JP2011250811A - Medical image processing apparatus and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a lesion candidate using the feature amount of the shape of an image.SOLUTION: A medical image processing apparatus includes: a candidate determining means for creating a similarity map showing the similarity between a region of interest and a specific shape for every region of interest set in the medical image for one or more different shapes by using data relating to the medical image (step S4) and detecting the region of interest as the region of a lesion candidate when the lesion candidate is determined to be included in the region of interest on the basis of the result of analysis of the feature amount of each similarity map created for the one or more different shapes (step S8); a display means; and a control means for displaying the detection result of the lesion candidate on the display means (step S10).

Description

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

  2. Description of the Related Art Conventionally, a medical image processing apparatus that detects an image region having a high possibility of a lesion as a lesion candidate by analyzing a medical image has been developed. Such a medical image processing apparatus is called CAD (Computer-Aided Diagnosis). According to CAD, the detection result of a lesion candidate can be provided to a doctor as objective judgment material, and the burden of interpretation work of the doctor can be reduced.

For medical images obtained by imaging the chest, many methods for detecting lung diseases called interstitial diseases have been disclosed (see, for example, Patent Documents 1, 2, and 3). An image region where an interstitial disease has occurred is generally characterized by a ground glass-like shadow on an X-ray image, specifically, a shadow that is uniformly white. In CAD, an image region having such characteristics is detected by image analysis. For example, in addition to a method of detecting a lesion candidate using a feature amount by a Fourier spectrum or a texture feature amount such as a density co-occurrence matrix, a feature amount analysis is performed by an artificial neural network as described in Patent Document 2 above. There is a method of detecting a lesion candidate of an interstitial disease.
JP-A-6-237925 JP-A-10-171910 US Pat. No. 5,319,549

  However, detection by focusing on the shadow of the lesion, that is, the shape (morphology) of the image has not been conventionally performed. The shape of the image is one of the characteristics for determining whether or not the disease is a true disease, and is considered useful for detecting a lesion candidate.

  The subject of this invention is detecting a lesion candidate using the feature-value of the shape of an image.

According to the invention of claim 1,
Using the medical image data, for each region of interest set in the medical image, create a similarity map indicating the similarity between the region of interest and a specific shape for one or a plurality of different shapes, When it is determined that a candidate lesion is included in the region of interest based on the result of analyzing the feature amount of each similarity map created for a plurality of different shapes, the region of interest is detected as a candidate lesion region. Candidate detection means to
Display means;
Control means for displaying the detection results of the lesion candidates on the display means;
A medical image processing apparatus is provided.

According to invention of Claim 2,
The candidate detecting means calculates a feature amount of the similarity map only for an image area in which the similarity is a certain value or more in each similarity map created for the one or a plurality of different shapes in the region of interest. The medical image processing apparatus according to claim 1 to be calculated is provided.

According to invention of Claim 3,
The candidate detection means obtains an isolated region in which pixels having a certain range of similarities are adjacent to each other in each of the similarity maps created for the one or a plurality of different shapes, and calculates a feature amount of the isolated region The medical image processing apparatus according to claim 2, wherein the feature amount analysis uses a feature amount of the similarity map and a feature amount of the isolated region.

According to invention of Claim 4,
The candidate detecting means creates a similarity map indicating the similarity to the shape corresponding to the normal tissue structure and a similarity map indicating the similarity to the shape corresponding to the lesion, respectively, and corresponds to the normal tissue structure The medical image processing apparatus according to any one of claims 1 to 3, wherein a feature amount of each similarity map created for a shape to be processed and a shape corresponding to a lesion is calculated and used for the feature amount analysis. .

According to the invention of claim 5,
Computer
Using the medical image data, for each region of interest set in the medical image, create a similarity map indicating the similarity between the region of interest and a specific shape for one or a plurality of different shapes, When it is determined that a candidate lesion is included in the region of interest based on the result of analyzing the feature amount of each similarity map created for a plurality of different shapes, the region of interest is detected as a candidate lesion region. Candidate detection means to
Control means for displaying the detection result of the lesion candidate on a display means;
A program for functioning as a server is provided.

  According to the present invention, it is possible to detect a lesion candidate from a medical image using the feature amount of the shape of the image.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 shows a functional configuration of a medical image processing apparatus 1 according to the present embodiment.
As shown in FIG. 1, the medical image processing apparatus 1 includes a control unit 11, an operation unit 12, a display unit 13, a communication unit 14, a storage unit 15, and a candidate detection unit 16.

The control unit 11 is a control unit configured to include a CPU (Central Processing Unit), a RAM (Random Access Memory), and the like. The control unit 11 executes various processes in cooperation with a program stored in the storage unit 15. In the processing, the control unit 11 performs various calculations and centrally controls the operation of each unit.
For example, when a detection result of a lesion candidate detected from a medical image is input from the candidate detection unit 16, the control unit 11 causes the display unit 13 to display a display screen indicating the detection result.

The operation unit 12 includes a keyboard, a mouse, a touch panel configured integrally with the display unit 13, etc., and generates operation signals corresponding to these operations and outputs them to the control unit 11.
The display unit 13 is a display unit having a display or the like, and displays medical images, detection results of lesion candidates, and the like according to display control of the control unit 11.

  The communication unit 14 includes a communication interface and communicates with an external device connected to the network. For example, the communication unit 14 receives a medical image that is a detection target of a lesion candidate from an imaging device that captures a medical image. In addition, the communication unit 14 transmits information on detection results of lesion candidates for the medical image together with the medical image to a server that manages the medical image.

  The storage unit 15 includes a memory such as a hard disk and stores programs, parameters, and the like. In addition, the storage unit 15 stores medical images to be detected as lesion candidates.

  The candidate detection unit 16 is candidate detection means for detecting a lesion candidate region included in the medical image using the medical image data. Information of the detection result is output to the control unit 11. The lesion candidate detection process can be realized by storing a detection processing program in the storage unit 15 and performing software processing in cooperation with the stored program and the CPU. Alternatively, a hardware circuit for detection processing may be constructed and realized by the hardware circuit.

Next, the operation of the medical image processing apparatus 1 will be described.
FIG. 2 is a flowchart showing processing executed by the medical image processing apparatus 1. In this process, lesion candidates for interstitial diseases that are lung diseases are detected for medical images obtained by X-ray imaging of the chest.
In the medical image processing apparatus 1, first, the candidate detection unit 16 performs lesion candidate detection processing. As shown in FIG. 2, the candidate detection unit 16 performs image analysis of the medical image and extracts a lung field region included in the medical image (step S1).

  The method for extracting the lung field region is not particularly limited, and a known method can be applied. For example, as disclosed in Japanese Patent No. 2998733, in the X-ray image, in the lung field region, the image density of the left and right lung portions is higher than the surrounding density. Therefore, the candidate detection unit 16 may create a density histogram, determine the image portion of the high density region corresponding to the lung field region from the shape and area of the density histogram, and extract the image portion as the lung field region. Alternatively, as disclosed in Japanese Patent Laid-Open No. 2003-6661, the candidate detection unit 16 detects the contour of the lung field region by performing template matching using a template that defines the contour of the standard lung field region. It is also possible to extract a lung field region within the contour.

When the lung field region is extracted, the candidate detection unit 16 performs image analysis of the medical image and extracts a rib region included in the medical image (step S2).
The method for extracting the rib region is not particularly limited, and a known method can be applied. For example, as disclosed in Japanese Patent Laid-Open No. 5-176919, the candidate detection unit 16 defines a number of contour lines in the vertical direction (head-to-foot direction) in the lung field region of the X-ray image. By applying a model function determined in advance, the contour portion of the rib region is estimated. Then, the candidate detection unit 16 determines a plurality of image regions to be processed in the estimated contour portion, obtains the gradient of each pixel in the image region and the direction corresponding to the gradient by the Sobel operator, The maximum value is set as the gradient of the image area. When the gradient and orientation of each image area are plotted as coordinates in the coordinate space, a set is created for each image area having the same gradient characteristics in the coordinate space, such as the upper and lower edges of the rib, the inside edge of the rib, and other than the rib It is formed. Therefore, the candidate detection unit 16 extracts an image region classified into the upper edge and the lower edge of the rib in the coordinate space as a rib area outline, and extracts an area surrounded by the outline as a rib area.

  Next, the candidate detection unit 16 sets one or a plurality of ROIs (Region Of Interest) for a region in the medical image that is a lung field region excluding the rib region (step S3). The ROI is an image area for which a similarity map, which will be described later, is to be created. For example, as shown in FIG. 3, several m × m pixel ROIs are set in the lung field region between the ribs. What is necessary is just to set the size of ROI suitably.

Next, the candidate detection unit 16 performs the following processing for any one of the set one or more ROIs.
The candidate detection unit 16 creates a similarity map for the ROI to be processed (step S4). The similarity map is a map showing the similarity between the ROI and a specific shape, and the similarity between each pixel and the specific shape is set for each pixel in the ROI.
The process for creating the similarity map will be described with reference to FIG.

  As shown in FIG. 4, the candidate detection unit 16 sets a specific shape J (step S41). When a similarity map is created for only one shape, the shape J may be selected as a shape that easily causes a lesion to be detected. In the case of an interstitial disease, for example, a shape such as a circle or a line may be mentioned, but a shape such as a star or an ellipse may be selected depending on the situation.

  Next, the candidate detecting unit 16 calculates the similarity between the set shape J and a certain pixel in the ROI (this is referred to as a target pixel) (step S42). As the degree of similarity, for example, a cross-correlation value or Euclidean distance can be used, but any index may be used as long as it is an index indicating similarity to the shape J.

上記式において、ｆ（ｎ）は特定の形状Ｊからなる画像を示す関数、ｇ（ｍ）は処理するＲＯＩの画像を示す関数である。ｎは特定の形状Ｊを示す画像の画素、ｍは処理対象とするＲＯＩ内の注目画素を示している。
上記式により求められる相互相関値は−１〜＋１の値域を有し、類似性が低いほど−１に近づき、類似性が高いほど＋１の値に近づく。 Here, the cross-correlation value is obtained by the following equation as the similarity.

In the above equation, f (n) is a function indicating an image having a specific shape J, and g (m) is a function indicating an ROI image to be processed. n is a pixel of an image showing a specific shape J, and m is a pixel of interest in the ROI to be processed.
The cross-correlation value obtained by the above formula has a range of −1 to +1, and approaches −1 as the similarity is lower and approaches +1 as the similarity is higher.

  When the similarity is calculated for the target pixel, the candidate detection unit 16 determines whether or not the processing for calculating the similarity is completed for all the pixels in the ROI (step S43). If there is an unprocessed pixel (step S43; N), the candidate detection unit 16 sets the unprocessed pixel as a target pixel (step S44), returns to step S42, and sets the similarity for the newly set target pixel. Repeat the calculation process.

Then, when the similarity is calculated for all the pixels in the ROI (step S43; Y), the candidate detection unit 16 sets the similarity calculated for each pixel for each pixel in the ROI. A degree map is created (step S45).
In the present embodiment, the case of creating a similarity map for one shape has been described. However, when creating a similarity map for a plurality of different shapes as well as for one shape, the shape is changed to the above-described one. What is necessary is just to repeat a process. Thereby, a similarity map can be created for each of a plurality of different shapes.

When the similarity map is created, the process proceeds to step S5 shown in FIG.
The candidate detection unit 16 creates a histogram of similarity using the similarity map created for the ROI to be processed. Then, the candidate detection unit 16 extracts, from the processing target ROI, an upper a% image area in the created histogram, that is, an image area whose similarity is equal to or greater than a certain value (step S5). For example, when a histogram of similarity as shown in FIG. 5 is obtained for a certain ROI, if a = 10, the similarity corresponding to 10% from the top of the histogram shown in FIG. It is. Therefore, pixels with a similarity of 0.9 or higher in the ROI are extracted as the top 10% or higher image areas.

上記式において、ｉは類似度マップの各画素を示し、ｉ_maxは類似度マップの最後に処理する画素、ｉ_minは最初に処理する画素を示している。また、Ｐ（ｉ）は類似度マップの各画素に対応する類似度を示している。
算出された特徴量は、後段の特徴量解析に用いられる。 Next, the candidate detection unit 16 calculates the feature amount of the similarity map for only the extracted image region (step S6). Here, examples of the feature amount of the similarity map include an average value, contrast, variance, standard deviation, skew, and the like represented by the following formula. Note that the feature amount of contrast indicates whether the distribution of the similarity in the similarity map for the extracted image region is spread flat or biased. Further, the skew is a feature amount indicating how much the distribution shape of the similarity in the similarity map for the extracted image region is deviated from a symmetrical shape as viewed from the average.

In the above equation, i represents each pixel of the similarity map, i _max represents a pixel to be processed at the end of the similarity map, and i _min represents a pixel to be processed first. P (i) indicates the similarity corresponding to each pixel of the similarity map.
The calculated feature value is used for the subsequent feature value analysis.

  Next, the candidate detection unit 16 obtains an image region (hereinafter referred to as an isolated region) in which pixels having a certain range of similarities are adjacent to each other in the similarity map created for the ROI to be processed, and features of the isolated region The amount is calculated (step S7). When an image forming a certain shape such as a circular image or a linear image is included in the ROI, it is considered that a similar degree of similarity is calculated for each pixel of the image. For example, in the similarity map in which the similarity with the circular shape is calculated, a relatively high similarity such as 0.8 or 0.9 is calculated for each pixel of the circular image. On the other hand, for each pixel of the linear image, a low similarity such as 0.3 or 0.4 is calculated. That is, it can be estimated that an isolated region where pixels having similar degrees of similarity calculated are adjacent is an image region having some shape.

  In addition, when an interstitial disease has occurred, the medical image includes many circular images often found in the interstitial disease, and all these circular images should have the same area. Therefore, the candidate detection unit 16 uses the number and area of image regions in which pixels with high similarity are adjacent in such a circular image, that is, the similarity map created for the circular shape, as the feature amount of the isolated region. calculate.

As a certain range of similarity that is determined as an isolated region, for example, it may be defined within the upper 5% in the similarity histogram, or may be defined by a numerical value within a range of similarity of 0.9 to 1. .
In the case of the upper 5% range, the candidate detection unit 16 creates a similarity histogram using the similarity map, and extracts pixels corresponding to the upper 5% in the histogram from the ROI pixels. Next, the candidate detection unit 16 refers to an adjacent pixel with respect to a certain pixel among the extracted pixels, and if the adjacent pixel is also an extracted pixel, it is the same as each pixel as a pixel constituting the same isolated region. An identification number is assigned. This identification number is a number for identifying an isolated region, and the same identification number is assigned to pixels in the same isolated region.

  If there is an extracted pixel in the adjacent pixel, the candidate detection unit 16 repeatedly performs the above processing for the adjacent pixel. Then, when there is no extracted pixel in the adjacent pixels, the above process is newly started for all the extracted pixels except for the pixels already assigned with the identification number. In this newly started process, an identification number different from the identification number already assigned is assigned. When identification numbers are assigned to all the pixels extracted in this way, as shown in FIG. 6, different identification is made for each isolated region in which pixels within a certain range having a similarity of the top 5% are adjacent. A number will be assigned.

  Next, the candidate detection unit 16 calculates the number of isolated regions, the area of each isolated region, the average value of the area, the standard deviation, and the like as the feature amount of the isolated region. These calculated feature values are used for subsequent feature amount analysis.

Next, the candidate detection unit 16 performs feature amount analysis on the feature amount calculated in steps S6 and S7, and based on the analysis result, the likelihood of ROI lesion, that is, the possibility that the ROI includes an image of an interstitial disease Judging.
As a technique of feature quantity analysis, for example, any technique such as LDA (Liner Discriminat Analysis), QDA (Quantitative DA), ANN (Artificial Neural Network), SVM (Support Vector Machine), AdaBoost, etc. May be applied.

  If it is LDA, the candidate detection unit 16 uses, as learning data, a feature amount obtained in advance for a medical image that is known to be normal (no lesion) or abnormal (with a lesion) as a learning data, A linear discriminant for discriminating the feature amount group related to is calculated. The candidate detection unit 16 determines, based on this linear discriminant, whether the feature quantity obtained for the ROI to be processed belongs to a normal feature quantity group or an abnormal feature quantity group. If the feature amount obtained for the ROI to be processed belongs to the feature amount group related to the abnormality, the candidate detection unit 16 determines that the ROI is likely to include an image of an interstitial disease, It is detected as a candidate region for a qualitative disease (step S8).

  Thus, when the process for one ROI to be processed is completed, the candidate detection unit 16 determines whether the process has been completed for all the set ROIs (step S9). If not completed (step S9; N), the process returns to the process of step S4, the unprocessed ROI is set as the processing target, and the candidate detection unit 16 repeats the processes of steps S4 to S8. When the processing is completed for all the ROIs (step S9; Y), the candidate detection unit 16 outputs information on detection results of lesion candidates for each ROI to the control unit 11. The control unit 11 displays the detection result of the lesion candidate on the display unit 13 (step S10).

FIG. 7 shows a display example.
The control unit 11 displays a medical image as shown in FIG. 7 on the display unit 13 as the detection result of the interstitial disease lesion candidate, and indicates the position of the ROI detected as the lesion candidate region in this medical image. Displays an arrow marker. When a plurality of adjacent ROIs are detected as candidates, a marker may be displayed so as to indicate the center position of these ROIs.
Note that the detection result display method is not limited to the method of indicating the position of a candidate by a marker, but may be other methods such as adding red to an ROI detected as a lesion candidate.

  As described above, according to the present embodiment, for each ROI set for the medical image, the candidate detection unit 16 uses the medical image data to calculate a similarity map between the ROI and a specific shape. Create for one shape. Further, the candidate detection unit 16 calculates the feature amount of the similarity map of each ROI created for one shape. Then, the candidate detection unit 16 detects, as a lesion candidate area, an ROI that is determined to include a lesion candidate based on an analysis result obtained by analyzing the calculated feature value.

  Thereby, a lesion candidate can be detected using the feature amount of the shape of the image included in the medical image. Since the detection result is displayed on the display unit 13 by the display control of the control unit 11, information can be provided to the doctor and the doctor's interpretation work can be supported.

  In addition, the candidate detection unit 16 calculates the feature amount of the similarity map as only one feature amount used for the feature amount analysis for only an image region in which the similarity degree is a certain value or more in the similarity map. As a result, feature quantities can be calculated only for image areas whose similarity is a certain value or higher and the similarity to a specific shape is high, and it is possible to accurately determine whether or not the lesion is a feature quantity analysis. Can do.

  In addition, the candidate detection unit 16 obtains an isolated region in which pixels having a certain degree of similarity are adjacent in the similarity map as one of the feature amounts used for feature amount analysis, and calculates the feature amount of the isolated region. . An isolated region having the same degree of similarity has some shape, and it is considered that the number and size of the isolated regions vary depending on the presence or absence of a lesion. Therefore, by using the feature amount of the isolated region for detection of the lesion candidate, multifaceted detection can be performed, and improvement in detection accuracy is expected.

  Further, since the candidate detection unit 16 sets the ROI in the lung field region and excludes the rib region, the image region for creating the similarity map can be narrowed down, and the processing efficiency can be improved.

In addition, the said embodiment is a suitable example of this invention, and is not limited to this.
For example, in the above-described embodiment, the example in which the feature amount analysis is performed using the similarity map created for one shape has been described. However, the similarity map may be created for a plurality of different shapes. The plurality of different shapes can all be shapes commonly found in lesions. For example, in the case of an interstitial disease, the candidate detection unit 16 can create a similarity map for a circle and an ellipse, and calculate the above-described feature amount using each similarity map.

The plurality of different shapes may be shapes often found in normal tissue structures in addition to shapes often found in lesions. Normal tissue structure refers to an anatomical structure free from lesions such as lungs, heart and blood vessels. Processing in the case of a shape often found in a normal tissue structure will be described with reference to FIG.
In FIG. 8, the processing steps from extracting the lung field region and rib region from the medical image and setting the ROI in the lung field region excluding the rib region are the same as the processing shown in FIG. 2. Therefore, in FIG. 8, the same step numbers are assigned to the same processing steps as in FIG.

  After setting the ROI, the candidate detection unit 16 creates a similarity map corresponding to the lesion as shown in FIG. 8 (step T11). The similarity map corresponding to a lesion refers to a similarity map created by calculating a similarity with a shape often seen in a lesion. The method of creating the similarity map itself is the same as the method described with reference to FIG. Then, the candidate detection unit 16 calculates a feature amount using the created similarity map (step T21). The processing for calculating the feature amount is the same as the processing in steps S5 to S7 shown in FIG. 2, and the feature amount of the similarity map and the feature amount of the isolated region are calculated.

  On the other hand, the candidate detection unit 16 creates a similarity map corresponding to a normal tissue structure (step T12), and calculates a feature amount using the created similarity map (step T22). The similarity map corresponding to a normal tissue structure is a similarity map created by calculating a similarity with a shape often found in a normal tissue structure. The method of creating the similarity map itself is the same as the method described with reference to FIG. 4, and the specific shape for calculating the similarity may be set to a shape often found in normal tissue structures. The processing for calculating the feature amount is the same as the processing in steps S5 to S7 shown in FIG. 2, and the feature amount of the similarity map and the feature amount of the isolated region are calculated.

  The candidate detection unit 16 performs feature amount analysis on the feature amount calculated using each of the similarity map corresponding to a lesion and the similarity map corresponding to a normal tissue structure, and detects a lesion candidate (step S8). The detection result is displayed on the display unit 13 by the control unit 11 (step S9). When an image of a lesion is included in the ROI, the similarity with the shape often seen in the lesion increases, and therefore the similarity map corresponding to the lesion tends to have a large similarity as a whole. On the other hand, since the similarity with the shape often found in the normal tissue structure is small, the similarity map corresponding to the normal tissue structure tends to have a small similarity as a whole. That is, the feature amount calculated from each of the similarity map corresponding to a lesion and the similarity map of a normal tissue structure is considered to show a certain tendency depending on whether or not a lesion is included in the ROI.

  Therefore, a similarity map is created that includes not only the shape often found in lesions but also the shape often seen in normal tissue structures, and the feature amount calculated using the similarity map is used to detect lesion candidates. As a result, multifaceted detection can be performed. As a result, improvement in detection accuracy can be expected.

  In addition, as an example of lesion candidate detection, an example of detecting an interstitial disease from an X-ray image obtained by imaging the chest was used. In this case, the present invention can be applied. The present invention can also be applied to the detection of lesion candidates such as aneurysms from not only X-ray images but also MRA (Magnetic Resonance Angiography) images obtained by imaging the head.

  In addition, as a computer-readable medium for the program according to the present invention, a non-volatile memory such as a ROM and a flash memory, and a portable recording medium such as a CD-ROM can be applied. Further, a carrier wave (carrier wave) is also applied to the present invention as a medium for providing program data according to the present invention via a communication line.

It is a figure which shows the functional structure of the medical image processing apparatus in this Embodiment. It is a flowchart which shows the process performed by the medical image processing apparatus. It is a setting example of ROI. It is a flowchart which shows the process which produces a similarity map. It is an example of a histogram of similarity. It is an example of the isolated area | region to which the identification number was attached | subjected. It is an example of a display of the detection result of the lesion candidate of interstitial disease. It is a flowchart which shows the other process example which detects a lesion candidate.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Medical image processing apparatus 11 Control part 12 Operation part 13 Display part 14 Communication part 15 Storage part 16 Candidate detection part

Claims (5)

Using the medical image data, for each region of interest set in the medical image, create a similarity map indicating the similarity between the region of interest and a specific shape for one or a plurality of different shapes, When it is determined that a candidate lesion is included in the region of interest based on the result of analyzing the feature amount of each similarity map created for a plurality of different shapes, the region of interest is detected as a candidate lesion region. Candidate detection means to
Display means;
Control means for displaying the detection results of the lesion candidates on the display means;
A medical image processing apparatus comprising:   The candidate detecting means calculates a feature amount of the similarity map only for an image area in which the similarity is a certain value or more in each similarity map created for the one or a plurality of different shapes in the region of interest. The medical image processing apparatus according to claim 1 to calculate.   The candidate detection means obtains an isolated region in which pixels having a certain range of similarities are adjacent to each other in each of the similarity maps created for the one or a plurality of different shapes, and calculates a feature amount of the isolated region The medical image processing apparatus according to claim 2, wherein the feature amount analysis uses a feature amount of the similarity map and a feature amount of the isolated region.   The candidate detecting means creates a similarity map indicating the similarity to the shape corresponding to the normal tissue structure and a similarity map indicating the similarity to the shape corresponding to the lesion, respectively, and corresponds to the normal tissue structure The medical image processing apparatus according to any one of claims 1 to 3, wherein a feature amount of each similarity map created for a shape to be processed and a shape corresponding to a lesion is calculated and used for the feature amount analysis. Computer
Using the medical image data, for each region of interest set in the medical image, create a similarity map indicating the similarity between the region of interest and a specific shape for one or a plurality of different shapes, When it is determined that a candidate lesion is included in the region of interest based on the result of analyzing the feature amount of each similarity map created for a plurality of different shapes, the region of interest is detected as a candidate lesion region. Candidate detection means to
Control means for displaying the detection result of the lesion candidate on a display means;
Program to function as.

Images