JP2009207727A - Vertebral body position determining device, method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve detection performance in detecting positions of both ends of vertebral bodies in a medical image. <P>SOLUTION: A central line of a vertebra is obtained from a plurality of medical images showing cross-sectional surfaces in different positions of one vertebra. A corpus cavernosum region in a body axial direction of each vertebral body is estimated from each pixel value in a specific region located in the vicinity of the central line of the vertebra in each of the plurality of the medical images. The amount of characteristic of each pixel value in the specific region is calculated in the estimated corpus cavernosum region of each vertebral body, and a three-dimensional center of the corpus cavernosum region of each vertebral body is detected from the amount of characteristic. Positions of the vertebral body are determined by calculating positions of both ends of the vertebral body in the body axial direction from the three-dimensional center of the estimated corpus cavernosum region of each vertebral body. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to vertebra detection processing, and more particularly to a vertebral body position determining apparatus and method suitable for automatic detection of both end positions of a vertebral body in a medical image, and a program therefor.

  Conventionally, in the medical field, automatically calculating the positions of both ends of a vertebral body based on a plurality of medical images obtained by scanning a subject from a chest to a base of a foot at a plurality of cutting positions segments each vertebral body. This is an indispensable technology.

Therefore, processing for automatically detecting the positions of both ends of the vertebral body in a plurality of medical images using a computer is performed. For example, Patent Literature 1 proposes a technique of extracting vertebral body endplates and intervertebral spaces by performing morphological operations on medical images and separating vertebral bodies. In addition, the intervertebral area candidate area is detected by performing a follow-up operation on the medical image, and the intervertebral area is detected using the projection value of the detected intervertebral area candidate area pixel to separate the vertebral bodies. A technique to do this is proposed in Patent Document 2.
"Construction of osteoporosis diagnosis support algorithm using multi-slice CT image", IEICE Technical Report MI106, pp.25-28, 2007.1.20 “Using mathematical morphology for the anatomical labeling of vertebrae from 3D CT-scan images”, Computerized Medical Imaging and Graphics, Vol. 31, pp. 141-156, 2007

  However, with the above method, it is difficult to detect the intervertebral portion of the subject when the subject suffers from osteoproliferative disease or when the slice interval taken by the CT apparatus is large. Especially in the upper part of the subject's thoracic vertebra and the cervical vertebra, the intervertebral part is integrated with the vertebral body endplate by the partial volume effect, so it does not become a typical "low CT area", so both end positions of the vertebral body are accurately detected There was a problem that could not be done.

  Therefore, in view of the above circumstances, the present invention has a vertebral body composed of a cortical region (outside) and a corpus cavernosum region (inside). The cavernous area of the vertebral body can be detected more stably than the interspace. By paying attention to this point, it is an object of the present invention to provide a vertebral body position determining apparatus and method capable of further improving the detection performance of both end positions of the vertebral body, and a program therefor.

  The vertebral body position determining device according to the present invention includes a vertebra centerline calculating means for obtaining a vertebra centerline from a plurality of medical images showing cross sections at different positions of one vertebra, and a vertebra center in each of the plurality of medical images. Based on each pixel value of the specific area existing near the line, the corpus cavernosum area estimation means for estimating the corpus cavernosum area at least in the body axis direction of each vertebral body, and in the estimated corpus cavernosum area of each vertebral body Calculate the feature value of each pixel value of the region, and based on the feature value, center detection means for detecting the three-dimensional center of the corpus cavernosum area of each vertebral body, and the detected three-dimensional center of the corpus cavernosum area of each vertebral body Based on this, the vertebral body position determining means for determining the vertebral body position by calculating the positions of both ends of each vertebral body in the body axis direction is provided.

  The “cross section at different positions of one vertebra” refers to a cross section that is substantially orthogonal to the body axis direction at different positions of one vertebra.

  The “medical image” is an axial image or the like used for image diagnosis, and examples include a radiographic image, CT image, MRI image, RI image, and PET image.

  In addition, the “specific region existing near the center line” refers to a region including a region passing through the center line existing around the center line.

  For example, the “center detection unit” obtains the feature amount of each pixel in a plurality of sample images including a region known to be a corpus cavernosum region by machine learning in advance, and calculates the feature amount of each pixel in a specific region. A score indicating the extent to which the corpus cavernosum area is included in the specific area is calculated based on the feature amount obtained by machine learning, and the three-dimensional center of the corpus cavernosum area is detected based on the score. Also good. Furthermore, it may have range data regarding the body axis direction width of each vertebral body in anatomy, and detect the three-dimensional center of the corpus cavernosum region based on the range data and the score.

  The “feature amount of each pixel in the specific region” may include, for example, a value of edge intensity projected in the body axis direction of the subject.

  The vertebral body position determination method of the present invention obtains a vertebra centerline from a plurality of medical images showing cross sections of different positions of one vertebra, and specifies a vertebral body centered in each of the plurality of medical images. Based on each pixel value of the region, the cavernous region in at least the body axis direction of each vertebral body is estimated, and the feature value of each pixel value of the specific region is calculated within the estimated cavernous region for each estimated vertebral body. , Detecting the three-dimensional center of the corpus cavernosum area for each vertebral body based on the feature quantity, and based on the detected three-dimensional center of the corpus cavernosum area for each vertebral body, the positions of both ends in the body axis direction of each vertebral body The vertebral body position is determined by calculating.

  The computer program for determining the position of the vertebral body according to the present invention has a function for obtaining a center line of the vertebra from a plurality of medical images showing cross sections at different positions of one vertebra, and a vertebral body position for each of the plurality of medical images. Based on each pixel value of the specific area existing near the center line, the function of estimating the corpus cavernosum area at least in the body axis direction of each vertebral body, and within the estimated corpus cavernosum area of each vertebral body, Calculate the feature value of the pixel value, and based on the feature value, detect the three-dimensional center of the corpus cavernosum area of each vertebral body, and based on the detected three-dimensional center of the corpus cavernosum area for each vertebral body, The function of determining the vertebral body position is executed by calculating the positions of both ends of the vertebral body in the body axis direction.

  According to the vertebral body position determining apparatus, method, and program of the present invention, a vertebral centerline is obtained, and each vertebral body is determined based on each pixel value of a specific region existing in the vicinity of the vertebral centerline in each of a plurality of medical images. The at least cavernous area in the body axis direction is estimated, and within the estimated cavernous area of each vertebral body, the feature value of each pixel value of the specific area is calculated, and the cavernous body of each vertebral body is calculated based on the feature value By detecting the three-dimensional center of the region and calculating the positions of both ends in the body axis direction of each vertebral body based on the detected three-dimensional center of the corpus cavernosum region for each vertebral body, The position can be detected stably.

  Hereinafter, an embodiment of a vertebral body position determining apparatus 10 according to the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a preferred embodiment of a vertebral body positioning device 10 of the present invention.

  The configuration of the vertebral body position determination device 10 as shown in FIG. 1 is realized by executing a vertebral body position determination program read into an auxiliary storage device (not shown) on a computer (for example, a personal computer). . The vertebral body position determination program is stored in an information storage medium such as a CD-ROM or distributed via a network such as the Internet and installed in a computer.

  The vertebral body position determining device 10 automatically calculates the positions of both ends of a vertebral body displayed in a medical image acquired by, for example, a CT image, and includes an image acquisition unit 1, a vertebra centerline calculation unit 5, It has a corpus cavernosum region estimation means 6, a center detection means 7, a vertebral body position determination means 8, a display means 9 and the like.

  The image acquisition means 1 acquires a medical image acquired by a CT image as shown in FIG. 4, for example. Note that the image acquisition unit 1 may acquire not only a CT image but also a medical image such as a so-called MRI image, RI image, PET image, X-ray image, and the like.

  The vertebra centerline calculation means 5 calculates the centerline of the vertebra from a plurality of medical images showing cross sections at different positions of one vertebra.

  The vertebra centerline calculation means 5 sets a peripheral region on the basis of the target pixel of each pixel in a plurality of medical images, and uses the discriminator generated by the machine learning method to determine whether the peripheral region is a spinal cord region. A spinal cord region detecting means 2 for determining whether or not a spinal cord center line constituted by a central point in a plurality of spinal cord regions detected by the spinal cord region detecting means 2 is generated in a longitudinal sectional image showing a longitudinal section of a vertebra, When the discriminator of the spinal cord region detection means 2 calculates a discriminant value indicating the degree of discriminating that the peripheral region is a spinal cord region, the target pixel serving as a reference for the peripheral region whose discriminant value is a predetermined value or more is determined as the spinal cord region A spinal centerline generating means 3 serving as a center point of the spine; and a vertebra centerline determining means 5 for determining a centerline of the vertebra based on a positional relationship between the spinal cord centerline generated by the spinal cord centerline generating means 3 and the vertebrae. Is.

  For each medical image, the vertebra centerline determination means 5 passes through the center point in the spinal cord region detected by the spinal cord region detection means 2 and the spinal region and passes through each pixel in the medical image on a straight line that does not pass through the heart. Luminance values are extracted, and each straight line having the extracted luminance values is generated by, for example, a longitudinal section image generating means (not shown) for generating a longitudinal section image as shown in FIG. 8 and a longitudinal section image generating means. Vertebral region detecting means (not shown) that detects two trabecular lines having a predetermined luminance value or more from a longitudinal cross-sectional image and detects an area including a spinal region sandwiched between the detected two trabecular lines as a vertebra. ).

  The display means 9 is a monitor that displays medical images, a CRT screen, or the like.

  Next, the process in the embodiment performed in the vertebral body position determining apparatus 10 having the above configuration will be described.

  FIG. 2 is a series of process flowcharts of the vertebral body position determining apparatus. First, as shown in FIG. 2, the spinal cord centerline calculating means 5 detects the centerline of the vertebra from a plurality of medical images showing cross sections at different positions of one vertebra acquired by the image acquiring means 1 (# 1 ).

  FIG. 3 is a flowchart of a specific series of processing for detecting the center line of the vertebra. First, as shown in FIG. 3, the spinal cord region detecting means 2 detects the spinal cord region (# 10). For example, FIG. 4 shows a typical CT image of the spinal cord that can be placed on a medical image (axial image) composed of images of each ring when the subject is cut from the chest to the base of the foot at a plurality of cutting positions. Since the spinal cord region has a clear pattern, it is possible to stably detect the spinal cord region from a medical image using image detection technology.

  As a spinal cord region detection method, a machine learning method based on Adaboost, which is a method for creating an integrated learning machine, may be used.

  A technique for detecting the spinal cord region can be used for detection of the spinal cord region in the vertebral body position determining apparatus of the present invention. The spinal cord region is tracked by using known methods such as feature point detection and learning data, and updating the weights sequentially when resampling. A machine learning technique based on Adaboost, which is a technique for creating machines, is used. In the learning sample image, for example, as shown in FIG. 5, the upper and lower points P1 and P2 of the spinal cord region are designated, the center between the two points is the center, and the length is twice the distance between the designated two points. A square area is set as a region of interest. The sizes of the quadrangular regions cut from various positions (thoracic vertebrae and lumbar vertebrae) of CT images of various patients are normalized and aligned with the same size. Let these regions of interest be positive learning samples.

  Next, the spinal cord region detection means 2 randomly cuts out square regions of various sizes from a region away from the designated region, aligns the sizes thereof, and uses them as negative learning samples.

  Next, the spinal cord region detection means 2 uses a combination of values of n pixel pairs selected at random as features for positive and negative learning sample images, for example, as shown in FIG. A discriminator that distinguishes positive and negative patterns is created by a machine learning technique based on this.

When detecting the spinal cord region, the medical image is scanned, and square regions of various sizes with the pixel of interest at the center are cut out, and feature amounts are calculated as shown in FIG.

  This is input to the discriminator obtained in the learning stage, the discriminant value is obtained, and the maximum value among them is used as the score of the pixel of interest. A position (pixel) at which the discriminant value is maximized is detected from the medical image and set as the center of the spinal cord region.

  In addition, as a technique for detecting the spinal cord region, a known template matching technique (for example, Reference 1; JP 2002-109548 A) or a demarcation technique using a ring model (for example, Reference 2; JP 2007-111533). Can be used for detecting the spinal cord region in the image processing apparatus of the present invention.

  Next, the spinal cord center line generating means 3 generates a spinal cord center line from the central point of the spinal cord region (# 11). As the center point, a predetermined pixel existing in the approximate middle in the spinal cord region is set. It does not necessarily have to be the exact center in the spinal cord area. In addition, a point that is substantially equidistant from the periphery of the spinal cord region or a predetermined both ends, or a barycentric point may be set.

  Specifically, a smooth three-dimensional spinal cord center line is obtained from the centers of the spinal cord regions of the plurality of medical images obtained in the previous step. As a method for calculating the spinal cord center line, a method of fitting a polyline or curve (polynomial curve, B-Spline curve, etc.) to the center point of the obtained spinal cord region (for example, Reference 3; JP-A-6-189934) There is. Also, RANSAC methods that randomly extract several samples and apply the least squares method (eg Reference 4; MAFischler and RCBolles (June 1981). “Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis. and Automated Cartography ”.Comm. of the ACM 24: 381-395).

  Then, when the spinal cord centerline generating means 3 sets the center point from the detected spinal cord region, it is the discriminant value obtained (the value representing the degree to discriminate that it is a spinal cord region, in other words, the spinal cord region (It is a value corresponding to the pattern likeness). At that time, only the center point of the spinal cord region where the discriminant value exceeds a certain threshold is selected, and a smooth three-dimensional curve of the spinal cord region is generated by the calculation method of the spinal cord center line.

  When using the spinal cord centerline calculation method described above, the discriminant value may be used as a weighting factor for the central point of each spinal cord region.

  Next, the vertical plane image generation means generates a cross-sectional image (# 12). Specifically, in order to calculate the spine centerline based on the spinal cord centerline calculated by the spinal cord centerline calculation method described above, first, the center of the spinal cord region obtained in each medical image is used. As shown in FIG. 7, the luminance value of each pixel in the medical image on the straight line inclined α degrees counterclockwise from the Y axis with respect to the center point as a reference is extracted, and each straight line having the extracted luminance value is extracted. A longitudinal section image is generated. The straight line generates a longitudinal section image. Further, since the heart part has many blood vessels and is not a stable pattern, the straight line may be set so as to pass through the center point and the spine region and not through the heart part.

  The vertebra detection means detects the boundary line of the vertebra in the longitudinal section image (# 13). Specifically, in the longitudinal cross-sectional image, the spinal cord center line becomes a single curve, and the spinal region is surrounded by two trabecular lines with high CT values (pixel values) on the left side of the spinal cord center line. A cancellous bone region having a low CT value (pixel value) appears.

  The vertebra detection means calculates the center line and width of the vertebra (# 14). Specifically, the centerline of the vertebra is calculated using the spinal cord centerline that is the curve L3 in FIG. First, a difference in the X direction is obtained for the longitudinal cross-sectional image of FIG. A large positive difference value is obtained at the ventral edge portion of the spine region, and a large negative difference value is obtained at the dorsal edge portion.

The vertebra detection means linearly transforms the spinal cord center line L3 to calculate the linear transformation numbers a and b using the equation (1) in order to fit the spine region to the dorsal edge line L2. The back edge line L5 is calculated.

  For the ventral edge curve L1, the edge curve L4 can be obtained by the same method.

  The fitting method is not limited to the above formula, and the curves L4 and L5 may be calculated directly from the gradient of the luminance value of the longitudinal section image.

  Curves L4 and L5 calculated as shown in FIG. 8 are the left and right boundary lines of the vertebra including the spine region. The center line and width of the vertebra can be calculated from the calculated left and right boundary lines.

  As described above, according to the embodiment of the present invention, the center of the vertebra can be accurately calculated based on the positional relationship with the spinal cord center line in the medical image (# 1).

  In the above embodiment, the vertebra centerline determination means 4 determines the vertebra centerline based on the positional relationship between the spinal cord centerline and the vertebrae generated by the spinal cord centerline generation means 3. Here, a process for calculating a vertebra centerline using a voxel image formed by stacking medical images will be described.

When the longitudinal section image generating means defines the center point in the spinal cord region detected by the spinal cord region detecting means 2 and the center line in the voxel image as shown in the following equation (2), P3 and P4 in FIG. A new vertical cross-sectional image obtained by cutting the voxel image can be expressed by the cross-section to be connected as shown in Expression (3).

  Θ used in Equation (3) is the inclination angle of the cross section connecting P3 and P4 in FIG. 7, and λ indicates the position on the cross section. The position range (range of λ) is determined by the frame range of the medical image.

  The vertebra region detecting means detects two trabecular lines having a predetermined luminance value or more from the new longitudinal section image generated by the longitudinal section image generating means, and is sandwiched between the detected two trabecular lines. An area including a spine area is detected as a vertebra.

Since the vertebra region detecting means obtains two trabecular lines, the edge of a new longitudinal section image can be obtained by the equation (4).

  Δ is a small value and is usually one pixel, but is not necessarily limited thereto.

Further, the vertebra region detection means can fit the dorsal vertebra wall using the equation (6) from the edge obtained by the equation (4), and the equation (7) The optimal parameters obtained from the conversion are expressed as in equation (5)

Then, a curve L5 that is an edge line on the dorsal side of the vertebra can be obtained.

  In addition, the vertebra region detecting means 7 can obtain the ventral edge curve L4 by the same method.

  Curves L4 and L5 calculated as shown in FIG. 8 are the left and right boundary lines of the vertebra including the spine region. The center line and width of the vertebra can be calculated from the calculated left and right boundary lines.

  Thus, the center of the vertebra can be accurately calculated based on the positional relationship with the spinal cord center line in the medical image using the voxel image from one embodiment of the present invention (# 1).

  Next, the corpus cavernosum area estimation means 6 is based on each pixel value of the specific area existing in the vicinity of the vertebra centerline calculated by the vertebra centerline calculation means 5 in each of the plurality of medical images. Is estimated (# 2).

  The corpus cavernosum area estimation means 6 uses each pixel of the area including the area passing through the center line existing around the vertebra center line calculated by the vertebra center line calculation means 5. The photographed medical image may show vertebral lesions and deformations, and a more stable feature amount can be obtained by using only a specific region rather than using the entire medical image. In addition, in Patent Document 1 and Patent Document 2 described above, segmentation is performed by first extracting the intervertebral portion. However, the present invention first extracts a corpus cavernosum region showing a relatively stable image pattern, Is used to obtain the intervertebral part.

  FIG. 9 conceptually shows vertebrae, vertebral bodies, and the like. The cylinder S1 shown in FIG. 9 represents a vertebral body, and the surface of the cylinder S1 corresponds to a cortical portion on the side of the vertebral body. The top and bottom plates of the cylinder correspond to the upper and lower end plates of the vertebral body, respectively. The protrusion S2 represents a bone growth portion. A red cylindrical portion T1 represents a stack of specific regions.

  When the cavernous body region estimation means 6 obtains the CT value (pixel value) of the red cylindrical portion T1 for each specific region as a feature amount (vertical direction), a graph as shown in FIG. 9 is obtained.

  As shown in the graph of FIG. 9, the intervertebral portion disappears due to the partial volume effect, but regions (for example, K1 to K3) with low CT values (pixel values) corresponding to the corpus cavernosum region appear. Further, when the entire vertebra region is used, the valley of the intervertebral portion may not be noticeable due to the influence of the bone growth portion of the process S2.

  Therefore, if the corpus cavernosum area estimation means 6 is a medical image showing a subject having a severe compression fracture, the corpus cavernosum area may be crushed and become an area having a high CT value (pixel value). If the edge feature quantity in the body axis direction is used instead of the sum of CT values (pixel values), it is possible to detect the corpus cavernosum area in the case of a compression fracture.

  Next, the center detecting means 7 calculates the feature quantity of each pixel value of the specific area within the cavernous area of each vertebral body estimated by the cavernous area estimation means 6, and for each vertebral body based on the feature quantity. The three-dimensional center of the corpus cavernosum region is detected (# 3).

  Further, the center detecting means 7 obtains the feature amount of each pixel in a plurality of sample images including a region that is known to be a cavernous region by machine learning in advance, and the feature amount of each pixel in the specific region, A score indicating the extent to which the corpus cavernosum region is included in the specific region is calculated based on the feature amount obtained by machine learning, and the three-dimensional center of the corpus cavernosum region is detected based on the score.

  The feature amount of each pixel value in the specific region includes, for example, the CT value (pixel value) itself, the sum of CT values (pixel values), the sum of edge strengths, the sum of body axis edge strengths, and the CT values (pixels) of the local region. Value) distribution, combination distribution of CT values (pixel values) at different positions in the local area, and combination distribution of edge values at different positions in the local area.

  The center detection means 7 detects the three-dimensional center of the corpus cavernosum region of the vertebral body from the feature value distribution of each pixel value of the specific region. For example, a machine learning method based on the above-described Adaboost may be used. In this method, the vertebral corpus cavernosum region and other region learning samples are cut out from several learning cases, and feature quantities of each pixel value in the specific region are calculated and learned.

  When detecting the corpus cavernosum region, the center detecting means 7 calculates the feature amount from the local region along the spine centerline, discriminates it by a machine learning method based on Adaboost, and calculates the score. The maximum point (peak value) is obtained from the score, and is set as the three-dimensional center of the corpus cavernosum region of the vertebral body.

  In addition, using the feature value of each pixel value in the specific area and the score calculated by the machine learning technique based on Adaboost, the three-dimensional center of the vertebral corpus cavernosum area is detected by an overall optimization technique. May be.

  The center detecting means 7 may further include range data relating to the anatomical width of each vertebral body in the body axis direction. Based on the above range data, the center detection means 7 has a vertebra width within a certain range, and under the condition of anatomy in which the vertebrae gradually increase from the cervical vertebra to the lumbar vertebra, the feature value of each pixel value and Adaboost described above The peaks or valleys that best satisfy the anatomical constraints as a whole may be searched from the scola calculated by the machine learning method based on the vertebral body, and the vertebral bodies may be separated using them.

  FIG. 10 is a series of processing flowcharts for detecting the three-dimensional center of the corpus cavernosum region. First, as shown in FIG. 10, the cavernous area detection means 7 projects the feature amount of each pixel value in the specific area in the body axis direction (# 20). The cavernous area detection means 7 selects the initial (first) peak value of the score obtained from the projected feature amount (# 21). Another peak value that satisfies the condition of the range data is searched (# 22). An overall evaluation value is calculated from the score value and the range data (# 24). The maximum evaluation value is recorded from the evaluation value (# 24).

  When there is a peak value that has not been searched yet (# 25; YES), the corpus cavernosum area detecting means 7 selects the initial (first) peak value of the score obtained from the projected feature amount. cure.

  On the other hand, when there is no peak value that has not been searched yet (# 25; NO), the corpus cavernosum area detection means 7 outputs the recorded maximum evaluation value (# 26).

  Next, the vertebral body position determining means 8 calculates the positions of both ends of each vertebral body in the body axis direction based on the three-dimensional center of the cavernous area of each vertebral body detected by the cavernous area region detecting means 7. Thus, the vertebral body position is determined (# 4).

  The vertebral body position determining means 8 sets an intermediate point between the three-dimensional centers of the cavernous areas of two adjacent vertebral bodies as the intervertebral portion of the two vertebral bodies. As another method, a vertebral body may be determined using the above-described feature amount. Specifically, the position of the cortical portion of the vertebral body is detected by detecting the maximum point of the sum (or edge strength) of the CT values (pixel values) between the three-dimensional centers of the cavernous areas of the two adjacent vertebral bodies. It is also possible to use a method of detecting vertebral bodies and determining the ends of vertebral bodies.

  The display means 9 uses the information on the position of the end of the vertebral body determined by the vertebral body position determining means 8 to display a three-dimensional medical image or the like showing the segmented vertebral body on a monitor, a CRT screen or the like.

  Thus, according to the present invention, the center line of the vertebra is obtained, and at least in the body axis direction of each vertebral body based on each pixel value of the specific region existing in the vicinity of the center line of the vertebra in each of a plurality of medical images. Estimate the corpus cavernosum area, calculate the feature value of each pixel value of the specific area within the estimated corpus cavernosum area of each vertebral body, and based on the feature quantity, determine the three-dimensional center of the corpus cavernosum area for each vertebral body Based on the detected three-dimensional center of the corpus cavernosum area of each vertebral body, the positions of both ends of each vertebral body in the body axis direction are detected, thereby stably detecting the positions of both ends of the vertebral bodies. be able to.

Functional block diagram of vertebral body positioning device A series of flowcharts in the embodiment of the present invention A series of flowcharts in the detection of the spinal cord centerline of the present invention Diagram showing an example of an axial image The figure which shows an example of the axial image part around a vertebra Conceptual diagram showing an example of feature values used for the Adaboost process A figure showing an example of an axial image and a longitudinal section image along the spinal cord center line Diagram showing vertebra boundary Conceptual diagram showing vertebrae and vertebral bodies A series of flowcharts in the center detection of the cavernous region of the present invention

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image acquisition means 2 Spinal cord area | region detection means 3 Spinal cord centerline production | generation means 4 Vertebral centerline determination means 5 Vertebral centerline calculation means 6 Cavernous area estimation means 7 Center detection means 8 Vertebral body position determination means 9 Display means 10 Vertebral center detection apparatus

Claims (6)

A vertebra centerline calculating means for determining a centerline of the vertebra from a plurality of medical images showing cross sections of different positions of one vertebra;
A corpus cavernosum area estimation means for estimating a corpus cavernosum area in at least the body axis direction of each vertebral body based on each pixel value of a specific area existing near the center line of the vertebra in each of the plurality of medical images;
Center detection means for calculating a feature quantity of each pixel value of the specific area in the estimated cavernous area of each vertebral body and detecting a three-dimensional center of the cavernous area of each vertebral body based on the feature quantity When,
Vertebral body position determining means for determining a vertebral body position by calculating positions of both ends of each vertebral body in the body axis direction based on the detected three-dimensional center of the corpus cavernosum area of each vertebral body. A vertebral body position determining apparatus characterized by being a thing.   The center detection means obtains the feature amount of each pixel in a plurality of sample images including a region known to be a cavernous region by machine learning in advance, and the feature amount of each pixel in the specific region and the machine A score indicating the extent to which the corpus cavernosum region is included in the specific region is calculated based on the feature amount obtained by learning, and a three-dimensional center of the corpus cavernosum region is detected based on the score. The vertebral body position determining device according to claim 1.   The vertebral body position determining apparatus according to claim 2, wherein the feature amount of each pixel in the specific region includes a value of edge intensity projected in the body axis direction of the subject.  The center detecting means further has range data relating to the body axis direction width of each vertebral body in anatomy, and detects the three-dimensional center of the corpus cavernosum region based on the range data and the score. The vertebral body position determining apparatus according to claim 2 or 3, wherein Determining a centerline of the vertebra from a plurality of medical images showing cross sections of different positions of a single vertebra;
Based on each pixel value of a specific region existing near the center line of the vertebra in each of the plurality of medical images, the cavernous region for each vertebral body is estimated,
At least in the cavernous region in the body axis direction of each estimated vertebral body, a feature amount of each pixel value of the specific region is calculated, and based on the feature amount, a three-dimensional center of the cavernous region of each vertebral body is calculated. Detect
Vertebral body position determination characterized in that, based on the detected three-dimensional center of the corpus cavernosum area of each vertebral body, the positions of the vertebral bodies are determined by calculating the positions of both ends in the body axis direction of each vertebral body. Method. On the computer,
A function for obtaining a centerline of the vertebra from a plurality of medical images showing cross sections of different positions of one vertebra;
A function of estimating the corpus cavernosum area of each vertebral body based on each pixel value of a specific area existing near the center line of the vertebra in each of the plurality of medical images;
In the estimated cavernous area of each vertebral body, a feature amount of each pixel value of the specific area is calculated, and a function of detecting a three-dimensional center of the cavernous area of each vertebral body based on the feature amount;
Based on the detected three-dimensional center of the corpus cavernosum area of each vertebral body, the function of determining the vertebral body position is executed by calculating the positions of both ends of each vertebral body in the body axis direction. program.