JP2016021248A - Medical diagnosis support apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently present, to a doctor, information to be preferentially confirmed by the doctor.SOLUTION: A medical image targeted for medical diagnosis is input (S201), one or more medical information pieces related to the medical image are acquired as already input information (S203), and an image feature amount is acquired from the medical image (S204). Then, a plurality of un-input information pieces associated with the image feature amount acquired in S204 are selected as un-input information candidates for presentation that are candidates for presentation from among the un-input information pieces which are the medical information pieces other than the already input information, and un-input information pieces for presentation are selected from among the un-input information pieces candidates on the basis of a plurality of inference results acquired by using each of the already input information and the un-input information pieces candidates (S205). Then, processing is executed, where the processing presents, to a doctor, the un-input information pieces for presentation which are selected in S205 (S206).SELECTED DRAWING: Figure 2

Description

  The present invention relates to a technique for providing information that supports medical diagnosis.

  In the medical field, a doctor displays a medical image obtained by imaging a patient on a monitor, interprets the medical image displayed on the monitor, and observes the state of a lesioned part and changes over time. As an apparatus for generating this type of medical image, for example, an X-ray CT (Computed Tomography) apparatus, an MRI (Magnetic Resonance Imaging) apparatus, an ultrasonic apparatus and the like can be cited. Diagnosis using these medical images (image diagnosis) can be divided into a process of finding an abnormal shadow or the like from a medical image and obtaining its characteristics, and a process of differential diagnosis of what the shadow is.

  2. Description of the Related Art Conventionally, for the purpose of assisting a differential diagnosis by a doctor, a medical diagnosis support apparatus has been developed that infers and presents what a shadow is as an input information using abnormal shadow characteristics (interpretation findings) and the like. For example, an apparatus that calculates a probability that a certain shadow of a chest X-ray CT image is a malignant tumor and a probability that it is a benign tumor and presents the same is considered. Usually, as a correct procedure when such an apparatus is used in an actual clinical site, a differential diagnosis is first performed by a doctor, and then a doctor refers to the inference result output by the medical diagnosis support apparatus as reference information.

  Here, when there is a large amount of uninput information, there arises a problem that the accuracy of inference by the medical diagnosis support apparatus is lowered. Therefore, an attempt has been made to obtain a more reliable inference result by the apparatus selecting uninput information necessary for inference and prompting the doctor to add the information.

  For example, in Patent Document 1 below, an inference result (current inference result) of an apparatus based on input information (existing input information) and an inference result when uninput information is added to the already input information are noted. A technique for selecting and presenting non-input information to be described is described. In this technique, the degree of influence of each non-input information on the current inference result is calculated, and the non-input information having a large degree of influence is presented. Further, in Patent Document 1 below, as a method of calculating the influence level, focusing on the diagnosis name having the highest probability in the current inference result, the change amount of the probability when the uninput information is entered is indicated as the influence degree. A method is disclosed. Further, Patent Document 1 below discloses a method in which the total sum of change amounts of each probability when uninput information is entered, taking into account the probability of each diagnosis name in the current inference result, is disclosed. Yes. And in patent document 1 below, it is possible to present uninput information that has a great influence on the inference result of the apparatus based on the already input information.

Japanese Patent No. 3226400

Kondo et al. “Classification of tumor shadows into aerated and solid types by 3D chest X-ray CT images and its application to benign / malignant discrimination”, IEICE Tech. Bulletin, MI2000-16, 2000-05. Tada et al: “Examination of bronchial extraction algorithm using multi-slice CT images”, IEICE Technical Report, MI 2004-37, 2004-09

  However, Patent Document 1 selects not-entered information to be noticed only on the basis of having a great influence on the inference result of the apparatus based on already-input information. For this reason, information (findings) that is unlikely to exist on the medical image may be presented as noticeable uninput information.

  The present invention has been made in view of such problems, and an object thereof is to enable efficient presentation of information that should be confirmed by a doctor with priority.

  The medical diagnosis support apparatus of the present invention is a medical diagnosis support apparatus that provides information for supporting a medical diagnosis, and relates to a medical image input unit that inputs a medical image to be a target of the medical diagnosis, and the medical image Medical information acquisition means for acquiring one or more pieces of medical information as already input information, image feature amount acquisition means for acquiring an image feature amount from the medical image, and uninput information that is medical information other than the already input information A presentation non-input information candidate selection means for selecting a plurality of non-input information associated with the image feature amount as a presentation non-input information candidate that is a candidate for presentation; An inference means for obtaining an inference result using each of the candidates; a presentation uninput information selection means for selecting presentation uninput information from the plurality of presentation uninput information candidates based on the plurality of inference results; And a presenting means for presenting the presentation uninput information selected in 示未 input information selection means.

  ADVANTAGE OF THE INVENTION According to this invention, the information which a doctor should confirm preferentially can be shown efficiently.

1 is a schematic diagram illustrating an example of a schematic configuration of a medical diagnosis support system including a medical diagnosis support apparatus according to an embodiment of the present invention. It is a flowchart which shows an example of the process sequence in the control method of the medical diagnosis assistance apparatus which concerns on embodiment of this invention. It is a schematic diagram which shows embodiment of this invention and shows an example of the probabilistic reasoning model using a Bayesian network. It is a flowchart which shows an example of the detailed process sequence in step S205 shown in FIG. It is a schematic diagram showing an embodiment of the present invention and showing an example of the relationship between the distance from each point of the edge to the center of gravity of the region and the angle, with the vertical axis representing the distance r and the horizontal axis representing the angle θ in the polar coordinate format. It is a schematic diagram which shows an example of the monitor display screen in embodiment of this invention.

  Hereinafter, embodiments (embodiments) for carrying out the present invention will be described with reference to the drawings.

  FIG. 1 is a schematic diagram illustrating an example of a schematic configuration of a medical diagnosis support system including a medical diagnosis support apparatus according to an embodiment of the present invention.

  As shown in FIG. 1, the medical diagnosis support system 10 includes a medical diagnosis support apparatus 100, a medical image database 200, a diagnostic record database 300, and a local area network (LAN) 400. That is, the medical diagnosis support system 10 shown in FIG. 1 is configured such that the medical diagnosis support apparatus 100 is connected to the medical image database 200 and the diagnostic record database 300 via the LAN 400.

  The medical diagnosis support apparatus 100 is an apparatus that provides information that supports medical diagnosis by a doctor or the like. Specifically, the medical diagnosis support apparatus 100 performs inference processing related to medical diagnosis based on a plurality of input medical information, and pays attention. It should present non-input information (which should be checked preferentially by doctors etc.).

  The medical diagnosis support apparatus 100 includes a control unit 110, a monitor 120, a mouse 130, and a keyboard 140.

  The control unit 110 controls the operation of the medical diagnosis support apparatus 100, and includes a central processing unit (CPU) 111, a main memory 112, a magnetic disk 113, a display memory 114, and a bus 115. Has been. Then, when the CPU 111 executes a program stored in the main memory 112, for example, various controls such as communication with the medical image database 200 and the diagnostic record database 300 and overall control of the medical diagnosis support apparatus 100 are executed. The

  The CPU 111 mainly controls the operation of each component of the medical diagnosis support apparatus 100 to control the operation of the medical diagnosis support apparatus 100 in an integrated manner.

  The main memory 112 stores a control program executed by the CPU 111 and provides a work area when the CPU 111 executes the program.

  The magnetic disk 113 stores an operating system (OS), peripheral device driver, various application software, and work data generated or used by various application software.

  The display memory 114 temporarily stores display data to be displayed on the monitor 120.

  The bus 115 connects the components of the medical diagnosis support apparatus 100 so that they can communicate with each other, and connects the medical diagnosis support apparatus 100 and the LAN 400 so that they can communicate with each other.

  The monitor 120 is composed of, for example, a CRT monitor or a liquid crystal monitor, and displays an image or the like based on display data in the display memory 114 under the control of the CPU 111. The execution status and execution result of the program executed by the CPU 111 are displayed on the monitor 120 as necessary. In the present embodiment, an example in which not-entered information to be noted is presented on the monitor 120 will be described. However, the present invention is not limited to this. For example, the present invention includes a form in which these non-input information and the like are output and presented to a printer or the like, and a form in which they are outputted and presented by voice or other means.

  The mouse 130 and the keyboard 140 are used for a pointing input and a character input by the user, respectively. A user (such as a doctor) inputs various commands (instructions / commands) to the medical diagnosis support apparatus 100 by operating the mouse 130 and the keyboard 140.

  The medical diagnosis support apparatus 100 of the present embodiment can read various image data from the medical image database 200 and can read various diagnostic record data from the diagnostic record database 300 via the LAN 400. Here, as the medical image database 200, for example, an existing PACS (Picture Archiving and Communication System) can be used. Moreover, as the diagnostic record database 300, for example, an electronic medical record system which is a subsystem of an existing HIS (Hospital Information System) can be used. Alternatively, an external storage device such as an FDD, HDD, CD drive, DVD drive, MO drive, ZIP drive or the like may be connected to the medical diagnosis support apparatus 100, and medical image data and diagnostic record data may be read from these drives. Good.

  The medical image database 200 stores various types of medical image data obtained by, for example, photographing with a medical image photographing device (not shown). The medical diagnosis support apparatus 100 is connected to the medical image database 200 via the LAN 400, and acquires necessary medical image data therefrom. Here, examples of the types of medical images used in the present embodiment include simple X-ray images (X-ray images), X-ray CT images, MRI images, PET images, SPECT images, and ultrasonic images.

  The diagnostic record database 300 stores various diagnostic record data. The medical diagnosis support apparatus 100 is connected to the diagnosis record database 300 via the LAN 400 and acquires necessary diagnosis record data therefrom. The diagnostic records used in the present embodiment include patient personal information (name, date of birth, age, sex, etc.), clinical information (various test values, chief complaints, medical history, treatment history, etc.), and medical image database 200. Reference information to the stored medical image data of the patient and finding information of the attending physician are described. Further, at the stage where diagnosis has progressed, the diagnosis name is described in the diagnosis record.

  The LAN 400 connects the medical diagnosis support apparatus 100 to the medical image database 200 and the diagnostic record database 300 so that they can communicate with each other.

  Next, a processing procedure in the control method of the medical diagnosis support apparatus 100 according to the present embodiment will be described.

  FIG. 2 is a flowchart showing an example of a processing procedure in the control method of the medical diagnosis support apparatus 100 according to the embodiment of the present invention. Specifically, the flowchart shown in FIG. 2 is realized by the CPU 111 executing a program stored in the main memory 112.

  FIG. 3 is a schematic diagram showing an embodiment of the present invention and an example of a probability inference model using a Bayesian network.

  In the following description, a case where medical diagnosis support apparatus 100 is used to support medical diagnosis related to an abnormal shadow of the lung will be described as an example. Of course, the inference target is not limited to this, and the diagnosis name options and medical information that can be input are merely examples for explaining the processing steps of the medical diagnosis support apparatus 100. .

  Further, in the following description, the choice of diagnosis name (abnormality type) handled by the medical diagnosis support apparatus 100 is described using “D”. In this embodiment, the primary lung cancer, lung metastasis of cancer, and other abnormalities shown in the state 303 of FIG. 3 are handled as options of diagnosis names. In the following, each diagnosis name is “D1”, They are expressed as “D2” and “D3”.

In the following description, the medical information handled by the medical diagnosis support apparatus 100 is denoted as “Ij”. In the present embodiment, the seven types of medical information I _{1 to} I ₇ shown in FIG. 3 are handled. For example, the “shape” of the medical information I ₁ in FIG. 3 represents the shape of the abnormal shadow. Further, “pleural retraction / involvement” in the medical information I ₅ of FIG. 3 indicates the presence / absence of retraction / involvement of the pleura in the organ.

Further, in the following description, states (discrete values) that each “Ij” can take are represented as “Sjk”. Here, the range of k varies depending on “Ij”. In the present embodiment, as shown in FIG. 3, for example, the “shape” of the medical information I ₁ is “shape-spherical” in S11, “shape-lobed” in S12, and “shape-unshaped” in S13. The state can be taken.

  In the following description, a set of “Ij” is represented as “N”. A set of “Sjk” is denoted as “C” and “E”.

  Here, “C” represents a simple set of “Sjk”. For example, when there are S11, S12, and S51 as elements, C = {S11, S12, S51}. In the example shown in FIG. 3, C = {shape-spherical, shape-lobed, pleural retraction / involvement-is}.

  On the other hand, “E” represents a set of “Sjk” that can be set. Here, “settable” indicates that “Sjk” regarding the same “Ij” does not exist simultaneously in one “E”. As in the previous example, if there are S11, S12, and S51 as elements, {S11, S51} can be taken, but {S11, S12} cannot be taken. This is because one piece of medical information can take only one state. In the example shown in FIG. 3, the {shape-spherical shape, pleural retraction / rolling-in} pair can be taken, but the {shape-spherical shape-shape-lobed} pair cannot be taken.

  First, in step S201 in FIG. 2, the CPU 111 performs a process of inputting desired medical image data from the medical image database 200 to the medical diagnosis support apparatus 100 in response to an input from the mouse 130 or the keyboard 140 (medical image input process). ).

  Hereinafter, the medical image data input in step S201 is referred to as an “interpretation target image”. As described above, the CPU 111 receives the desired medical image data as the interpretation target image via the LAN 400 from the medical image database 200 that stores the captured medical image data. Done. In step S201, the CPU 111 interprets desired medical image data from various storage media such as an FDD, CD-RW drive, MO drive, and ZIP drive connected to the medical diagnosis support apparatus 100. A form of reading out as a target image may be used.

  Subsequently, in step S202, the CPU 111 performs processing for displaying the image to be interpreted input in the medical diagnosis support apparatus 100 in step S201 on the monitor 120.

  In step S203, the CPU 111 performs a process of acquiring an interpretation finding input by a doctor using the mouse 130, the keyboard 140, or the like as medical information while viewing the image to be interpreted displayed on the monitor 120 (medical information acquisition step). .

  The processing in step S203 is performed by, for example, using a template-type interpretation finding input support method for a function that can select any one of the states Sjk with respect to each of the medical information Ij using the GUI. This is realized by having 100.

  Here, in the following description, the set of medical information Ij acquired in step S203 is referred to as “Nf”, the set of the state Sjk as “Ef”, and these pieces of information are referred to as “already input information”. For example, the medical information acquired in step S203 includes “shape-spherical” in S11 in FIG. 3, “radial-not” in S33, “bronchial retraction / involvement—is” in S41, and “pleural retraction / involvement” in S53. It is assumed that “winding-not” and S71 “size-small”. In this case, {I1, I3, I4, I5, I7} is acquired as the already input information Nf, and {S11, S33, S41, S53, S71} is acquired as the already input information Ef.

  When there are a plurality of tumors on the image to be interpreted, it is necessary to identify by any method which tumor is the subject of interpretation (input of findings) by the doctor. This is realized, for example, when the doctor designates the position of the tumor using the mouse 130 on the interpretation image displayed on the monitor 120. Here, as a method for specifying the position of the tumor, for example, a method of clicking the center of the tumor or a method of specifying a rectangular region surrounding the tumor can be used.

  Subsequently, in step S204, the CPU 111 performs image processing on the image to be interpreted input in step S201, and acquires various image feature amounts related to the tumor designated as an image interpretation target by the doctor using the mouse 130 or the like in step S203. (Image feature amount acquisition step).

Specifically, in the process of step S204, the following information is acquired as the image feature amount.
First, a region of a tumor designated as an interpretation target is extracted by binarization processing, and the edge of the region is converted based on the degree of coincidence of the regions obtained by performing distance conversion and inversely converting only pixels having a maximum value. Is calculated (for example, see Non-Patent Document 1 described above for details). Further, after extracting the region of the tumor designated as the interpretation target, the distance from each point of the edge to the center of gravity of the region and its variance are calculated. In addition, the bronchus is extracted by using the method described in Non-Patent Document 2 and the like, and it is determined whether or not the bronchus is present inside the tumor. Based on the concentration, it is determined whether or not an annular structure of air is observed. In addition, the maximum diameter of the tumor is calculated.
Note that these image feature amounts are merely examples shown for explaining the image feature amounts in the present embodiment, and the acquired image feature amounts are not limited thereto.

  Further, since the image feature amount acquired in step S204 is for narrowing down the non-input information, only the image feature amount that affects the non-input information may be obtained. In other words, the process for obtaining the image feature amount that affects only the already input information Nf obtained in step S203 may be omitted. For example, when “size-small” in S71 shown in FIG. 3 is obtained as already input information, the maximum diameter of the tumor need not be calculated.

  Subsequently, in step S205, the CPU 111, based on the medical information of the image to be interpreted acquired in step S203 and the image feature amount acquired in step S204, non-input information (hereinafter, not presented) presented in step S206. Input information).

  In the present embodiment, of the inference results based on the already-input information Ef, the one with the highest probability is set as the predicted diagnosis name Df, the uninput information Eh that most affirms the expected diagnosis name Df, and the undetermined unpredictable diagnosis name Df. Input information El is estimated, and this is selected as presentation non-input information.

Details of the processing procedure in step S205 shown in FIG. 2 will be described below.
FIG. 4 is a flowchart showing an example of a detailed processing procedure in step S205 shown in FIG.

  When the process of step S205 in FIG. 2 is started, first, in step S401 in FIG. 4, the CPU 111 receives the already input information Ef and inputs the probability P (i = 1 to 3) of each diagnosis name Di (i = 1 to 3). Di | Ef) is obtained by calculation using a probabilistic inference model. At this time, the CPU 111 saves Di having the highest probability among the probabilities P (Di | Ef) as the predicted diagnosis name Df in the main memory 112, and also calculates the probability P (Di | Ef) (hereinafter referred to as this). “Pf”) is stored in the main memory 112.

  In the following description, the case where the maximum probability Pf = 66.2% is obtained with respect to “cancer lung metastasis” of the diagnosis name D2 when the input information Ef is inferred as an input will be described as an example. The calculation of the inference probability Pf is the second inference step in the present embodiment, and the obtained inference probability Pf corresponds to the second inference result.

  Although there are various inference models for performing inference processing for calculating the inference probability belonging to the diagnostic name of interest using input information, in this embodiment, as a specific example, as shown in FIG. Use the Bayesian Network. Of course, the example shown below is an example of an inference model, and the inference method, the structure of the inference model, and the like are not limited thereto.

  The Bayesian network shown in FIG. 3 is a model in which a phenomenon is expressed by a causal relationship between a plurality of events. In the Bayesian network shown in FIG. 3, the relationship between events is represented by a probability, the event constituting the target phenomenon is represented by a node 301, and the relationship between nodes is represented by a link 302.

  As shown in FIG. 3, the link 302 is represented by an arrow, and a node corresponding to the root of the arrow is referred to as a parent node, and a node corresponding to the tip of the arrow is referred to as a child node. Each node has a plurality of states 303 indicating the state of the node, and an occurrence probability (referred to as “prior probability”) is assigned to each state. The relationship between a parent node and a child node is given by a conditional probability with the parent node as a condition. This table of conditional probabilities is called a conditional probability table (CPT). A CPT 304 shown in FIG. 3 is an example of a CPT between a parent node 301 and a node I7 that is a child node. Information indicating the state (such as 303) of at least one node of the target model is referred to as evidence. Using this evidence, CPT (304), and Bayes' theorem, the probability of a desired node (referred to as “posterior probability”) can be obtained by the probability propagation method.

Returning to the description of FIG.
When the process of step S401 ends, the process proceeds to step S402.
In step S402, the set of medical information states (excluding the already input information) excluding the already input information Nf from all the enterable medical information, that is, a set of unentered medical information states (hereinafter referred to as “Cv”). ”).
For example, in the example of FIG. 3, if the already-input information Nf is {I1, I3, I4, I5, I7}, each set of states I2, I6 {S21, S22, S23, S61 as “Cv”. , S62, S63} are acquired.

Subsequently, in step S403, the CPU 111 acquires a set of medical information states (hereinafter referred to as “Cif”) that can correspond to the image feature amount from the image feature amount acquired in step S204.
For example, the “shape” state of I1 shown in FIG. 3 is selected using the unevenness degree of the edge that is one of the image feature amounts. For example, when it is determined that the degree of unevenness of the edge is large (greater than or equal to a predetermined threshold value), there is a possibility that {shape-unshaped} in S12 and {shape-leaved shape} in S13 may correspond. As an element of “Cif”. On the other hand, if it is determined that the degree of unevenness of the edge is small (below a predetermined threshold value), {shape-spherical shape} in S11 is set as an element of “Cif” as a possible state.

  Also, based on the edge distance, which is one of the image feature amounts, the state of “cut” in I2 is selected. Hereinafter, an example of the selection method will be described with reference to FIG.

FIG. 5 shows an embodiment of the present invention, and is a schematic diagram showing an example of the relationship between the distance from each point on the edge to the center of gravity of the region and the angle in the polar coordinate format in which the vertical axis is the distance r and the horizontal axis is the angle θ. FIG.
First, the barycentric point G (white circle in FIG. 5) of each region 511 to 531 and the point Ma (black circle in FIG. 5) on the edge where the distance from the barycentric point G is maximum (r0) are obtained. Then, the direction of the line segment connecting these two points (hereinafter referred to as “reference direction”) is obtained.

  At this time, an arbitrary point on the edge is defined as a distance r between the center of gravity G and an angle θ (a clockwise direction) formed by a line segment connecting the center of gravity G and the point on the edge and the reference direction. Can be represented by Next, an angle range θsk (k ≧ 1) in which the distance r from the center of gravity G is equal to or less than the threshold value rth is obtained. Also, r (hereinafter referred to as “rmink”) that is minimum within the range of θsk is obtained. When θsk is within a predetermined range and θsk is acquired once or twice, it is determined that there is a high possibility that there is a cut (however, whether weak or strong is unknown).

In the example of the upper part (511 and 512) in FIG. 5, θsk is acquired twice, but since neither θsk is within a predetermined range (not shown), it is determined that there is a low possibility of being cut. The Then, {notch-weak} of S22 and {notch-not} of S23 shown in FIG. 3 are set as elements of “Cif” as possible states.
On the other hand, in the example of the middle stage (521 and 522) in FIG. 5, θsk is acquired once and falls within a predetermined range (not shown), so it is determined that there is a high possibility that there is a cut. Then, {notch-strong} of S21 and {notch-weak} of S22 shown in FIG. 3 are set as elements of “Cif” as states that may correspond.
Moreover, in the example of the lower stage (531 and 532) of FIG. 5, since θsk is acquired six times, it is determined that there is a low possibility of being cut. Then, {notch-weak} of S22 and {notch-not} of S23 shown in FIG. 3 are set as elements of “Cif” as possible states.

  Further, the “radial” state of I3 shown in FIG. 3 is selected based on the dispersion of the edge distance, which is one of the image feature amounts. For example, if it is determined that the variance is large (greater than or equal to a predetermined threshold), {radial-strong} in S31 and {radial-weak} in S32 shown in FIG. Cif ”element. On the other hand, if it is determined that the variance is small (below a predetermined threshold), S32 {radial-weak} and S33 {radial-not} shown in FIG. The element is “Cif”.

Similarly, selection (determination) of the state of “bronchial transparency image” of I6 shown in FIG. 3 is performed based on information on whether or not a bronchus is present in the tumor, which is one of the image feature amounts.
For example, when it is determined that there is no bronchi in the tumor, all the states related to the bronchial transparency image are excluded from the elements of “Cif”. For example, there is a mass on the side of the main bronchus or near the chest wall. This is because it can be determined that the information regarding the bronchial transparency image does not exist in the image. Note that the state of {bronchial bronchial image-not} in S63 indicates a state in which there is no translucency when the bronchus is present in the tumor, and thus indicates a state different from the above example.
On the other hand, when it is determined that the bronchus is present in the mass, the “bronchial transparent image” of I6 shown in FIG. Is further determined. For example, when a tubular structure of air is recognized inside the tumor, the “bronchial brilliant image—some” in S61 is set as an element of “Cif” as a state that may correspond. On the other hand, if a ring structure is not recognized, it is difficult to judge, so all of S61 {bronchial transillumination image-yes}, S62 {bronchial translucency image-unknown}, and S63 {bronchial translucency image-not} Is an element of “Cif” as a state with a corresponding possibility.

Further, the “size” state of I7 shown in FIG. 3 is selected (determined) from the size of the tumor which is one of the image feature amounts.
When the acquired tumor size is smaller than the predetermined threshold, {size-small} in S71 and {size-medium} in S72 are set as elements of “Cif” as possible states. On the other hand, when the number is larger than the predetermined threshold value, {size-large} in S73 and {size-medium} in S72 are set as elements of “Cif” as possible states.

  For medical information that is not determined from the image feature amount, all the states are elements of “Cif”.

  The above-described example is an example showing the correspondence between the image feature quantity and the state of the medical information, and the image feature quantity and the method for selecting the state to be used are not limited to this. For example, only a part of the determination process exemplified above may be used. For example, when it is determined that there is no bronchus inside the tumor, only the process of excluding all the states related to the “bronchial transparency image” of I6 shown in FIG. 3 from “Cif” may be performed. Further, any image feature amount other than those exemplified above may be used, and any determination process for determining the possibility of existence of a state from the image feature amount may be used.

For example, in the process of step S204,
・ Roughness unevenness: below a predetermined threshold,
-Shape of edge distance: θsk has been acquired 0 times,
-Dispersion of edge distance: below a predetermined threshold,
・ The bronchi are present in the mass,
・ A ring structure of air is observed inside the mass,
-Size of the tumor: below a predetermined threshold,
In step S403, {S11, S22, S23, S32, S33, S41, S42, S43, S51, S52, S53, S61, S71, S72} is set as “Cif”.

Returning to the description of FIG.
When the process of step S403 ends, the process proceeds to step S404.
In step S404, the CPU 111 obtains a common state between “Cv” acquired in step S402 and “Cif” acquired in step S403, and a set of possible uninput medical information states (hereinafter referred to as a set of states). , “Cuf”). In the above-described example, when a common state is taken between “Cv” and “Cif”, “Cuf” is expressed as “S22“ notch-weak ”, S23“ notch-not ”, and S61“ bronchial image ”. -"" Is obtained. In other words, “notch—strong” in S21, “bronch bronchial image—don't know” in S62, and “bronch bronchial image—not” in S63 are excluded from the presentation candidates as having a low possibility.

  Subsequently, in step S405, the CPU 111 prepares a variable m and sets m = 1.

  Subsequently, in step S <b> 406, the CPU 111 first selects one or more “Sjk” from “Cuf”, which is a set of candidate medical information states that have not been input. Then, a virtual Em set “Sjk” Em is created (obtained) and used as a presentation non-input information candidate (presentation non-input information candidate selection step). Here, when m> 1 (that is, when the process of step S406 is performed for the second time or later), a group different from E1 to Em-1 created so far is created (acquired). The number of “Sjk” belonging to Em may be plural or less than the plural, and the doctor may designate the number. In the following description, the number of “Sjk” is assumed to be 1. For example, in the above-described example, “Cuf” is “notch-weak” in S22, “not-break” in S23, and “bronch bronchial image-is” in S63 shown in FIG. In this case, for example, E1 is created (acquired) as {S22 “notch-weak”}.

  In step S <b> 407, the CPU 111 calculates a probability P (Df | Ef, Em) belonging to the predicted diagnosis name Df based on the inference model by using a set of the already input information Ef and the presentation uninput information candidate Em as an input. Then, the obtained inference probability (hereinafter referred to as “Pm”) is stored in the main memory 112 in association with the presentation uninput information candidate Em.

  The calculation of the inference probability Pm is an inference process in the present embodiment, and the obtained inference probability Pm corresponds to the inference result. For example, assuming that the predicted diagnosis name Df is “pulmonary metastasis of cancer”, a set of Ef = {S11, S33, S41, S53, S71} and E1 = {S22} is input to the inference model of FIG. The inference probability belonging to “lung metastasis” is calculated to be 60.2%. Then, this inference probability is stored in association with E1 = {S22}.

  Subsequently, in step S408, the CPU 111 performs a process of adding 1 to the variable m.

  Subsequently, in step S409, the CPU 111 determines whether to repeatedly execute the processing from step S406 to step S408. Specifically, in step S409, the CPU 111 determines whether or not the value of the variable m is larger than the number of combinations that Em can take (hereinafter referred to as “CE”).

  If the value of the variable m is not greater than CE (ie, the value of the variable m is less than CE) as a result of the determination in step S409, not all Pm has been obtained, so the process returns to step S406 and again Then, the processing after step S406 is performed.

  On the other hand, if the value of the variable m is larger than CE as a result of the determination in step S409, the process proceeds to step S410. For example, in the example described above, the number of combinations that Em can take is CE = 3. Therefore, the processing from step S406 to step S409 is repeated until E3. Then, when m = 4, m becomes larger than CE, so the process proceeds to step S410.

  In step S410, the CPU 111 selects one of the Ems created and acquired in step S406 as the non-input information that is most likely to be presented to the doctor (presentation non-input information selection step).

  Specifically, Em that maximizes the inference probability Pm stored in the main memory 112 by the above processing is stored in the main memory 112 as uninput information Eh that most affirms the predicted diagnosis name. Further, the inference probability (hereinafter referred to as “Ph”) at this time is stored in the main memory 112 in association with the non-input information Eh. Similarly, Em that minimizes the inference probability Pm is stored in the main memory 112 as uninput information El that most denies the predicted diagnosis name. Further, the inference probability at this time (hereinafter referred to as “Pl”) is stored in the main memory 112 in association with the non-input information El.

  Table 1 shows examples of E1 to E3 and the inference probability P (Df | Ef, Em) in the above example. In this example, the inference probability is 71.6% at S23 in FIG. 3 at E2 at {not--not cut}, and the minimum at 44.1% at {bronch bronchial image-is} at S61 in FIG. 3 at E3. It becomes. Therefore, in this case, Eh = {S23} and El = {S61} are stored in the main memory 112.

When the process of step S410 ends, the process of the flowchart of FIG. 4 ends.
As described above, the process of step S205 shown in FIG. 2 is executed through the process of steps S401 to S410 of FIG.

Returning to the description of FIG.
When the process of step S205 ends, the process proceeds to step S206.
When the processing proceeds to step S206, the CPU 111 performs a process of displaying the uninputted information to be noted (presentation uninput information etc.) selected in step S205 on the monitor 120 and presenting it (presentation step). That is, the CPU 111 displays Eh (non-input information for affirming Df), El (non-input information for Negating Df), and the like stored in the main memory 112 on the monitor 120.

  FIG. 6 is a schematic diagram showing an example of a monitor display screen in the embodiment of the present invention. In the example shown in FIG. 6, the image to be interpreted 610 displayed in step S <b> 202 is shown on the monitor display screen 600. In addition, on the monitor display screen 600, the predicted diagnosis name Df (cancer lung transfer) acquired in step S401 is shown in the predicted diagnosis name 620, and the medical information (existing input information) Ef acquired in step S203 is already input information. 630. In addition, the monitor display screen 600 shows not-input information 640 to be noted. Specifically, “not-not-noted” of the uninput information Eh that affirms the predicted diagnosis name is shown as the not-entered information 641 to be noted, and the uninput to deny the predictive diagnosis name as the not-entered information 642 to be noted The information “El bronchial bronze image-is” is shown.

  By using the medical diagnosis support apparatus 100 according to the present embodiment, a doctor can obtain medical information to be preferentially viewed by referring to the presented non-input information, and information that assists diagnosis. Can be obtained.

  In the medical diagnosis support apparatus 100 according to the present embodiment, it is possible to present non-input information that is likely to exist in the image to be interpreted and that affects the predicted diagnosis name. Accordingly, medical information that is preferential and assists in diagnosis can be presented to a doctor or the like. Further, in the medical diagnosis support apparatus 100 according to the present embodiment, the image feature amount is used only for the purpose of excluding a state with a clearly low possibility from the presentation candidates. Therefore, unlike the process of automatically determining the findings from the image feature quantity, even if the detection accuracy of the image feature quantity and the reliability of the association with the state are not perfect, the diagnosis using the information effectively Support is feasible.

  As described above, according to the medical diagnosis support apparatus 100 according to the present embodiment, since uninput information that has a large influence on the inference result and is highly likely to exist is selected and presented, the doctor preferentially confirms it. Information to be presented can be efficiently presented.

<Modification>
In the embodiment described above, in step S203 of FIG. 2, the doctor interprets the medical image displayed by the medical diagnosis support apparatus 100 and acquires the findings as medical information. In the present invention, the method for acquiring medical information is not limited to this form. For example, medical information of medical examination data including past interpretation reports and medical records about the examinee (patient) and other information that can be used for diagnosis support processing is set as a processing target (existing input information / non-input information). You can also.

  Further, the image feature amount acquisition processing in step S204 of FIG. 2 may be a pre-process of step S203 of FIG. 2 or may be after step S402.

  In the above-described embodiment, description has been made assuming that all medical information takes discrete values. However, continuous values may be acquired as input information, and the values may be discretized.

  Further, in the above-described embodiment, regarding the findings in which the presence / absence determination is not performed based on the image feature amount in the process of step S403 in FIG. 4, all the states are elements of “Cif”. However, in this invention, it is not limited to this form, The structure of excluding all the states from "Cif" may be sufficient.

  Further, the presentation non-input information selected in step S205 in FIG. 2 is selected for the diagnosis name with the highest probability obtained in step S401 in FIG. However, the present invention is not limited to this form. For example, it is a configuration in which uninput information for which the probability of each diagnosis name (D1, D2, D3) is the highest (or lowest) is selected, and this is associated with each diagnosis name and presented. May be.

(Other embodiments)
The present invention can also be realized by executing the following processing.
That is, software (program) that realizes the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and a computer (or CPU, MPU, or the like) of the system or apparatus reads the program. It is a process to be executed.

10: Medical diagnosis support system, 100: Medical diagnosis support apparatus, 110: Control unit, 111: Central processing unit (CPU), 112: Main memory, 113: Magnetic disk, 114: Display memory, 115: Bus, 120: Monitor , 130: mouse, 140: keyboard, 200: medical image database, 300: diagnostic record database, 400: local area network (LAN)

Claims (1)

A medical diagnosis support apparatus that provides information for supporting medical diagnosis,
Medical image input means for inputting a medical image to be subjected to the medical diagnosis;
Medical information acquisition means for acquiring one or more pieces of medical information relating to the medical image as already input information;
Image feature amount acquisition means for acquiring an image feature amount from the medical image;
A presentation uninput information candidate for selecting a plurality of non-input information associated with the image feature amount as a presentation non-input information candidate that is a presentation candidate from non-input information that is medical information other than the already input information A selection means;
An inference means for obtaining an inference result using each of the already input information and the presented uninput information candidates;
Based on a plurality of inference results, a presentation non-input information selection means for selecting presentation non-input information from the presentation non-input information candidates,
A medical diagnosis support apparatus, comprising: presentation means for presenting presentation non-input information selected by the presentation non-input information selection means.

Images