JP6458166B2 - MEDICAL IMAGE PROCESSING METHOD, DEVICE, SYSTEM, AND PROGRAM - Google Patents

Description

  The present invention relates to a medical image processing method, apparatus, system, and program.

In medical image processing, for example, there is a demand for extracting (segmentation) information (blood vessels, nerves, cancer tissue, etc.) necessary for diagnosis from a medical image. Since the image contains a lot of unnecessary information, it is necessary to select information by extraction for accurate examination.
For example, Non-Patent Document 1 discloses a technique related to region extraction using luminance value continuity. This is called “Region growing” method. When the luminance values are similar between adjacent pixels, both the pixels are integrated and expanded assuming that they represent the same tissue.
Non-Patent Document 2 describes a technique related to local geometric analysis. The local geometric shape analysis is for determining a geometric shape (lumps, lines, and surfaces) for each pixel in an image, and an area can be emphasized and extracted using the relationship between eigenvalues.

Fabijanska, Anna. “Two-pass region growing algorithm for segmenting airway tree from MDCT chest scans.” Computerized Medical Imaging and Graphics 33.7 (2009): 537-5. Frangi, Alejandro Fetal. “Multiscale Vessel enhancement filtering”, “Medical Image Computing and Computer Assisted Interpretation-MICCAI'98, Springer Berlin Heidelberg (1998), pp. 130-137.

  In diagnosis and surgical examination using a medical image, it is necessary to extract necessary information from the image and scrutinize it. In cancer diagnosis using conventional segmentation, for example, a cancer shape is selected and a tumor is extracted. However, in general, there are various cancer shapes, and it is not always easy to deal with all cases using specific feature amounts. In addition, there is a demand for extracting anatomically correct linear tissues such as blood vessels. Conventionally, an extraction method that considers the continuity of pixel values by the Region Growing method has been proposed for this problem, as in Non-Patent Document 1, but for example, an organ having similar brightness is in contact with the subject. In this case, it is assumed that erroneous extraction of surrounding tissues occurs. In addition, as in Non-Patent Document 2, etc., information enhancement and extraction methods have been used using local geometric analysis, but extraction of intersections and branches of linear tissues such as blood vessels and bronchial branches is extracted. May be difficult and accuracy may not be sufficient.

  In view of the above points, an object of the present invention is to provide a medical image processing method, apparatus, system, and program for extracting a linear tissue probability image in consideration of spatial continuity.

According to the first solution of the present invention,
A medical image processing method comprising:
The processing unit obtains a probability P _f indicating whether the pixel value of each pixel seems to be linear based on a local shape analysis performed according to the pixel value for each pixel with respect to a three-dimensional medical image,
The processing unit refers to probability information that is stored in advance in the storage unit and stores probability information P _i that indicates whether the pixel is a target organ for which pixel value is specified for each organ identification information, and the pixel of each pixel Based on the value, a probability P _i indicating whether each pixel is a target organ is obtained,
The processing unit obtains the probability P _c by accumulating the probability that surrounding linear tissue pixels pass on the pixel to be processed based on the continuity of each pixel,
The processing unit obtains a linear tissue probability P indicating whether each pixel is linear by adding or weighting the product of the probability P _i and the probability P _c and the probability P _f ,
A medical image processing method is provided in which the processing unit generates a linear tissue probability image according to the linear tissue probability P, displays the linear tissue probability image on a display unit, and / or stores the linear tissue probability image in the storage unit.

According to the second solution of the present invention,
A medical image processing apparatus,
A storage unit that stores a captured three-dimensional medical image, probability information indicating a probability P _i indicating whether a pixel is designated with respect to a pixel value for each organ identification information, and a linear tissue probability image;
A display unit;
A processing unit;
With
The processing unit obtains a probability P _f indicating whether the pixel value of each pixel seems to be linear based on a local shape analysis performed according to the pixel value for each pixel with respect to a three-dimensional medical image,
The processing unit refers to the probability information stored in advance in the storage unit, and obtains a probability P _i indicating whether each pixel is a target organ based on a pixel value of each pixel,
The processing unit obtains the probability P _c by accumulating the probability that surrounding linear tissue pixels pass on the pixel to be processed based on the continuity of each pixel,
The processing unit obtains a linear tissue probability P indicating whether each pixel is linear by adding or weighting the product of the probability P _i and the probability P _c and the probability P _f ,
A medical image processing apparatus is provided in which the processing unit generates a linear tissue probability image according to the linear tissue probability P, displays the linear tissue probability image on a display unit, and / or stores the linear tissue probability image in the storage unit.

According to the third solution of the present invention,
A medical image processing system,
An image capturing device for capturing and acquiring a three-dimensional medical image;
A medical image processing apparatus as described above for inputting the medical image from the image capturing apparatus and generating a linear tissue probability image based on the medical image;
A medical image processing system is provided.

According to the fourth solution of the present invention,
A medical image processing program,
Obtaining a probability P _f indicating whether the pixel value of each pixel seems to be linear based on a local shape analysis performed according to the pixel value for each pixel with respect to a three-dimensional medical image by the processing unit;
The processing unit refers to probability information that is stored in advance in the storage unit and stores probability information P _i that indicates whether the pixel is a target organ for which pixel value is specified for each organ identification information, and the pixel of each pixel Obtaining a probability P _i representing whether each pixel is a target organ based on the value;
The processor determines a probability P _c by accumulating the probability that surrounding linear tissue pixels pass over the pixel to be processed based on the continuity of each pixel;
The processing unit obtains a linear tissue probability P representing whether each pixel seems to be linear by adding or weighting the product of the probability P _i and the probability P _c and the probability P _f. When,
The processing unit generates a linear tissue probability image according to the linear tissue probability P, and causes the computer to execute a step of displaying the linear tissue probability image on a display unit and / or storing it in the storage unit. A medical image processing program is provided.

  According to the present invention, it is possible to provide a medical image processing method, apparatus, system, and program for extracting a linear tissue probability image in consideration of spatial continuity.

1 is a configuration diagram of a medical image processing system. The flowchart (1) of medical image processing. Is a diagram showing the relationship between the probability P _i based on the pixel values (luminance values). Explanatory drawing regarding calculation of probability _Pc based on continuity. Explanatory drawing which shows the comparison of extraction of a linear tissue probability image. Flowchart (2) of medical image processing.

1. Outline The present invention and / or the present embodiment mainly realizes extraction of a topologically correct linear tissue image of a branching portion, which has been difficult in the past, with high accuracy. In particular, the present invention and / or the present embodiment calculates the linear tissue probability P with high accuracy by adding a branch region extraction method to the local shape analysis method.
The medical image processing apparatus calculates the linear tissue probability P as the sum of the linear region probability P _f based on the local shape analysis and the linear region probability P _b in the branch region as in the following equation. The linear region probability P _b is calculated based on the product of the extraction target organ probability P _i based on the pixel value and the probability P _c based on the geometric continuity of the surrounding linear tissue.

P = P _f + αP _b
= P _f + αP _c P _i
here,
P: Linear tissue probability P _f : Linear region probability based on local shape analysis P _b : Linear region probability based on branch region extraction method α: Weight parameter P _c : Probability based on continuity P _i : Based on luminance value probability

The above formula is obtained by adding a probability based on continuity (continuity probability) P _c and a probability based on luminance value (luminance value probability) to correspond to a branch region as an additional item to P _f obtained using local shape analysis. _{P b} = _{αP c P i} is added is a term consisting of the product of P _i.

Compared with the probability image created by the conventional method (local shape analysis), the linear tissue probability image created according to the present invention and / or the present embodiment can clearly extract the blood vessel bifurcation.
In addition, according to the present invention and / or the present embodiment, for example, cancer is included by extracting / removing a linear tissue probability image that is an organ having a stable feature amount from the original captured image. Regions can be extracted. Examples of the main stable feature amount include linear tissue (blood vessels, bronchi, nerves, etc.). Further, by extracting / removing a linear tissue probability image representing a linear structure from a medical image, it can be applied to extracting a disease (possibly) region. Although it is assumed that the conventional linear structure image extraction method is unstable, according to the present invention and / or the present embodiment, a stable linear structure image extraction method can be realized.

2. Medical image processing

FIG. 1 shows a configuration diagram of a medical image processing system.
This system includes a medical image processing apparatus 10 and an image photographing apparatus 20.
The medical image processing apparatus 10 includes a processing unit 1, an input unit 2, a display unit 3, an interface unit (I / F) 4, and a storage unit 5. The I / F 4 is an interface for connecting to the image capturing device 20 and inputting a captured image. The storage unit 5 includes an image file 51, a probability _Pi information file 52, and a linear tissue probability image file 53.
The image capturing device 20 captures 3D image information such as an organ. The medical image processing apparatus 10 stores the image information acquired from the image capturing apparatus 20 in the storage unit 5. The captured image is, for example, a three-dimensional medical image (mainly X-ray CT image, MRI, etc., and the imaging device is hereinafter referred to as modality).

  Here, a case where a luminance value is mainly used as a pixel value will be described as an example. However, the present invention is not limited thereto, and appropriate pixel information such as a value indicating a color may be used.

FIG. 2 shows a flowchart (1) of medical image processing.
Hereinafter, a procedure for medical image processing will be described with reference to a flowchart.
(Step S101)
The processing unit 1 acquires an image stored in the storage unit 5 (image file 51) in accordance with the designation by the input unit 2 or the like.
As an example, it is assumed that the processing unit 1 stores an image measured by the image capturing device 20 through the I / F 4 and stores the image in the storage unit 5 in advance for each predetermined ID that identifies the image. Can do. In this case, the processing unit 1 selects and reads the identification information and image of the organ to be processed from the storage unit 5 according to the ID specified by the input unit 1 or the like. Alternatively, when executing this step, the processing unit 1 may acquire the image measured by the image capturing device 20 via the I / F 4. Note that the processing unit 1 may omit the ID and do not perform the processing for each ID when the number of objects is one, or is designated in advance (the same applies to the following processing). Further, a processing specific area in the image may be set by the input unit 2 or the like.
(Step S103)
The processing unit 1 performs local shape analysis for each pixel in the image to be processed, and obtains P _f for each pixel (details will be described later).
(Step S105)
The processing unit 1 selects one pixel from the acquired image.
(Step S107)
The processing unit 1 acquires or calculates the probability P _i of the extraction target organ (blood vessel, bronchus, etc.) based on the pixel value for the pixel selected in step S105 (details will be described later). The extraction target organ can be determined in advance by the user using the input unit 1. The information for acquiring the probability P _i may be stored in advance in advance in the storage unit 5 (probability P _i information file 52) as the prior information corresponding to the identification information of the extraction-corresponding organ, or the operation It is also possible for the person to arbitrarily set by the input unit 2 or an external device.
(Step S109)
The processing unit 1 calculates a probability P _c based on the geometric continuity of the surrounding linear tissue for the pixel selected in step S105. The probability P _c based on continuity is based on, for example, the absolute value of the inner product of the eigenvector of the third principal component of the peripheral pixel and the direction vector from the peripheral pixel to the selected pixel, and the distance between the selected pixel and the peripheral pixel Alternatively, the linear tissue probability of peripheral pixels may be multiplied as a weight (details will be described later).
(Step S111)
Processing unit 1, as in the following equation, by multiplying the _{P c} obtained in a probability Pi and step S109 obtained in step S107, a linear region probability _{P b} in the branch region, give the probability _{P b} at step S103 A linear tissue probability P is obtained by adding to the obtained P _f . In addition, weighting may be performed with an appropriate predetermined weight parameter α at the time of addition.
(Step S113)
The processing unit 1 performs steps S105 to S111 for all the pixels to be processed.
(Step S115)
The processing unit 1 generates a linear tissue probability image for the linear tissue probability P obtained up to step S113, and displays it on the display unit 3 and / or stores it in the storage unit 5 (details will be described later).

Hereinafter, processing in each step will be described in detail.

[Step S103: Linear Region Probability P _f Based on Local Shape Analysis]
The processing unit 1 can determine the geometric shape (lump, line, surface) for each pixel and calculate the linear tissue probability by local geometric analysis. This analysis method emphasizes and extracts linear tissue pixels using the relationship between eigenvalues. The physical meanings of eigenvalues and eigenvectors represent the dispersion (change amount) of pixel values and their directions. For example, in the third eigenvector direction (the direction in which the eigenvalue is the smallest), the pixel value variation is small. As will be described in detail below, when the third eigenvalue is sufficiently smaller than the first and second eigenvalues, it is considered that pixel values similar to the third eigenvector direction are continuous. That is, it can be considered as a geometric object having a linear structure such as a blood vessel or a bronchus.

The local shape analysis is a method proposed by Frangi in Non-Patent Document 2 and is a method for evaluating the linear texture of each pixel of an image. This method uses an evaluation function that takes a pixel value to be evaluated and a local pixel group around it as an input and increases the value for a line. As will be described in detail below, the distribution of the pixel values around the pixel is analyzed by eigenvalues to calculate the distribution direction of the pixel values and the intensity thereof as eigenvalues and eigenvectors. Is sufficiently smaller than the first and second eigenvalues. The value obtained by the evaluation function becomes so large that the surrounding geometric shape seems to be a line, and this value is defined as a linear region probability P _f . Since this method means analyzing a local geometric shape around the pixel, it is called “local shape analysis” in this specification. Note that “linear” in the present invention and / or the present embodiment has a thickness, not a thickness of 0, as commonly used in this technical field. Including bar shape.

Hereinafter, a specific calculation method of the probability P _f based on the local geometric shape will be described.
It should be noted that, mainly in equations, scalar variables are written in italic characters and vector variables are written in bold characters. Further, the pixel value of the three-dimensional image is assumed to be I (x, y, z) or I (r). Here, (x, y, z) represents the coordinates of the image coordinate system. (Note that there are some places that cannot always be written because of restrictions such as fonts in application documents of the JPO.)

A geometric structure can be estimated by evaluating the amount of change in surrounding pixel values for a certain pixel. For example, when the pixel values are continuous in a certain direction and the pixel values are not continuous in the other direction, the probability of the linear structure is large.
Further, there is an eigenvalue of a pixel as a method for quantitatively representing the amount of change in the pixel value. The eigenvalue of the pixel is obtained by calculating the eigenvalue of a Hessian matrix (Hesse matrix) H expressed by the following equation.

一般に、固有値の幾何学的解釈は次の通りである。
・塊 ：｜λ_１｜＞｜λ_２｜＞｜λ_３｜＞＞０
・線状：｜λ_１｜＞｜λ_２｜＞＞｜λ_３｜≒０）
・面状：｜λ_１｜＞＞｜λ_２｜＞｜λ_３｜≒０）

In general, the geometric interpretation of eigenvalues is as follows.
Lump: | λ ₁ | >> | λ ₂ | >> | λ ₃ | >> 0
-Linear: | λ ₁ | >> | λ ₂ | >> | λ ₃ | ≈0)
・ Surface shape: | λ ₁ | >> | λ ₂ | >> | λ ₃ | ≈0)

As eigenvalues, λ _1, λ _{2, and} λ ₃ are obtained in descending order of absolute values (| λ ₁ |> | λ ₂ |> | λ ₃ |). At the same time, eigenvectors e _1, e _2, e ₃ corresponding to the respective eigenvalues are obtained. As an eigenvalue, for example, a negative value can be used according to the pixel value, but is not limited thereto.

For linear tissue probabilities, the interpretation of eigenvalues is as follows.
-The direction of λ ₁ (e ₁ ) has a large amount of change in pixel value, and the direction of λ ₃ (e ₃ ) is small.
-Since the amount of change is the absolute value of the eigenvalue λ, if λ ₁ and λ ₂ are sufficiently larger than λ ₃ , the probability of a pixel having a linear structure is high (| λ ₁ | >> | λ ₂ | >> | λ ₃ | ≈0). Also, the direction of λ ₃ (e ₃ ) is the direction of the linear structure.

Based on the above, the linear tissue probability P _f is calculated from the eigenvalue as in the following equation (see Non-Patent Document 2). The meaning of the expression is that “in the case of a linear structure, it is not a planar shape in terms of eigenvalues but seems to be linear, and surrounding pixels are not uniform and change”. As a result, V (r) in the following equation is multiplied by three functions at the underlined portion.

[Step S107 Probability P _i Based on Pixel Value (Luminance Value)]
Figure 3 is a diagram showing the relationship between the probability P _i based on the pixel values (luminance values).
As an example, this figure shows the relationship of the probability P _i based on the luminance value I of the blood vessel with respect to the lung field image. For example, stored in advance in the storage unit 5 the information of the diagram showing the relationship (probability P _i information file 52), the processing unit 1, see probability P _i information stored in the probability P _i information file 52 Thus, the probability P _i can be obtained according to the pixel value (luminance value).
When a certain organ is imaged, the pixel values indicated by the organ for each modality are approximately the same. However, pixels having the same value are not necessarily the same organ. For example, a CT image has a low contrast of soft tissue and is photographed with similar pixel values even though they are different organs. Even when different organs are imaged with the same pixel value, for example, if the imaging region is fixed in the hospital imaging flow, the probability that each pixel value is the target organ can be estimated roughly (eg, the same if it is the same) Even if there are 100 pixels having a value, if it is known that 30 of them are target organs, the probability that the pixel value is the target organ can be estimated to be 0.3.) . Thus, with the pixel value, it is an organ to be extracted, that the probability that, is defined as P _i "extraction target organ probability based on the pixel value". In this way, the probability P _i based on the luminance value is calculated from the linear tissue pixel value occupied in the luminance value distribution (histogram) of the image, assuming that there is an inherent probability as organ information such as a blood vessel in advance for each luminance value. be able to.

The probability P _i information described above may be applied as a template stored in advance when this processing is executed, or a user such as a doctor may specify on the pixel histogram for each application target image. You may make it do.

Next, a description will be given of a method to create a template of P _i.
As an example, the probability of the vascular tissue in the lung field image can be obtained as follows.

In the above formula, at the stage of making the template, both denominator numerators are measured values. The denominator is the pixel value group of the whole organ such as the lung, and the numerator is the pixel value group of the blood vessel region, for example. The probability P _i information calculated in this way can be stored in advance in the probability P _i information file 52, and the processing unit 1 can refer to it.

[Step S109: Probability P _c Based on Continuity]
The branching means that the lines intersect with each other, and is a place where the lines extending from the surroundings intersect. That is, as the number of intersecting lines extending from surrounding pixels increases, the probability of branching increases. The processing unit 1 calculates the probability P _c based on continuity by accumulating the probability that the linear information of peripheral pixels passes over the target pixel, assuming that the branch region is a region where a plurality of linear tissues overlap. Since this utilizes the fact that geometric information is continuous, it is defined as “probability based on geometric continuity of surrounding linear tissue” P _c .

FIG. 4 is an explanatory diagram regarding the calculation of the probability P _c based on continuity. Hereinafter, a method for calculating the probability P _c will be described with reference to specific figures.
The processing unit 1 can obtain the probability P _c by the following equation.

This will be described below with reference to the drawings. First, one target pixel and surrounding pixels are selected (designated). In the figure, the target pixel is ri and one surrounding pixel is rj. Then perform local shape analysis in rj, retrieve the third eigenvector direction e _3. This direction is the line direction when rj is a pixel in a linear region. In the above equation, it is evaluated whether e ₃ extends in the direction of ri in the “direction identity term”. This portion can be evaluated, for example, by simply taking the inner product in the direction, and if the absolute value is close to 1, it is in the same direction, and if it is close to 0, it is not in the same direction. In the expression shown in the example, the weight is given using a Gaussian function. Furthermore, a distance term whose value decreases according to the distance is included. A sigmoid function is used as an example. P _line is obtained by multiplying the values obtained above.
Further, the probability P _f as the line-like term of the peripheral pixels is multiplied. As described above, the processing unit 1 calculates the probability P _f in the local shape analysis method.

When the above is added to all the peripheral pixels of ri, the degree to which a line extends from the peripheral pixels to ri can be evaluated. The value obtained in this way is defined as the probability P _c based on continuity.

[Step S111: Linear tissue probability P]
In summary, the processing unit 1 calculates the linear tissue probability P by adding the branch region extraction method to the local shape analysis method as in the following equation.

[Step S115: Linear tissue probability image]

The linear tissue probability P is calculated for all the pixels to be processed, and the probability image itself can be acquired as a linear tissue probability image such as a blood vessel structure image. In the obtained probability image, the value of the portion that seems to be a linear tissue is large, and the processing unit 1 forms the linear tissue probability image by binarizing the linear region by, for example, specifying a threshold value in advance by the user. Can be generated as a multi-value with a plurality of threshold values, or can be generated as a probability image of the linear tissue probability P itself. The processing unit 1 displays the linear tissue probability image on the display unit 3 and / or stores it in the storage unit 5.

  As described above, according to the present embodiment, the branch region can be extracted, so that the linear tissue pixels can be extracted in an anatomically correct form. This makes it possible to emphasize and extract the linear tissue pixels more accurately during diagnosis and surgical examination, and it is expected that the diagnosis and examination accuracy will be improved.

FIG. 5 is an explanatory diagram showing a comparison of extraction of linear tissue probability images.
This is an example of pulmonary blood vessels as an example. In the figure, linear tissue probability images are respectively shown by manual segmentation by a doctor, probability P _f , probability P _C by local shape analysis, and linear tissue probability P according to the present invention and / or the embodiment of the present invention. .
Compared with the probability image created by the conventional method (local shape analysis), the linear tissue probability image created according to the present invention and / or the present embodiment can extract the blood vessel bifurcation.
In addition, as a result of setting a threshold value for the linear tissue probability image and extracting the blood vessel region, it was also found that it was closer to a manual extraction result by a doctor than the conventional method.
As described above, by evaluating the continuity of the geometric information, it is possible to more clearly extract the branch region, which has been difficult in the past.

(Application to time change)
The processing unit 1 may further acquire and form the linear tissue probability image of the above-described embodiment at predetermined time intervals. And the process part 1 memorize | stores these information in the memory | storage part 5 for every ID, and can display a time change with a moving image on the display part 3, or can display with several continuous still images.

3. Application to Lung Field Information and Blood Vessel Structure According to the present embodiment, a method for extracting an image region such as a cancer region (such as a lung) whose shape is unstable is proposed. For example, by extracting and removing an image region such as an organ (bronchi, blood vessel, etc.) having a stable shape, the remaining region can be extracted as a pixel region such as a cancer candidate. In order to realize this extraction method, a linear tissue probability calculation method as in the present invention and / or the present embodiment is provided. The linear tissue probability image extracted using this is realized by, for example, evaluating the continuity of the geometric information, and the image region of the bifurcation region, which has been difficult in the past, can be extracted.
In addition, although the image about the lung was used here as an example, this invention and / or this Embodiment are not restricted to this, It can apply to a heart, a brain, various organs, etc., and cancer. Not only the area but also an appropriate area can be designated and extracted.

FIG. 6 shows a flowchart (2) of the medical image processing.
(Step S201)
The processing unit 1 applies the above-described medical image processing to the input lung image, obtains a linear tissue probability P, obtains a linear tissue probability image, and performs linear organ extraction. In this example, the processing unit 1 extracts a linear tissue probability image P for lung bronchi and / or blood vessels.
(Step S202)
The processing unit 1 extracts a cancer region by removing the linear tissue probability image (linear organ region) obtained in step S201 from the input image. The processing unit 1 displays the extracted cancer area on the display unit 3 and / or stores it in the storage unit 5.

In recent years, diagnosis and surgical examination using medical images are beginning to be considered as indispensable processes. Conventionally, however, it has not always been easy to accurately extract linear organ information from an image. Therefore, these treatments have a large time cost in daily clinical practice. Since this problem can be improved by the present invention, there is a high need in the field and the industrial utility value is high.
As an application of the present invention, for example, medical image segmentation is expected to be used for diagnosis and surgical examination, for example. In addition, with regard to the extraction of linear information from other images, for example, it can be applied to feature analysis of actual images, geological surveys, marine surveys, and others.

  The medical image processing method or medical image processing apparatus / system of the present invention includes a medical image processing program for causing a computer to execute each procedure, a computer-readable recording medium storing the medical image processing program, and a medical image processing program. It can be provided by a program product that can be loaded into the internal memory of a computer including a computer, a computer such as a server that includes the program, and the like.

DESCRIPTION OF SYMBOLS 10 Medical image processing apparatus 20 Image imaging device

Claims (11)

A medical image processing method comprising:
The processing unit obtains a probability P _f indicating whether the pixel value of each pixel seems to be linear based on a local shape analysis performed according to the pixel value for each pixel with respect to a three-dimensional medical image,
The processing unit refers to probability information that is stored in advance in the storage unit and stores probability information P _i that indicates whether the pixel is a target organ for which pixel value is specified for each organ identification information, and the pixel of each pixel Based on the value, a probability P _i indicating whether each pixel is a target organ is obtained,
The processing unit obtains the probability P _c by accumulating the probability that surrounding linear tissue pixels pass on the pixel to be processed based on the continuity of each pixel,
The processing unit obtains a linear tissue probability P indicating whether each pixel is linear by adding or weighting the product of the probability P _i and the probability P _c and the probability P _f ,
A medical image processing method in which the processing unit generates a linear tissue probability image according to a linear tissue probability P, displays the linear tissue probability image on a display unit, and / or stores the linear tissue probability image in the storage unit.
The medical image processing method according to claim 1,
The processing unit obtains a plurality of eigenvalues and eigenvectors for each eigenvalue from the Hessian matrix of the pixel values of each pixel, and each pixel according to a predetermined function that increases the value if the pixel is a linear tissue pixel based on the plurality of eigenvalues. the medical image processing method characterized by calculating the probability P _f.

The medical image processing method according to claim 1 or 2,
The storage unit stores in advance, as the probability P _i of the probability information, a ratio of the probability density based on the pixel value of the linear tissue pixel to the probability density based on the pixel value of the whole organ, which is obtained based on the image captured in advance. A medical image processing method.
The medical image processing method according to any one of claims 1 to 3,
The processor is
Specify the target pixel and surrounding pixels for processing,
The line-likeness term of the target pixel is obtained by the identity term indicating the degree of the direction of the line of the surrounding pixel toward the direction of the target pixel and the distance term whose value decreases according to the distance between the target pixel and the surrounding pixel. ,
By accumulating the product of the line-likeness term of the peripheral pixel obtained based on the local shape analysis and the line-likeness term of the target pixel for a plurality of peripheral pixels, the degree of the line extending from the peripheral pixel to the target pixel A medical image processing method characterized by obtaining a certain probability P _c based on continuity.
The medical image processing method according to any one of claims 1 to 4,
The processing unit generates a linear tissue probability image by binarization or multi-leveling with one or a plurality of predetermined thresholds with respect to the linear tissue probability P, or the linear tissue probability of each pixel A medical image processing method, wherein a linear tissue probability image is generated based on a distribution of P.
A linear tissue probability image is generated by the medical image processing method according to any one of claims 1 to 5,
A cancer region or other target image is extracted by removing the linear tissue probability image from the medical image, and the cancer region or the other target image is displayed on a display unit and / or the storage unit A medical image processing method characterized by storing the data in a memory.
The medical image processing method according to any one of claims 1 to 6,
The processing unit further acquires and forms the medical image and / or the linear tissue probability image at predetermined time intervals or times, stores them in the storage unit, and / or the display unit. A medical image processing method characterized in that a time change is displayed as a moving image or a plurality of continuous still images.
The medical image processing method according to any one of claims 1 to 7,
2. The medical image processing method according to claim 1, wherein each of the linear tissue probability images is a linear tissue probability image representing one or more of a blood vessel shape, a bronchial shape, a lymph vessel shape, and a nerve shape.
A medical image processing apparatus,
A storage unit that stores a captured three-dimensional medical image, probability information indicating a probability P _i indicating whether a pixel is designated with respect to a pixel value for each organ identification information, and a linear tissue probability image;
A display unit;
A processing unit;
With
The processing unit obtains a probability P _f indicating whether the pixel value of each pixel seems to be linear based on a local shape analysis performed according to the pixel value for each pixel with respect to a three-dimensional medical image,
The processing unit refers to the probability information stored in advance in the storage unit, and obtains a probability P _i indicating whether each pixel is a target organ based on a pixel value of each pixel,
The processing unit obtains the probability P _c by accumulating the probability that surrounding linear tissue pixels pass on the pixel to be processed based on the continuity of each pixel,
The processing unit obtains a linear tissue probability P indicating whether each pixel is linear by adding or weighting the product of the probability P _i and the probability P _c and the probability P _f ,
A medical image processing apparatus in which the processing unit generates a linear tissue probability image according to a linear tissue probability P, displays the linear tissue probability image on a display unit, and / or stores the linear tissue probability image in the storage unit.
A medical image processing system,
An image capturing device for capturing and acquiring a three-dimensional medical image;
The medical image processing apparatus according to claim 9, wherein the medical image is input from the image capturing apparatus and a linear tissue probability image is generated based on the medical image;
A medical image processing system.
A medical image processing program,
Obtaining a probability P _f indicating whether the pixel value of each pixel seems to be linear based on a local shape analysis performed according to the pixel value for each pixel with respect to a three-dimensional medical image by the processing unit;
The processing unit refers to probability information that is stored in advance in the storage unit and stores probability information P _i that indicates whether the pixel is a target organ for which pixel value is specified for each organ identification information, and the pixel of each pixel Obtaining a probability P _i representing whether each pixel is a target organ based on the value;
The processor determines a probability P _c by accumulating the probability that surrounding linear tissue pixels pass over the pixel to be processed based on the continuity of each pixel;
The processing unit obtains a linear tissue probability P representing whether each pixel seems to be linear by adding or weighting the product of the probability P _i and the probability P _c and the probability P _f. When,
The processing unit generates a linear tissue probability image according to the linear tissue probability P, and causes the computer to execute a step of displaying the linear tissue probability image on a display unit and / or storing it in the storage unit. Medical image processing program.

Images