optical system 15 attached to an observation window (not shown) arranged at a position adjacent to the above-described illumination window, and an imaging position of the objective optical system 15 are arranged. An image pickup unit 17 having an image pickup element 16 made of a CCD or the like is provided.

  The image sensor 16 is connected to the CCU 8 via a signal line. The image sensor 16 is driven based on the drive signal output from the CCU 8 and outputs an image signal obtained by imaging the subject imaged by the objective optical system 15 to the CCU 8.

  The imaging signal input to the CCU 8 is signal-processed by a signal processing circuit (not shown) provided in the CCU 8 to be converted into a video signal and output. The video signal output from the CCU 8 is input to the monitor 9 and the image processing device 3. As a result, an image of the subject based on the video signal output from the CCU 8 is displayed on the monitor 9.

  The image processing device 3 includes an image input unit 21 that performs processing such as A / D conversion on the video signal output from the medical observation device 2 to generate image data, a CPU, and the like. An arithmetic processing unit 22 that performs various processes on the image data output from 21, a program storage unit 23 that stores a program (and software) related to processing executed in the arithmetic processing unit 22, and an image An image storage unit 24 that can store image data output from the input unit 21 and an information storage unit 25 that can store the processing result of the arithmetic processing unit 22 are included. A medical image obtained by imaging the biological mucous membrane or the like in the medical observation device 2 is input to the image input unit 21.

  Further, the image processing apparatus 3 can store a processing result of a storage device I / F (interface) 26 connected to a data bus 30 described later and an arithmetic processing unit 22 output via the storage device interface 26. The hard disk 27, a display processing unit 28 for generating and outputting an image signal for displaying the processing result of the arithmetic processing unit 22 on the monitor 4, and an input device such as a keyboard are provided. And an input operation unit 29 capable of inputting parameters in 22 processes, operation instructions for the image processing apparatus 3, and the like.

  The image input unit 21, the arithmetic processing unit 22, the program storage unit 23, the image storage unit 24, the information storage unit 25, the storage device interface 26, the display processing unit 28, and the input operation unit 29 of the image processing apparatus 3 are They are connected to each other via a data bus 30.

  FIG. 2 is a diagram showing main processing functions provided in the arithmetic processing unit, FIG. 3 is a diagram for explaining processing in the feature amount distribution information calculating unit, and FIG. 4 is a calculation of distance between distributions. It is a figure for demonstrating the process in a part.

  As illustrated in FIG. 2, the arithmetic processing unit 22 includes a processing function of the feature amount distribution information calculation unit 31, a processing function of the inter-distribution distance calculation unit 32, a processing function of the representative value calculation unit 33, and a determination unit 34. And a processing function.

  As shown in FIG. 3, a medical image 41 obtained by imaging a biological mucous membrane or the like with the imaging element 16 of the endoscope 6 is input from the image input unit 21 to the feature amount distribution information calculation unit 31. From this medical image 41, for example, a feature region such as a blood vessel or a fine structure of the surface of the mucosa (mucosal surface fine structure) is extracted by a feature region extraction unit (not shown).

  The feature quantity distribution information calculation unit 31 calculates, for example, feature quantities such as the length and width of blood vessels extracted from the medical image 41, and derives the calculated feature quantity distribution information (hereinafter referred to as feature quantity distribution information). . This feature quantity distribution information is represented by a histogram or a signature represented by a representative feature quantity vector representing a feature quantity in a predetermined region of the feature quantity distribution information and a weight coefficient for the representative feature quantity vector. Thus, the feature amount distribution information calculated by the feature amount distribution information calculating unit 31 is output to the inter-distribution distance calculating unit 32. The feature quantity distribution information calculation unit 31 includes a feature quantity calculation unit that calculates at least one feature quantity from the input medical image 41, and a distribution information derivation unit that derives distribution information of the feature quantity calculated by the feature quantity calculation unit. And configure.

  In the inter-distribution distance calculation unit 32, in addition to the feature amount distribution information calculated by the feature amount distribution information calculation unit 31, a set of feature amount distribution information in a predetermined biological mucosal state as a reference feature amount distribution information group (FIG. 4). Then, a feature amount distribution information group (three feature amount distribution information) is input. A feature amount distribution information group that is a set of feature amount distribution information in the predetermined biological mucosa state is stored in the hard disk 27. For example, as shown in FIG. 4, class A, which is a predetermined biological mucous membrane state, corresponds to image A1 to A3, each feature quantity of images A1 to A3, and feature quantity distribution information (histogram or signature). It is attached. Note that only the feature amount distribution information may be stored in the hard disk 27, or only the images A1 to A3 or the feature amount are stored, and the arithmetic processing unit 22 performs the feature from the image A1 to A3 or the feature amount. The amount distribution information may be calculated.

  The hard disk 27 as a reference feature quantity distribution information group holding unit is a feature quantity distribution information which is a set of feature quantity distribution information stored in a predetermined biological mucosa state via the storage device I / F 26 and the data bus 30. The group is stored as a reference feature amount distribution information group, and is output to the inter-distribution distance calculation unit 32 of the arithmetic processing unit 22.

  The inter-distribution distance calculation unit 32 is an inter-distribution distance that is a measure representing the similarity of distribution between the feature quantity distribution information group output from the hard disk 27 and the feature quantity distribution information calculated by the feature quantity distribution information calculation unit 31. (For example, a plurality of EMD: Earth Mover's Distance) described in Non-Patent Document 1 are calculated. The inter-distribution distance is not limited to EMD, and Kullback-Leibler divergence described in Non-Patent Document 2, which is a known technique, may be used.

  The representative value calculation unit 33 calculates a representative value of a plurality of inter-distribution distances from the plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit 32. In the present embodiment, the representative value calculation unit 33 calculates an average value of a plurality of inter-distribution distances as a representative value.

  The determination unit 34 determines whether the input medical image 41 belongs to the class A by determining the threshold value of the representative value calculated by the representative value calculation unit 33, that is, an average value of a plurality of inter-distribution distances. To determine the state of the biological mucosa. The determination result is displayed on the monitor 4 by the display processing of the display processing unit 28, for example.

  Next, the operation of the image processing apparatus configured as described above will be described.

  FIG. 5 is a flowchart illustrating an example of a processing flow of the image processing apparatus according to the first embodiment.

  First, the medical image 41 imaged by the imaging element 16 of the endoscope 6 is input (step S1). Next, a plurality of feature amounts of the medical image 41 (input image) are calculated by the feature amount distribution information calculation unit 31 (step S2), and feature amount distribution information is calculated from the plurality of calculated feature amounts (step S3). . Next, the feature amount distribution information group stored in the hard disk 27 is read out by the storage device I / F 26 (step S4).

  Next, the inter-distribution distance calculation unit 32 calculates a plurality of inter-distribution distances between the feature amount distribution information calculated by the feature amount distribution information calculation unit 31 and the read feature amount distribution information group (step S5). . Then, the representative value calculation unit 33 performs a representative value calculation process for the plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit 32 (step S6). Based on the representative value calculated by the representative value calculation process, the determination unit 34 determines whether the medical image 41 belongs to class A (step S7), and ends the process.

  Here, the representative value calculation processing in step S6 will be described with reference to FIG. FIG. 6 is a flowchart for explaining an example of the flow of the representative value calculation process in step S6.

  In the representative value calculation process of step S6, the average value of the calculated plurality of distribution distances is calculated as a representative value (step S8), and the process ends.

  As described above, the image processing apparatus 3 calculates the feature amount distribution information from the plurality of feature amounts of the input medical image 41, and the feature amount distribution information and the reference feature amount distribution information in the hard disk 27 as a database. A plurality of inter-distribution distances with the feature amount distribution information group stored as a group are calculated. Then, the image processing device 3 calculates the average value of the calculated distances between the distributions, and determines the biological mucosa state based on the average value.

  Further, the hard disk 27 as a reference feature quantity distribution information group holding unit is not limited to one that holds a feature quantity distribution information group that is a set of stored feature quantity distribution information in a predetermined biological mucosa state. Alternatively, distribution information of a plurality of reference feature amounts calculated from one or more medical images different from the medical image 41 may be held as reference feature amount distribution information. Further, the hard disk 27 further holds attribute information such as imaging conditions and patient information of one or more medical images different from the input medical image 41, and the image processing apparatus 3 receives the input medical image 41. The attribute information may be determined based on the inter-distribution distance between the feature amount distribution information derived from the reference feature distribution information and the reference feature amount distribution information.

According to the configuration of the image processing apparatus 3 of the present embodiment as described above, it is possible to accurately identify a lesion and compare with another image regardless of the feature amount distribution shape.
(Modification 1)

  Next, Modification 1 of the first embodiment will be described.

  FIG. 7 is a diagram illustrating main processing functions included in the arithmetic processing unit according to the first modification. In FIG. 7, the same components as those in FIG. 2 are denoted by the same reference numerals and description thereof is omitted.

  As shown in FIG. 7, the arithmetic processing unit 22a is configured by adding the processing function of the outlier determination unit 35 to the configuration of the arithmetic processing unit 22 of FIG.

  The outlier determination unit 35 determines whether or not there are outliers in the plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit 32, and outputs the determination result to the representative value calculation unit 33. As a method for detecting an outlier, a known method, for example, a Smirnov-Grubbs test or a Thompson test is applied to a plurality of calculated distances between distributions.

  The representative value calculation unit 33 calculates the median of a plurality of inter-distribution distances when it is determined by the outlier determination unit 35 that there is an outlier, and the plurality of inter-distribution distances when it is determined that there is no outlier. The average value of is calculated. The determination unit 34 determines whether the input medical image 41 belongs to class A by determining the threshold value of the median value or average value calculated by the representative value calculation unit 33, thereby determining the biological mucosal state of the biological mucosa. Determine.

  Next, an operation of the image processing apparatus according to the first modification configured as described above will be described. Note that the processing flow of the image processing apparatus according to the first modification is the same as that in FIG. 5, and the content of the representative value calculation processing in step S6 is different from that in the first embodiment.

  FIG. 8 is a flowchart for explaining an example of the flow of the representative value calculation process in step S6 of the first modification.

  When a plurality of inter-distribution distances are calculated in step S5, the outlier determination unit 35 determines the presence or absence of an outlier (step S11). When it is determined that there is an outlier, the representative value calculation unit 33 calculates a median value of a plurality of inter-distribution distances as a representative value (step S12). On the other hand, when it is determined that there is no outlier, the representative value calculator 33 calculates an average value of a plurality of inter-distribution distances as a representative value (step S13). In step S7 in FIG. 5, it is determined whether the input image belongs to class A based on the median value calculated in step S12 or the average value calculated in step S13.

In the case where the outlier jumps out and is large, the influence of the outlier increases when the average value of the distances between the distributions is obtained as in the first embodiment. On the other hand, when there is an outlier as in Modification 1, by calculating the median, the influence of the error that the outlier has on the average value can be reduced.
(Modification 2)

  Next, Modification 2 of the first embodiment will be described.

  FIG. 9 is a diagram illustrating main processing functions included in the arithmetic processing unit according to the second modification. In FIG. 9, the same components as those in FIG. 2 are denoted by the same reference numerals and description thereof is omitted.

  As shown in FIG. 9, the arithmetic processing unit 22b is configured by adding the processing function of the variation determining unit 36 to the configuration of the arithmetic processing unit 22 of FIG.

  The variation determination unit 36 determines the magnitude of variation in the plurality of inter-distribution distances by calculating a parameter representing variation with respect to the plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit 32, and the determination result Is output to the representative value calculation unit 33. Here, the definition of the variation is, for example, that the variation is large when the standard deviation or the variance of the distances between the distributions is equal to or greater than a predetermined threshold. In another embodiment, the distribution may be large when the shape of the distribution is concave. Here, the term “concave” refers to a state where the frequency near the average value / median value of the distribution is small and the frequency outside the distribution is large. In this modification, a normalized histogram is created from the distances between multiple distributions, and the frequency of the class including the mean and median is compared with the frequency of the class outside the distribution, and the difference is greater than or equal to a predetermined threshold. Suppose that the variation is large.

  The representative value calculation unit 33 calculates the minimum value of the plurality of distribution distances when the variation determination unit 36 determines that the variation is large, and when it is determined that the variation is small, the average value of the plurality of distribution distances Is calculated. The determination unit 34 determines whether the input medical image 41 belongs to class A by determining the threshold value of the minimum value or average value calculated by the representative value calculation unit 33, thereby determining the biological mucosa state of the biological mucosa. Determine.

  Next, an operation of the image processing apparatus according to the second modification configured as described above will be described. Note that the processing flow of the image processing apparatus according to the second modification is the same as that in FIG. 5, and the contents of the representative value calculation processing in step S6 are different from those in the first embodiment.

  FIG. 10 is a flowchart for explaining an example of the flow of the representative value calculation process in step S6 of the second modification.

  When a plurality of inter-distribution distances are calculated in step S5, the variation determining unit 36 determines the magnitude of variation (step S21). When it is determined that the variation is large, the representative value calculation unit 33 calculates the minimum value of the plurality of distribution distances as the representative value (step S22). On the other hand, when it is determined that the variation is small, the representative value calculation unit 33 calculates an average value of a plurality of inter-distribution distances as a representative value (step S23). In step S7 in FIG. 4, it is determined whether the input image belongs to class A based on the minimum value calculated in step S22 or the average value calculated in step S23.

According to the configuration of Modification 2 as described above, even when the variation in the reference feature amount distribution information group is large within the same class, it is difficult to be affected by the variation.
(Modification 3)

  Next, Modification 3 of the first embodiment will be described.

  FIG. 11 is a diagram illustrating main processing functions provided in the arithmetic processing unit of the third modification. In FIG. 9, the same components as those in FIGS. 2, 7, and 9 are denoted by the same reference numerals and description thereof is omitted.

  As shown in FIG. 11, the arithmetic processing unit 22c is configured by adding the processing functions of the outlier determination unit 35 of FIG. 7 and the variation determination unit 36 of FIG. 9 to the configuration of the arithmetic processing unit 22 of FIG. ing.

  The representative value calculation unit 33 calculates a minimum value of a plurality of inter-distribution distances when the outlier determination unit 35 determines that there is an outlier and the variation determination unit 36 determines that the variation is large. On the other hand, the representative value calculation unit 33 calculates a median of a plurality of inter-distribution distances when the outlier determination unit 35 determines that there is an outlier and the variation determination unit 36 determines that the variation is small. .

  The representative value calculation unit 33 calculates the minimum value of the distances between the distributions when the outlier determination unit 35 determines that there is no outlier and the variation determination unit 36 determines that the variation is large. . On the other hand, the representative value calculation unit 33 calculates an average value of a plurality of inter-distribution distances when it is determined by the outlier determination unit 35 that there is no outlier and the variation determination unit 36 determines that the variation is small. . The determination unit 34 determines whether the input medical image 41 belongs to class A by determining the threshold value of the minimum value, median value, or average value calculated by the representative value calculation unit 33, thereby determining the living body of the living mucous membrane. Determine mucosal condition.

  Next, the operation of the image processing apparatus of the third modification configured as described above will be described. Note that the processing flow of the image processing apparatus according to the third modification is the same as that in FIG. 5, and the content of the representative value calculation processing in step S6 is different from that in the first embodiment.

  FIG. 12 is a flowchart for explaining an example of the flow of the representative value calculation process in step S6 of the third modification.

  When a plurality of inter-distribution distances are calculated in step S5, the outlier determination unit 35 determines the presence or absence of an outlier (step S31). If it is determined that there is an outlier, the variation determination unit 36 determines the magnitude of the variation in step S32. When it is determined that there is an outlier and the variation is large, the representative value calculation unit 33 calculates a minimum value of a plurality of inter-distribution distances as a representative value (step S33). On the other hand, if it is determined that there is an outlier and the variation is small, the representative value calculation unit 33 calculates a median value of a plurality of inter-distribution distances as a representative value (step S34).

  If it is determined in step S31 that there is no outlier, in step S35, the variation determination unit 36 determines whether the variation is large or small. When it is determined that there is no outlier and the variation is large, the representative value calculation unit 33 calculates a minimum value of a plurality of inter-distribution distances as a representative value (step S36). On the other hand, when it is determined that there is no outlier and variation is small, the representative value calculation unit 33 calculates an average value of a plurality of inter-distribution distances as a representative value (step S37). In step S7 in FIG. 5, the input is made based on the minimum value calculated in step S33, the median value calculated in step S34, the minimum value calculated in step S36, or the average value calculated in step S37. It is determined whether the image belongs to class A.

According to the configuration of the modification example 3 as described above, it is possible to reduce the influence of the error that the outlier has on the average value and the variation in the reference feature amount distribution information group in the same class is large. However, it can be made less susceptible to that effect.
(Modification 4)

  Next, Modification 4 will be described.

  In the fourth modification, a case where the hard disk 27 has a plurality of biological mucous membrane states (classes) will be described. The configuration of the arithmetic processing unit 22 is the same as that in FIG.

  FIG. 13 is a diagram for explaining processing in the inter-distribution distance calculation unit of the fourth modification.

  The hard disk 27 stores classes A to N, which are different states of the biological mucous membrane. Each class A to N is associated with a feature quantity distribution information group which is a set of three feature quantity distribution information. The hard disk 27 outputs to the inter-distribution distance calculation unit 32 a plurality of feature amount distribution information groups of the biological mucous membrane states.

  The inter-distribution distance calculation unit 32 is based on a plurality of feature quantity distribution information groups output from the hard disk 27 and the feature quantity distribution information calculated by the feature quantity distribution information calculation unit 31 for each state of the biological mucous membrane (for each class). ) To calculate a plurality of inter-distribution distances.

  The representative value calculation unit 33 calculates an average value of a plurality of distribution distances for each state of the biological mucosa calculated by the distribution distance calculation unit 32. In addition, it is set as the structure which has the outlier determination part 35 and / or the variation determination part 36 of the modifications 1-3, and the representative value calculation part 33 is based on the determination result of the outlier determination part 35 and / or the variation determination part 36. A median value, minimum value, or average value of a plurality of distribution distances for each state of the mucous membrane may be calculated.

  The determination unit 34 compares and determines the average value of the plurality of distribution distances for each state of the biological mucosa calculated by the representative value calculation unit 33, and the input medical image 41 belongs to any class of classes A to N. By determining whether or not the biological mucosa state of the biological mucosa.

  Next, the operation of the image processing apparatus of the modification 4 configured as described above will be described.

  FIG. 14 is a flowchart illustrating an example of a processing flow of the image processing apparatus according to the fourth modification. In FIG. 14, the same processes as those in FIG.

  When the feature amount distribution information of the input medical image 41 is calculated in step S3, a plurality of feature amount distribution information groups stored in the hard disk 27 are read out by the storage device I / F 26 (step S41). Next, the inter-distribution distance for each state of the biological mucous membrane between the feature amount distribution information calculated by the feature amount distribution information calculation unit 31 and the read-out feature amount distribution information group by the inter-distribution distance calculation unit 32 Are calculated (step S42). In step S6, a representative value calculation process of the inter-distribution distance for each state of the biological mucous membrane is performed, and to which class A to N the input medical image 41 belongs based on the calculated representative values. Determination is made (step S43), and the process is terminated.

According to the configuration of the fourth modification, it is possible to easily determine which of the different biological mucous membrane states the input medical image 41 belongs to.
(Second Embodiment)

  Next, a second embodiment will be described.

  FIG. 15 is a diagram illustrating main processing functions included in the arithmetic processing unit according to the second embodiment.

  As illustrated in FIG. 15, the arithmetic processing unit 22 d is configured by deleting the representative value calculating unit 33 of the arithmetic processing unit 22 of FIG. 2 and using an inter-distribution distance calculating unit 32 a instead of the inter-distribution distance calculating unit 32. Has been.

  FIG. 16 is a diagram for explaining processing in the inter-distribution distance calculation unit according to the second embodiment.

  As shown in FIG. 16, the hard disk 27 serving as a representative feature quantity distribution information holding unit holds representative reference feature quantity distribution information obtained by integrating a plurality of feature quantity distribution information in class A. More specifically, as shown in FIG. 18, it is assumed that all feature amounts calculated from a plurality of reference images A1 to A5 belonging to class A form one distribution, and representative feature amount distribution information of class A Is derived.

  Further, the present invention is not limited to the method of deriving the representative feature amount distribution information of class A using all the feature amounts calculated from the plurality of reference images A1 to A5 belonging to class A. As shown in FIG. Among the feature amounts calculated from the plurality of reference images belonging to the feature amount, the less frequent feature amount may be excluded as an outlier, and the class A representative feature amount distribution information may be derived using the remaining feature amounts. In the example of FIG. 19, the feature amount calculated from the reference image A1 is excluded as an outlier, and the class A representative feature amount distribution information is derived using the feature amounts of the reference images A2 to A5. Further, the outliers are not limited to feature values with low frequency, but an average value of all feature amounts may be calculated, and the top 5% may be set as outliers in order from the value farthest from the average value.

  The inter-distribution distance calculation unit 32 a calculates the inter-distribution distance between the representative feature amount distribution information output from the hard disk 27 and the feature amount distribution information calculated by the feature amount distribution information calculation unit 31. Then, the determination unit 34 determines the distance between the distributions as a threshold, and determines whether the input medical image 41 belongs to class A.

  Next, the operation of the image processing apparatus configured as described above will be described.

  FIG. 17 is a flowchart illustrating an example of a processing flow of the image processing apparatus according to the second embodiment. In FIG. 17, the same processes as those in FIG.

  When the feature amount distribution information of the input medical image 41 is calculated in step S3, the representative feature amount distribution information stored in the hard disk 27 is read by the storage device I / F 26 (step S51). Next, the inter-distribution distance calculation unit 32a calculates the inter-distribution distance between the feature amount distribution information calculated by the feature amount distribution information calculation unit 31 and the read representative feature amount distribution information (step S52). . Then, based on the calculated inter-distribution distance, whether or not the medical image 41 belongs to class A is determined as a threshold value (step S53), and the process ends.

  According to the configuration of the image processing apparatus 3 of the present embodiment as described above, similarly to the first embodiment, lesion identification and comparison with other images can be performed accurately regardless of the feature shape distribution shape. It can be carried out.

  It should be noted that the steps in the flowcharts in this specification may be executed in a different order for each execution by changing the execution order and performing a plurality of steps at the same time, as long as the steps are not contrary to the nature.

  The present invention is not limited to the above-described embodiments and modifications, and various changes and modifications can be made without departing from the scope of the present invention.

  DESCRIPTION OF SYMBOLS 1 ... Medical system, 2 ... Medical observation apparatus, 3 ... Image processing apparatus, 4 ... Monitor, 6 ... Endoscope, 7 ... Light source device, 8 ... CCU, 9 ... Monitor, 11 ... Insertion part, 12 ... Operation part, DESCRIPTION OF SYMBOLS 13 ... Light guide, 14 ... Tip part, 15 ... Objective optical system, 16 ... Imaging element, 17 ... Imaging part, 21 ... Image input part, 22 ... Arithmetic processing part, 23 ... Program memory part, 24 ... Image memory part, 25 ... Information storage unit, 26 ... Storage device I / F, 27 ... Hard disk, 28 ... Display processing unit, 29 ... Input operation unit, 30 ... Data bus, 31 ... Feature quantity distribution information calculation unit, 32 ... Calculation of distance between distributions Reference numeral 33: Representative value calculation unit 34: Discrimination unit 35: Outlier determination unit 36: Variation determination unit 41: Medical image

Claims (17)

A reference feature quantity distribution information group holding unit that holds a feature quantity distribution information group that is a set of feature quantity distribution information in a predetermined biological mucosa state as a reference feature quantity distribution information group;
An image input unit for inputting a medical image obtained by imaging a biological mucous membrane;
A feature amount calculation unit that calculates at least one feature amount related to the biological mucous membrane from a medical image input to the image input unit;
A distribution information deriving unit for deriving distribution information of the feature amount calculated by the feature amount calculating unit;
A plurality of inter-distribution distances, which are measures representing the similarity between the reference feature quantity distribution information group held by the reference feature quantity distribution information group holding section and the distribution information derived by the distribution information deriving section. An inter-distribution distance calculation unit to calculate,
A representative value calculating unit that calculates a representative value of the plurality of inter-distribution distances from a plurality of inter-distribution distances calculated by the inter-distribution distance calculating unit;
A discriminator for discriminating the state of the biological mucous membrane of the inputted medical image based on the representative value;
An image processing apparatus comprising:   The image according to claim 1, wherein the representative value is any one of an average value, a median value, and a minimum value of a plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit. Processing equipment. An outlier determination unit that determines the presence or absence of an outlier with respect to a plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit;
The said representative value calculation part calculates the median value of the some distance between distributions calculated by the said distance calculation part between distributions based on the determination result of the said outlier determination part. Image processing device. By calculating a parameter representing variation with respect to the plurality of inter-distribution distances calculated by the inter-distribution distance calculating unit, the size of the variation of the plurality of inter-distribution distances calculated by the inter-distribution distance calculating unit is determined. A variation determination unit;
4. The representative value calculation unit calculates a minimum value of a plurality of inter-distribution distances calculated by the inter-distribution distance calculation unit based on a determination result of the variation determination unit. An image processing apparatus according to 1.   The image processing apparatus according to claim 4, wherein the parameter is a standard deviation or a variance of a plurality of inter-distribution distances calculated by the inter-distribution distance calculating unit. The reference feature amount distribution information group holding unit further holds at least one feature amount distribution information group of a state of a biological mucosa different from the state of the predetermined biological mucosa,
The inter-distribution distance calculation unit is configured for each state of the biological mucous membrane from the plurality of feature amount distribution information groups held by the reference feature amount distribution information group holding unit and the distribution information derived by the distribution information deriving unit. 6. The image processing apparatus according to claim 2, wherein a plurality of distribution distances are calculated. A feature region extraction unit that extracts a predetermined feature region from the input medical image;
The image processing apparatus according to claim 6, wherein the feature amount calculation unit calculates a feature amount in the predetermined feature region of the input medical image.   The feature quantity distribution information derived by the distribution information deriving section and the feature quantity distribution information constituting the reference feature quantity distribution information group held by the reference feature quantity distribution information holding section are each feature quantity. The image processing apparatus according to claim 7, wherein the image processing apparatus is represented by a signature represented by a representative feature quantity vector representing a feature quantity in a predetermined region and a weight coefficient for the representative feature quantity vector.   The feature amount distribution information derived by the distribution information deriving unit and each feature amount distribution information constituting the reference feature amount distribution information group held by the reference feature amount distribution information holding unit are represented by a histogram. The image processing apparatus according to claim 7.   The image processing apparatus according to claim 7, wherein the predetermined feature region is a blood vessel or a fine structure of a surface of the biological mucosa. A representative feature quantity distribution information holding unit for holding representative feature quantity distribution information calculated from a plurality of feature quantity distribution information in a predetermined biological mucosal state;
A medical image obtained by imaging a biological mucous membrane is input, and a feature amount calculation unit that calculates at least one feature amount from the medical image;
A distribution information deriving unit for deriving distribution information of the feature amount calculated by the feature amount calculating unit;
Based on the representative feature amount distribution information held by the representative feature amount distribution information holding unit and the distribution information derived by the distribution information deriving unit, an inter-distribution distance that is a measure representing similarity between distributions is calculated. An inter-distribution distance calculator;
A discriminating unit for discriminating the state of the biological mucosa of the biological mucous membrane based on the inter-distribution distance calculated by the inter-distribution distance calculating unit;
An image processing apparatus comprising: The representative feature amount distribution information holding unit further holds at least one representative feature amount distribution information of a biological mucosal state different from the predetermined biological mucosal state,
The inter-distribution distance calculation unit is configured to calculate a distribution interval for each state of the biological mucous membrane from a plurality of representative feature amount distribution information held by the representative feature amount distribution information holding unit and the distribution information derived by the distribution information deriving unit. The image processing apparatus according to claim 11, wherein a distance is calculated. A feature region extraction unit that extracts a predetermined feature region from the input medical image;
The image processing apparatus according to claim 12, wherein the feature amount calculation unit calculates a feature amount in the predetermined feature region of the input medical image.   The feature amount distribution information derived by the distribution information deriving unit and each feature amount distribution information constituting the representative feature amount distribution information held by the representative feature amount distribution information holding unit is a predetermined amount of each feature amount. The image processing apparatus according to claim 13, wherein the image processing apparatus is represented by a signature expressed by a representative feature quantity vector representing a feature quantity in a region and a weighting coefficient for the representative feature quantity vector.   The feature amount distribution information derived by the distribution information deriving unit and each feature amount distribution information constituting the representative feature amount distribution information held by the representative feature amount distribution information holding unit are represented by a histogram. The image processing apparatus according to claim 13, characterized in that:   The image processing apparatus according to claim 13, wherein the predetermined feature region is a blood vessel or a fine structure of a surface of the biological mucosa.   The image processing apparatus according to claim 11, wherein the representative feature quantity distribution information holding unit holds average feature quantity distribution information that is an average of the plurality of feature quantity distribution information as the representative feature quantity distribution information. .