BLOOD VESSEL DETECTING APPARATUS, MAGNETIC RESONANCE IMAGING APPARATUS, AND PROGRAM

Abstract

An MRI apparatus 1 comprising an image producing unit 101 for producing a plurality of axial images D1 to D10 in a plurality of slices defined in a body part to be imaged containing a blood vessel; a classifying unit 102 for classifying the plurality of axial images D1 to D10 into a plurality of classes I to IV based on which a portion of the imaged body part each of the plurality of axial images D1 to D10 represents; and a defining unit 103 for defining a search region for searching for a blood vessel from within an axial image based on within which of the plurality of classes I to IV the axial image falls.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

The present application claims priority to Japanese Patent Application No. 2016-085200, filed on Apr. 21, 2016, the entirety of which is incorporated herein by reference.

BACKGROUND

The present invention relates to a blood vessel detecting apparatus for detecting a blood vessel, a magnetic resonance imaging apparatus comprising the blood vessel detecting apparatus, and a program applied to the blood vessel detecting apparatus.

Magnetic resonance imaging apparatuses for imaging a subject using a contrast medium are known.

An example of the method of imaging a subject using a contrast medium is one in which an operator produces an image in each of a plurality of axial planes intersecting a torso, visually finds a position of an aorta from within/among an image/images, and defines a tracker region for detecting the contrast medium at the position of the aorta.

An aorta is detected by increasing the area of a Z_dephaser in a slice gradient magnetic field to diminish as much as possible signals from blood in the aorta rendered in an axial image. Blood in the aorta, however, sometimes exhibits high signals in cardiac systole because the blood flow velocity in the aorta lowers then. Therefore, in practice, signals from blood in the aorta sometimes cannot be fully diminished due to an effect of the high-signal blood. This poses a problem that a tissue different from the aorta is wrongly detected as an aorta, or the aorta cannot be detected. Accordingly, there has been disclosed a technique capable of detecting a position of the aorta even when signals from blood in the aorta are not fully diminished.

The above-referenced technique is capable of detecting a position of the aorta even when signals from blood in the aorta are not fully diminished. In some axial images, however, a cross section of a body part, different from the cross section of the aorta of interest to be detected, may exhibit a signal pattern resembling that of the aorta. In this case, the cross section of the body part different from the aorta may be wrongly detected as a cross section of an aorta. When the aorta is wrongly detected, a tracker is defined at a position offset from the aorta, resulting in a problem that accuracy in detecting a contrast medium is deteriorated.

Accordingly, there is a need for developing a technique for improving the accuracy of detection of a blood vessel in an image.

SUMMARY

In a first aspect, is a blood vessel detecting apparatus comprising an image producing unit for producing a plurality of images in a plurality of slices defined in a body part to be imaged containing a blood vessel; a classifying unit for classifying said plurality of images into a plurality of classes based on which portion in said imaged body part each of said plurality of images represents; and a defining unit for defining a search region for searching for said blood vessel from within said image based on within which of said plurality of classes said image falls.

The present invention, in its second aspect, is a magnetic resonance imaging apparatus comprising scanning section for performing a scan on a body part to be imaged containing a blood vessel, said apparatus comprising the blood vessel detecting apparatus in the first aspect.

The present invention, in its third aspect, is a program for causing a computer to execute image producing processing of producing a plurality of images in a plurality of slices defined in a body part to be imaged containing a blood vessel; classifying processing of classifying said plurality of images into a plurality of classes based on which portion in said imaged body part each of said plurality of images represents; and defining processing of defining a search region for searching for said blood vessel from within said image based on within which of said plurality of classes said image falls.

A plurality of images are classified into a plurality of classes, and a search region for searching for the blood vessel from within an image is defined according to within which of the plurality of classes the image falls; thus, accuracy of detection of a blood vessel within an image may be improved.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic diagram of a magnetic resonance imaging apparatus in one embodiment of the present invention;

FIG. 2 is a diagram explaining unit the processing apparatus 10 implements;

FIG. 3 is a diagram showing scans performed in the present embodiment;

FIG. 4 is a diagram schematically showing a body part to be imaged;

FIG. 5 is a diagram showing an exemplary flow in performing a localizer scan LS and a main scan MS;

FIG. 6 is a diagram explaining the localizer scan LS;

FIG. 7 is a diagram schematically showing axial images DA₁ to DA_n, DA_n+1 to DA_n+a, . . . , DA_j+1 to DA_k for ‘m’ subjects SU₁ to SU_m;

FIG. 8 is a diagram schematically showing images each obtained by cropping an image portion in a rectangular region circumscribing a torso region C1 from an axial image;

FIG. 9 is a diagram schematically showing images after their resolution is scaled down;

FIG. 10 is a diagram showing a vector determined for each of the scaled-down images a₁′ to a_n′;

FIG. 11 is a diagram showing a vector α_i obtained for each axial image for the subjects SU₁ to SU_m;

FIG. 12 is a diagram schematically showing an adjacency matrix MA;

FIG. 13 is a diagram showing a classification map CM;

FIG. 14 is a diagram showing an exemplary flow of Step ST2;

FIG. 15 is a diagram schematically showing an axial image D₁ in a slice SL₁;

FIG. 16 is a diagram schematically showing a binary image DB₁ obtained by binarizing the axial image D₁ in the slice SL₁;

FIG. 17 is a diagram schematically showing a binary image DC₁ in which arm portions are erased;

FIG. 18 is a diagram schematically showing an image obtained by cropping an image portion in a rectangular region circumscribing a torso region HB from the axial image D₁;

FIG. 19 is a diagram schematically showing an image D₁″ after its resolution is scaled down;

FIG. 20 is a diagram showing a vector β₁ determined for the scaled-down image D₁″;

FIG. 21 is a diagram showing a point E projected onto the classification map CM;

FIG. 22 is a diagram explaining a method of identifying N points;

FIG. 23 is a diagram showing within which class each of the axial images D₁ to D₁₀ falls;

FIG. 24 is a diagram explaining a probability distribution model;

FIG. 25 is a diagram explaining a method of defining a search region;

FIG. 26 is a diagram schematically showing multiplied images MD₁′ to MD₁₀′ and search regions R₁ to R₁₀ determined for the axial images D₁ to D₁₀;

FIG. 27 is a diagram showing a position of a detected cerebrospinal fluid CSF;

FIG. 28 is a diagram showing lines Ls and Le;

FIG. 29 is a diagram showing a window W;

FIG. 30 is an explanatory diagram wherein an aorta A is detected using a classifier C;

FIG. 31 is a diagram schematically showing positions PA1 to PA10 of the aorta located respectively for the axial images D₁ to D₁₀ according to Step ST2;

FIG. 32 is a diagram schematically showing a plane FS;

FIG. 33 is a diagram explaining an exemplary method of defining a tracker region; and

FIG. 34 is a diagram schematically showing an example involving cropping an image portion so that a body surface portion of the torso region C1 is not included, and determining vectors α1 to αk.

DETAILED DESCRIPTION

Now embodiments for practicing the invention will be described hereinbelow, although the present invention is not limited thereto.

FIG. 1 is a schematic diagram of a magnetic resonance imaging apparatus in one embodiment of the present invention. The magnetic resonance imaging apparatus (referred to as “MRI apparatus” hereinbelow) 1 comprises a magnet 2, a table 3, a receive coil 4, and a contrast injection apparatus 5.

The magnet 2 has a reception space 21 in which a subject 14 is received. Moreover, the magnet 2 has a superconductive coil 22, a gradient coil 23, and an RF coil 24. The superconductive coil 22 applies a static magnetic field, the gradient coil 23 applies a gradient pulse, and the RF coil 24 applies an RF pulse. A permanent magnet may be employed in place of the superconductive coil 22.

The table 3 has a cradle 3a for carrying the subject 14. It is by the cradle 3a that the subject 14 is carried into the reception space 21.

The receive coil 4 is attached to the subject 14. The receive coil 4 receives magnetic resonance signals from the subject 14. The contrast injection apparatus 5 injects a contrast medium into the subject 14.

The MRI apparatus 1 further comprises a control section 6, a transmitter 7, a gradient power supply 8, a receiver 9, a processing apparatus 10, a storage section 11, an operating section 12, and a display section 13.

The control section 6 receives from the processing apparatus 10 data containing waveform information, the time for application, etc. of the RF pulse and gradient pulse used in a sequence. The control section 6 then controls the transmitter 7 based on the data for the RF pulse, and controls the gradient power supply 8 based on the data for the gradient pulse. The control section 6 also performs control of the start time for injection of the contrast medium by the contrast injection apparatus 5, control of movement of the cradle 3a, etc. While the control section 6 performs control of the contrast injection apparatus 5, transmitter 7, gradient power supply 8, cradle 3a, etc. in FIG. 1, the control of the contrast injection apparatus 5, transmitter 7, gradient power supply 8, cradle 3a, etc. maybe performed by a plurality of control sections. For example, there may be separately provided a control section for controlling the contrast injection apparatus 5, that for controlling the transmitter 7 and gradient power supply 8, and that for controlling the cradle 3a.

The transmitter 7 supplies electric current to the RF coil 24 based on the data received from the control section 6. The gradient power supply 8 supplies electric current to the gradient coil 23 based on the data received from the control section 6.

The receiver 9 applies processing, such as demodulation/detection, to magnetic resonance signals received by the receive coil 4, and outputs the resulting signals to the processing apparatus 10. It should be noted that a combination of the magnet 2, receive coil 4, control section 6, transmitter 7, gradient power supply 8, and receiver 9 constitute the scanning section.

The storage section 11 stores therein programs executed by the processing apparatus 10, and the like. The storage section 11 may be a non-transitory storage medium, such as a hard disk or CD-ROM. The processing apparatus 10 loads a program stored in the storage section 11, and operates as a processor executing processing written in the program. By executing the processing written in the program, the processing apparatus 10 implements several kinds of units. FIG. 2 is a diagram explaining the units the processing apparatus 10 implements.

Image producing unit 101 produces an image in each of a plurality of slices intersecting a body part to be imaged in the subject 14.

Classifying unit 102 classifies the plurality of images into a plurality of classes based on which portion in the imaged body part each of the plurality of images obtained by the image producing unit 101 represents. The class will be discussed later.

Search-region defining unit 103 defines a search region for searching for a blood vessel on an image-by-image basis based on within which of the plurality of classes each of the plurality of images falls.

Detecting unit 104 detects a position of an aorta from within the search region. Plane calculating unit 105 calculates a plane longitudinally cutting the aorta based on positions PA1 to PA10 (see FIG. 31) of the aorta A detected by the detecting unit 104.

Tracker-region defining unit 106 defines a tracker region R_t (see FIG. 33) for detecting a contrast medium.

The MRI apparatus 1 comprises a computer including the processing apparatus 10.

The processing apparatus 10 implements the image producing unit 101 to tracker-region defining unit 106, etc. by loading programs stored in the storage section 11. The processing apparatus 10 may implement the image producing unit 101 to tracker-region defining unit 106 by a single processor, or by two or more processors. The programs executed by the processing apparatus 10 may be stored in a single storage section, or separately in a plurality of storage sections. The processing apparatus 10 constitutes the blood vessel detecting apparatus.

Referring back to FIG. 1, the description will be continued. The operating section 12 is operated by an operator to input several kinds of information to the computer 8. The display section 13 displays several kinds of information. The MRI apparatus 1 is configured as described above.

FIG. 3 is a diagram showing scans performed in the present embodiment, and FIG. 4 is a diagram schematically showing a body part to be imaged. In the present embodiment, a localizer scan LS, a main scan MS, etc. are performed. The localizer scan LS is a scan for acquiring an image used in defining slice positions or a tracker region R_t (see FIG. 33). The tracker region R_t is a region defined for detecting a contrast medium flowing through the aorta A. Subsequent to the localizer scan LS, the main scan MS is performed.

In the main scan MS, a contrast medium is injected into the subject, and a sequence for detecting the contrast medium from the tracker region R_t is repetitively performed. Once a predetermined amount of the contrast medium has been injected into the tracker region, an imaging sequence for acquiring an image of a liver is performed. Now an exemplary flow in performing the localizer scan LS and main scan MS will be described below.

FIG. 5 is a diagram showing an exemplary flow in performing the localizer scan LS and main scan MS. At Step ST1, the localizer scan LS (see FIG. 2) is performed.

FIG. 6 is a diagram explaining the localizer scan LS. The localizer scan LS is a scan for acquiring an image of a body part including a liver and organs surrounding the liver in the subject. FIG. 6 shows slices in performing the localizer scan LS. Although the localizer scan LS acquires axial, sagittal, and coronal images, FIG. 6 shows a plurality of slices (ten slices SL₁ to SL₁₀ here) for acquiring axial images for convenience of explanation.

In performing the localizer scan LS, the control section 6 (see FIG. 1) sends data for an RF pulse in a sequence used in the localizer scan LS to the transmitter 7, and data for a gradient pulse in the sequencer used in the localizer scan LS to the gradient power supply 8. The transmitter 7 supplies electric current to the RF coil 24 based on the data received from the control section 6, while the gradient power supply 8 supplies electric current to the gradient coil 23 based on the data received from the control section 6. Thus, the RF coil 24 applies an RF pulse, while the gradient coil 23 applies a gradient pulse. By performing the localizer scan LS, an MR signal is generated from the body part to be imaged. The MR signal is received by the receive coil 4 (see FIG. 1). The receive coil 4 receives the MR signal and outputs an analog signal containing information on the MR signal. The receiver 9 applies signal processing, such as demodulation/detection, to the signal received from the receive coil 4, and outputs data resulting from the signal processing to the processing apparatus 10.

The image producing unit 101 (see FIG. 2) produces images D₁ to D₁₀ in the slices SL₁ to SL₁₀ based on the data collected by the localizer scan LS (which images will be referred to as “axial images” hereinbelow). In the drawing, axial images D₁, D₄, D₆, and D₁₀ are schematically shown. Since the slices SL₁ to SL₁₀ intersect the aorta A, the axial images D₁ to D₁₀ render cross sections of the aorta A.

An axial image includes three regions. The first one is a region (referred to hereinbelow as “torso region”) HB representing a cross section of the torso of the subject, the second is a region (referred to hereinbelow as “left-arm region”) HL representing a cross section of the left arm of the subject, and the third is a region (referred to hereinbelow as “right-arm region”) HR representing across section of the right arm of the subject. After producing the axial images D₁ to D₁₀, the flow goes to Step ST2.

At Step ST2, the classifying unit 102 (see FIG. 2) classifies the axial images D₁ to D₁₀ into a plurality of classes. Now a method of classifying the axial images D₁ to D₁₀ will be described below. In the following example, a method of classifying the axial images using a Laplacian eigenmaps method will be described.

In the present embodiment, the axial images D₁ to D₁₀ are classified into four classes as defined as follows:

Class I: axial images having a large cross-sectional area of the lungs;

Class II: axial images intersecting the vicinity of a border between the lungs and liver;

Class III: axial images having a large cross-sectional area of the liver; and

Class IV: axial images intersecting the liver and kidneys.

In the present embodiment, before imaging the subject, a classification map is prepared in advance for classifying each of the axial images D₁ to D₁₀ according to within which of the four classes I to IV the axial image falls, and the classification of the axial images is achieved using the classification map. Now an exemplary method of creating the classification map will be described below.

FIGS. 7 to 13 are diagrams explaining a method of creating a classification map. First, a plurality of axial images of the abdomen for use to create a classification map are prepared. FIG. 7 shows an example in which a plurality of axial images DA₁ to DA_n, DA_n+1 to DA_n+a, . . . , DA_j+1 to DA_k of the abdomen are prepared from ‘m’ subjects SU₁ to SU_m. Each axial image displays a torso region C1, a left-arm region C2, and a right-arm region C3. The present embodiment shows an example in which the ‘k’ axial images DA₁ to DA_k are prepared from the ‘m’ subjects SU₁ to SU_m.

Next, a map creator who creates a classification map decides within which of Classes I to IV described above each of the axial images DA₁ to DA_k falls. For example, the axial images DA₁, DA₂, etc. are decided to fall within Class I, while the axial image DA_k, etc. are decided to fall within Class IV.

Next, an image portion in a rectangular region circumscribing the torso region C1 is clopped from each of the axial images DA₁ to DA_k. FIG. 8 is a diagram schematically showing images each obtained by cropping an image portion in a rectangular region circumscribing the torso region C1 from the axial image. In FIG. 8, only the subject SU₁ is chosen from among the subjects SU₁ to SU_m to show images clopped from the axial images for convenience of explanation.

Referring to FIG. 8, images clopped from the axial images DA₁ to DA_n are designated by symbols a₁, a₂, . . . , a_n. Moreover, the size of the clopped images a₁ to an is represented by the number of pixels: v_i by w_i (i=1 to n). Since the rectangular region circumscribes the torso region C1, the left-arm region C2 and right-arm region C3 may be excluded from the clopped images a₁ to a_n.

After cropping the image portion of the torso from the axial images DA₁ to DA_n, the resolution of the clopped images a₁ to a_n is scaled down. FIG. 9 is a diagram schematically showing images after their resolution is scaled down. The scaled-down images are designated by symbols a₁′, a₂′, . . . , a_n′. The scaled-down images a₁′ to a_n′ are scaled down to a resolution (v₀ by w₀) of the same size. For example, v₀=44 and w₀=22. By thus scaling down the images, images a₁′ to a_n′ of the same resolution may be produced.

Next, a vector for each of the scaled-down images a₁′ to a_n′ is determined. FIG. 10 is a diagram showing a vector determined for each of the scaled-down images a₁′ to a_n′.

First, let us consider a vector α₁ for an i-th image a₁′ of the scaled-down images a₁′ to a_n′. The vector α₁ is a vector defined so that its elements are pixel values of pixels in the image a_i′. Representing the number of pixels of the image a_i′ as t (=v₀ by w₀), the vector α_i may be expressed by the following equation:

α_i=(α_i1, α_i2, α_i3, α_i4, α_i5, . . . , α_it)  (1)

where α_i1, α_i2, α_i3, α_i4, α_i5, . . . , α_it denote pixel values of pixels in the image a₁′, so that the vector α_i is expressed by the vector having ‘t’ elements.

Therefore, a vector when i=1, for example, that is, a vector α₁ for the image a₁′, may be expressed by the following equation:

α₁=(α₁₁, α₁₂, α₁₃, α₁₄, α₁₅, . . . , α_1t)  (2)

Referring to FIG. 10, representative vectors α₁, α₂, α₃, α₆, and α_n when i=1, 2, 3, 6, and n are shown.

While FIGS. 8 to 10 explain an example in which the vectors α₁ to α_n for the axial images DA₁ to DA_n for the subject SU₁ are determined, the vector α_i may be determined for axial images for the other subjects SU₂ to SU_m as well according to the method described above. FIG. 11 shows vectors α_i obtained for the subjects SU₁ to SU_m on an axial image-by-axial image basis. In the present embodiment, ‘k’ axial images DA₁ to DA_k are prepared from the subjects SU₁ to SU_m, so that ‘k’ vectors α₁ to α_k are obtained.

After determining these vectors α₁ to α_k, an adjacency matrix MA is determined. FIG. 12 schematically shows the adjacency matrix MA. Each element in the adjacency matrix MA represents a distance d_ij between two vectors.

After determining the adjacency matrix MA, it is weighted by a heat kernel according to the distance d_ij to thereby obtain a matrix W. Then, a diagonal matrix D whose diagonal elements are respective sums of columns in the matrix W is determined, and a Laplacian matrix is obtained based on the matrices W and D.

After obtaining the Laplacian matrix, an eigenproblem for the Laplacian matrix is solved to obtain ‘k’ eigenvectors. After obtaining the ‘k’ eigenvectors, an eigenvector F1 having a smallest eigenvalue, and an eigenvector F2 having a second smallest eigenvalue are identified from among the ‘k’ eigenvectors. Thus, the two eigenvectors F1 and F2 may be used to obtain a map representing a relationship among the axial images DA₁ to DA_k (see FIG. 7) for the subjects SU₁ to SU_m. FIG. 13 schematically shows an example of a resulting map. A horizontal axis of the map CM represents an element of the eigenvector F1, while a vertical axis represents an element of the eigenvector F2. Positions of the axial images DA₁ to DA_k in the map CM may be determined by a coordinate point (f1, f2) (f1: the element of the eigenvector F1, and f2: the element of the eigenvector F2). In FIG. 13, points representing the positions of the axial images DA₁ to DA_k are represented using four marks (a cross, a hollow square, a hollow triangle, and a hollow circle) for convenience of explanation. The cross indicates a point representing the position of an axial image falling within Class I (an axial image having a large cross-sectional area of the lungs). The hollow square indicates a point representing the position of an axial image falling within Class II (an axial image intersecting the vicinity of a border between the lungs and liver). The hollow triangle indicates a point representing the position of an axial image falling within Class III (an axial image having a large cross-sectional area of the liver). The hollow circle indicates a point representing the position of an axial image falling within Class IV (an axial image intersecting the liver and kidneys).

In the present embodiment, the map CM shown in FIG. 13 is used as a classification map for classifying the axial images D₁ to D₁₀ (see FIG. 6) obtained in the localizer scan LS. The classification map CM is stored in the storage section 11 (see FIG. 1).

At Step ST2, the classifying unit 102 classifies the axial images D₁ to D₁₀ obtained in the localizer scan LS into the plurality of classes I to IV based on the classification map CM stored in the storage section 11. Now a procedure of classifying the axial images D₁ to D₁₀ into the plurality of classes I to IV based on the classification map CM will be described referring to FIG. 14.

FIG. 14 is a diagram showing an exemplary flow of Step ST2. Step ST2 has Steps ST21 to ST29. Now Steps ST21 to ST29 will be described one by one.

At Step ST21, the classifying unit 102 selects one axial image from among the axial images D₁ to D₁₀ in the slices SL₁ to SL₁₀. Assume here that the axial image D₁ in the slice SL₁ intersecting the lungs is selected. FIG. 15 schematically shows the selected axial image D₁ in the slice SL₁. In the axial image D₁, a torso region HB in the subject, and in addition, a left-arm region HL and a right-arm region HR are rendered. After selecting the axial image D₁ in the slice SL₁, the flow goes to Step ST22.

At Step ST22, the classifying unit 102 binarizes the axial image D₁ in the slice SL₁. FIG. 16 schematically shows a binary image DB₁ obtained by binarizing the axial image D₁ in the slice SL₁. The classifying unit 102 executes binarization processing so that the torso region HB, left-arm region HL, and right-arm region HR are rendered with a logical value of one, while regions outside of the body are rendered with a logical value of zero. This provides the binary image DB₁ in which regions inside of the body are rendered with a logical value of one and regions outside of the body are rendered with a logical value of zero. A method of binarization that may be employed is, for example, a region growing method as described in “Med. Imag. Tech., Vol. 31, No. 2, March 2013.” After applying binarization, the flow goes to Step ST23.

At Step ST23, the classifying unit 102 executes processing of erasing arm portions from the binary image DB₁.

FIG. 17 is a diagram schematically showing a binary image DC₁ in which the arm portions are erased. At Step ST23, image processing for erasing the left-arm and right-arm regions from the binary image DB₁ is executed. This provides the binary image DC₁ in which the arm portions are erased. A method of erasing arm portions that may be employed is, for example, a method executing erosion processing, region growing processing, and dilation processing, as described in “Med. Imag. Tech., Vol. 31, No. 2, March 2013.” After obtaining the binary image DC₁ in which the arm portions are erased, the flow goes to Step ST24.

At Step ST24, the classifying unit 102 clops an image portion in a rectangular region circumscribing the torso region HB from the axial image D₁ based on the binary image DC₁ obtained at Step ST23. FIG. 18 schematically shows an image obtained by cropping the image portion in a rectangular region circumscribing the torso region HB from the axial image D₁. Since in the binary image DC₁, the torso region has a logical value of one and regions outside of the torso are assigned with a logical value of zero, the torso region HB within the axial image D₁ maybe identified by using the binary image DC₁. Thus, a rectangular region circumscribing the torso region HB may be clopped from the axial image D₁. In FIG. 18, the image clopped from the axial image D₁ is designated by symbol D₁′. Moreover, the size of the clopped image D₁′ is represented by a number of pixels of p₁ by q₁. Since the rectangular region circumscribes the torso region HB, the left-arm region HL and right-arm region HR may be excluded from the rectangular region. After cropping the image portion of the torso region from the axial image D₁, the flow goes to Step ST25.

At Step ST25, the classifying unit 102 scales down the resolution of the clopped image D₁′. FIG. 19 is a diagram schematically showing an image D₁″ after its resolution is scaled down. The scaled-down image D₁″ has been scaled down to a resolution (v₀ by w₀) of the same size as the images a₁′ to a_n′ (see FIG. 10) used in creating the classification map CM. After scaling down the resolution, the flow goes to Step ST26.

At Step ST26, the classifying unit 102 determines a vector for the scaled-down image D₁″ (see FIG. 20).

FIG. 20 is a diagram showing a vector β₁ determined for the scaled-down image D₁″. The vector β₁ is a vector defined so that its elements are pixel values of pixels in the scaled-down image D₁″. The vector i maybe expressed by the following equation:

β=(β₁₁, β₁₂, β₁₃, β₁₄, β₁₅, . . . , β_1t)  (3)

where β₁₁, β₁₂, β₁₃, β₁₄, β₁₅, . . . , β_1t denote pixel values of pixels in the scaled-down image D₁″. After determining the vector β₁, the flow goes to Step ST27.

At Step ST27, the classifying unit 102 determines a distance between each of the ‘k’ vectors α₁ to α_k (see FIG. 11) used to create the classification map CM and the vector β₁ for the axial image D₁. The distance is then weighted with a heat kernel according to the distance value to determine a matrix Y. Next, the two eigenvectors F1 and F2 identified in creating the classification map CM are used to project the matrix Y onto the classification map CM (see FIG. 13). FIG. 21 shows a point E representing the matrix Y projected onto the classification map CM. In FIG. 21, the point E lies at coordinates (f11, f21). In this way, the point E corresponding to the axial image D₁ is mapped onto the classification map CM. After the mapping, the flow goes to Step ST28.

At Step ST28, the classifying unit 102 identifies N points from among a plurality of points defined on the classification map CM in an ascending order of distance to the point E corresponding to the axial image D₁.

FIG. 22 is a diagram explaining a method of identifying N points. FIG. 22 shows the vicinity of the point E mapped onto the classification map CM in an enlarged view.

The classifying unit 102 calculates a distance between each of the plurality of points defined on the classification map CM and the point E, and identifies N points in an ascending order of distance. Thus, N points may be identified in an ascending order of distance to the point E. Assume here that N=7. Therefore, seven points e1, e2, e3, e4, e5, e6, and e7 are identified in an ascending order of distance to the point E. After identifying the seven points e1 to e7, the flow goes to Step ST29.

At Step ST29, the classifying unit 102 first identifies within which of the four classes I to IV each of the seven points e1 to e7 falls. Here, six points (e1, e2, e4, e5, e6, and e7) of the seven points e1 to e7 are identified as falling within Class I, and one point (e3) as falling within Class II. After identifying within which class each point falls, the classifying unit 102 determines how many of the seven points fall within each class, and identifies a class having a largest number of the points falling there. Since six of the seven points e1 to e7 fall within Class I here, Class I is identified. Therefore, the axial image D₁ is decided to fall within Class I of the four classes I to IV, so that the axial image D₁ is classified into Class I.

While the preceding description is addressed to classification of the axial image D₁, the other axial images D₂ to D₁₀ are also classified according to the flow from Step ST21 to Step ST29. Since information on each of Classes I to IV is reflected in the classification map CM, which portion of the imaged body part each of the axial images D₁ to D₁₀ represents may be identified by using the classification map CM. Thus, the axial images D₁ to D₁₀ may be classified into four classes.

FIG. 23 is a diagram showing within which class each of the axial images D₁ to D₁₀ falls. FIG. 23 shows within which class four representative axial images D₁, D₄, D₆, and D₁₀ of the axial images D₁ to D₁₀ fall. The axial image D₁ is classified into Class I, the axial image D₄ into Class II, the axial image D₆ into Class III, and the axial image D₁₀ into Class IV. After classifying the axial images D₁ to D₁₀, the flow goes to Step ST3.

At Step ST3, the search-region defining unit 103 (see FIG. 2) defines an aorta search region for each of the axial images D₁ to D₁₀. Now a method of defining the aorta search region will be described below.

In the present embodiment, a probability distribution model expressing the probability of presence of the aorta for each of Classes I to IV is stored in the storage section 11.

FIG. 24 is a diagram explaining the probability distribution model. The storage section 11 stores therein four probability distribution models M1 to M4.

The probability distribution model M1 represents the probability of presence of the aorta in Class I. The probability of presence of the aorta in Class I may be determined based on the position of the aorta appearing in axial images (for example, axial images DA₁, DA_n+1, DA_j+1) having a large cross-sectional area of the lungs among the plurality of axial images DA₁ to DA_k (see FIG. 7) used to create the classification map CM. A region RM1 in the probability distribution model M1 represents a region having a high probability of presence of the aorta. Referring to an a-a cross section of the probability distribution model M1, it is shown that the probability of presence of the aorta increases from an outer periphery of the region RM1 toward the inside thereof.

The probability distribution model M2 represents the probability of presence of the aorta in Class II. The probability of presence of the aorta in Class II may be determined based on the position of the aorta appearing in axial images (for example, axial images DA₃, DA_n+3, DA_j+3) intersecting the vicinity of a border between the lungs and liver among the plurality of axial images DA₁ to DA_k (see FIG. 7) used to create the classification map CM. A region RM2 in the probability distribution model M2 represents a region having a high probability of presence of the aorta. Referring to an a-a cross section of the probability distribution model M2, it is shown that the probability of presence of the aorta increases from an outer periphery of the region RM2 toward the inside thereof.

The probability distribution model M3 represents the probability of presence of the aorta in Class III. The probability of presence of the aorta in Class III may be determined based on the position of the aorta appearing in axial images (for example, axial images DA₆, DA_n+6, DA_j+6) having a large cross-sectional area of the liver among the plurality of axial images DA₁ to DA_k (see FIG. 7) used to create the classification map CM. A region RM3 in the probability distribution model M3 represents a region having a high probability of presence of the aorta. Referring to an a-a cross section of the probability distribution model M3, it is shown that the probability of presence of the aorta increases from an outer periphery of the region RM3 toward the inside thereof.

The probability distribution model M4 represents the probability of presence of the aorta in Class IV. The probability of presence of the aorta in Class IV may be determined based on the position of the aorta appearing in axial images (for example, axial images DA_n, DA_n+a, DA_k) intersecting the liver and kidneys among the plurality of axial images DA₁ to DA_k (see FIG. 7) used to create the classification map CM. A region RM4 in the probability distribution model M4 represents a region having a high probability of presence of the aorta. Referring to an a-a cross section of the probability distribution model M4, it is shown that the probability of presence of the aorta increases from an outer periphery of the region RM4 toward the inside thereof.

The search-region defining unit 103 defines a search region for searching for the aorta in the axial images D₁ to D₁₀ based on the probability distribution models M1 to M4. Now a method of defining a search region in the axial images D₁ to D₁₀ will be described below. Since the method of defining a search region in the axial images D₁ to D₁₀ is the same for all the axial images, the axial image D₁ is chosen from the axial images D₁ to D₁₀ to describe the method of defining a search region in the following description.

FIG. 25 is a diagram explaining the method of defining a search region. The search-region defining unit 103 selects a probability distribution model used to determine a search region in the axial image D₁ from among the probability distribution models M1 to M4. Since the axial image D₁ is classified into Class I (see FIG. 23), the search-region defining unit 103 selects the probability distribution model M1 corresponding to Class I from among the probability distribution models M1 to M4.

Next, the search-region defining unit 103 multiplies the clopped image D₁′ (see FIG. 18) obtained from the axial image D₁ by the probability distribution model M1 to obtain a multiplied image MD₁′.

After obtaining the multiplied image MD₁′, the search-region defining unit 103 determines a region R1 corresponding to the region RM1 in the probability distribution model M1 from within the multiplied image MD₁′ based on the information on the position of the region RM1 in the probability distribution model M1. The thus-determined region R1 is defined as the aorta search region R₁ in the axial image D₁.

While the method of defining the aorta search region R₁ in the axial image D₁ is explained in FIG. 25, a search region is determined for each of the other axial images D₂ to D₁₀ as well by determining a multiplied image according to a similar procedure. FIG. 26 schematically shows multiplied images MD₁‘ to MD₁₀’ and search regions R₁ to R₁₀ determined for the axial images D₁ to D₁₀. In FIG. 26, multiplied images MD₁′, MD₄′, MD₆′, and MD₁₀′, and search regions R₁, R₄, R₆, and R₁₀ determined for the four representative axial images D₁, D₄, D₆, and D₁₀ among the axial images D₁ to D₁₀ are schematically shown. After determining the search regions R₁ to R₁₀, the flow goes to Step ST4.

At Step ST4, the detecting unit 104 (see FIG. 2) detects a position of the aorta from within the search region R1. Now a method of detecting a position of the aorta will be described below. Since the method of detecting a position of the aorta in the axial images D₁ to D₁₀ is the same for all the axial images, the axial image D₁ is chosen from the axial images D₁ to D₁₀ to describe the method of detecting a position of the aorta in the following description.

At Step ST4, a cerebrospinal fluid CSF is first detected from within the multiplied image MD₁′ for the axial image D₁. A method of detecting a cerebrospinal fluid CSF that may be employed is, for example, one described in “Med. Imag. Tech., Vol. 31, No. 2, March 2013.” FIG. 27 shows a position of the detected cerebrospinal fluid CSF.

After detecting the cerebrospinal fluid CSF, the detecting unit 104 defines lines Ls and Le tangent to the search region R1 with respect to the cerebrospinal fluid CSF. FIG. 28 shows the lines Ls and Le.

After defining the lines Ls and Le, the detecting unit 104 defines a window on the line Ls for use in detecting the aorta A (see FIG. 29).

FIG. 29 is a diagram showing a window W defined. The size of the window W is represented by a number of pixels of a by b. The detecting unit 104 defines the window W on the line Ls with respect to the cerebrospinal fluid CSF. After defining the window W, the detecting unit 104 changes a rotation angle θ of the window W by rotating the window W around the cerebrospinal fluid CSF within a range delineated by the lines Ls and Le, and moreover changes the size (the numbers a and b of pixels) of the window W. It then detects the aorta A from within the search region R1 using a classifier C (see FIG. 30), which will be discussed below.

FIG. 30 is an explanatory diagram wherein the aorta A is detected using a classifier C. The detecting unit 104 changes θ between θ1 and θ2, and changes values of a and b for the window W each time θ is changed. Then, each time any one of the three parameters (θ, a, b) for the window W is changed, data for pixels in the window W is extracted. The classifier C is for outputting an output value Out for deciding whether the possibility that the aorta A in the axial image D₁ inscribes a distal portion of the window W is high or low based on the extracted data, and the classifier C is created in advance before imaging the subject. The classifier C may be created by, for example, learning training data containing information on a signal value for the aorta, and training data containing information on a signal value for a tissue different from the aorta. In the present embodiment, the classifier C is configured so that the output value Out is greater for a higher possibility that the aorta A inscribes the distal portion of the window W. Therefore, it may be decided that the possibility that the aorta A in the axial image D₁ inscribes the distal portion of the window W is highest when the output value Out from the classifier C is largest. Assume here that the output value Out from the classifier C is largest when the rotation angle θ=1, the number a of pixels=a₁, and the number b of pixels=b₁. It is then decided that the aorta A in the axial image D₁ inscribes the distal portion of the window W when the parameters (θ, a, b) for the window W=(θ₁, a₁, b₁), and thus, the aorta A may be detected from within the search region R1.

While the method of detecting the aorta A in the axial image D₁ is explained in the preceding description, the aorta is detected from within the search region using the classifier C for the other axial images D₂ to D₁₀ as well. Thus, the position of the aorta may be located for the axial images. FIG. 31 schematically shows positions PA1 to PA10 of the aorta located according to Step ST2 respectively for the axial images D₁ to D₁₀. The positions PA1 to PA10 of the aorta are not necessarily the center of the aorta, and may lie in the inside of the cross section of the aorta. After locating the position of the aorta in each of the axial images D₁ to D₁₀, the flow goes to Step ST5.

At Step ST5, the plane calculating unit 105 (see FIG. 2) calculates a plane longitudinally cutting the aorta based on the positions PA1 to PA10 of the detected aorta A. Now a method of calculating the plane will be described below.

First, based on the positions PA1 to PA10 of the aorta, the plane calculating unit 105 identifies a region of a cross section of the aorta on an axial image-by-axial image basis. A method of identifying a region of the cross section of the aorta that may be employed is, for example, a segmentation technique such as a Level Set method. After identifying the region of the cross section of the aorta, a center of the cross section of the aorta is located. For example, a centroid of the cross section of the aorta A maybe located as a center of the aorta A. The center of the aorta A may be located by considering the cross section of the aorta A to have a generally circular shape.

After locating the center of the aorta A, the plane calculating unit 105 determines a plane FS longitudinally cutting the aorta A generally in parallel with a direction of the course of the aorta A based on the information on the position of the center of the aorta A located for each of the axial images D₁ to D₁₀. FIG. 32 schematically shows the determined plane FS. In the present embodiment, the plane calculating unit 105 calculates a square sum of the distance between the center of the aorta A located for each of the axial images D₁ to D₁₀ and the plane, and determines as the plane FS a plane when the square sum is minimized. After determining the plane FS, the flow goes to Step ST4.

At Step ST4, a tracker region for detecting a contrast medium is defined. Now a method of defining the tracker region will be described below.

FIG. 33 is a diagram explaining an exemplary method of defining a tracker region. The operator operates the operating section 12 (see FIG. 1) to input a command for displaying the plane FS determined at Step ST3 in the display section 13. Once the command has been input, the plane FS determined at Step ST3 is displayed in the display section 13.

After the plane FS is displayed, the operator operates the operating section 12 to input information for defining a tracker region referring to a positional relationship among organs and the aorta displayed in the plane FS. Once the information has been input, the tracker-region defining unit 106 (see FIG. 2) defines a tracker region based on the input information. FIG. 32 shows a case in which a tracker region Rt is defined to lie in the inside of the aorta A. It should be noted that the tracker region Rt may be automatically defined based on the position of the aorta A. After defining the tracker region R_t, the flow goes to Step ST5.

At Step ST5, a main scan MS (see FIG. 3) is performed. In the main scan MS, the subject is injected with a contrast medium, and a sequence for detecting the contrast medium from the tracker region R_t is repetitively performed. Once a predefined amount of the contrast medium has been injected in the tracker region R_t, a scan for acquiring an image of the liver is performed, and the flow is terminated.

In the present embodiment, within which class an axial image falls is determined, and a probability distribution model corresponding to that class is used to determine an aorta search region. Therefore, detection of an aorta may be achieved while focusing upon a region having a high probability of presence of the aorta, thus improving accuracy of detection of the aorta.

Moreover, in the present embodiment, an image portion is clopped from each of the axial images D₁ to D₁₀ so as to circumscribe the torso region HB, and classification of the axial images D₁ to D₁₀ is performed. Therefore, the axial images D₁ to D₁₀ may be classified without being affected by the signal value in the left-arm region HL and right-arm region HR, thus fully reducing a risk that the axial images are wrongly classified.

While the window W is defined with respect to the cerebrospinal fluid CSF in the present embodiment, the window W may be defined with respect to a body part different from the cerebrospinal fluid CSF. Moreover, while the aorta A is detected in the present embodiment, the present invention may be applied to a case in which a blood vessel different from the aorta A is to be detected.

While the present embodiment detects an aorta based on an axial image, the aorta may be determined based on an image in a plane other than the axial plane (for example, an oblique plane intersecting the axial plane at an angle).

While in the present embodiment the window W is rectangular, it may have a different shape (an elliptical shape, for example).

In creating the classification map CM in the present embodiment, an image portion in a rectangular region circumscribing the torso region C1 is clopped from each of the axial images (see FIG. 8) to determine vectors α₁ to α_k (see FIG. 11). However, the vectors α₁ to α_k may be determined by cropping the image portion so that the center portion of the torso region C1 is included and a body surface portion is not included. FIG. 34 is a diagram schematically showing an example involving cropping an image portion so that the body surface portion of the torso region C1 is not included, and determining the vectors α₁ to α_k. According to the method in FIG. 34, since the vectors α₁ to α_k are determined based on the image portion (hatched portion) including the center portion of the torso region C1 and not including the body surface portion, a classification map CM′ (not shown) may be created with a reduced effect of fat in the body surface portion. When the classification map CM′ is created in place of the classification map CM (see FIG. 13), again, the axial images maybe classified by performing Step ST2. When classifying axial images using the classification map CM, an image portion is clopped from the axial image D₁ so as to circumscribe the torso region HB at Step ST24. When classifying axial images using the classification map CM′ in place of the classification map CM, however, an image portion may be clopped from the axial image D₁ so as not to include the body surface portion of the torso region HB at Step ST24.

Furthermore, in the present embodiment, the classification map is created by cropping the image portion so as not to include the left-arm and right-arm regions. However, the classification map may be created by cropping an image portion so as to include the left-arm and right-arm regions insofar as axial images are correctly classified.

While the method of classifying axial images using the Laplacian eigenmaps method is described in the present embodiment, the classification method in the present embodiment is exemplary, and axial images may be classified using any other technique.

Claims

1. A blood vessel detecting apparatus comprising:

an image producing unit for producing a plurality of images in a plurality of slices defined in a body part to be imaged containing a blood vessel;

a classifying unit for classifying said plurality of images into a plurality of classes based on which portion in said imaged body part each of said plurality of images represents; and

a defining unit for defining a search region for searching for said blood vessel from within said image based on within which of said plurality of classes said image falls.

2. The blood vessel detecting apparatus as recited in claim 1, wherein a model representing a probability of presence of said blood vessel is made corresponding to each said class; and said defining unit defines said search region based on said model.

3. The blood vessel detecting apparatus as recited in claim 1, wherein said plurality of classes includes

a first class corresponding to images having a large cross-sectional area of lungs;

a second class corresponding to images intersecting a vicinity of a border between the lungs and a liver;

a third class corresponding to images having a large cross-sectional area of the liver; and

a fourth class corresponding to images intersecting the liver and kidneys.

4. The blood vessel detecting apparatus as recited in claim 1, wherein said classifying unit classifies said plurality of images into a plurality of classes based on a map for identifying which portion of said imaged body part each of said plurality of images represents.

5. The blood vessel detecting apparatus as recited in claim 4, wherein said classifying unit determines a vector whose elements are pixel values of pixels in said image, maps a point corresponding to said image onto said map based on said vector, and determines within which of said plurality of classes said image falls based on a position at which said point is mapped in said map.

6. The blood vessel detecting apparatus as recited in claim 5, wherein each of said plurality of images includes a first region representing a cross-section of a torso of a subject, a second region representing a cross-section of a left arm of said subject, and a third region representing a cross-section of a right arm of said subject; and said classifying unit crops from said image an image portion that includes said first region and does not include said second region nor said third region, and determines said vector based on an image obtained by cropping said image portion.

7. The blood vessel detecting apparatus as recited in claim 5, wherein each of said plurality of images includes a first region representing a cross-section of a torso of a subject, a second region representing a cross-section of a left arm of said subject, and a third region representing a cross-section of a right arm of said subject; and said classifying unit crops from said image an image portion that includes a central portion of said first region and does not include a body surface portion of said first region, said second region, nor said third region, and determines said vector based on an image obtained by cropping said image portion.

8. The blood vessel detecting apparatus as recited in claim 1, comprising a detecting unit for detecting a position of said blood vessel from said search region.

9. A method comprising:

producing a plurality of images in a plurality of slices defined in a body part to be imaged containing a blood vessel;

classifying said plurality of images into a plurality of classes based on which portion in said imaged body part each of said plurality of images represents; and

defining a search region for searching for said blood vessel from within said image based on within which of said plurality of classes said image falls.

10. The method as recited in claim 9, further comprising making a model representing a probability of presence of said blood vessel corresponding to each said class; and defining said search region based on said model.

11. The method as recited in claim 9, wherein said plurality of classes includes:

a first class corresponding to images having a large cross-sectional area of lungs;

a second class corresponding to images intersecting a vicinity of a border between the lungs and a liver;

a third class corresponding to images having a large cross-sectional area of the liver; and

a fourth class corresponding to images intersecting the liver and kidneys.

12. The method as recited in claim 9, wherein said classifying includes classifying said plurality of images into a plurality of classes based on a map for identifying which portion of said imaged body part each of said plurality of images represents.

13. The method as recited in claim 12, wherein said classifying includes:

determining a vector whose elements are pixel values of pixels in said image;

mapping a point corresponding to said image onto said map based on said vector; and

determining within which of said plurality of classes said image falls based on a position at which said point is mapped in said map.

14. The method as recited in claim 13, wherein each of said plurality of images includes a first region representing a cross-section of a torso of a subject, a second region representing a cross-section of a left arm of said subject, and a third region representing a cross-section of a right arm of said subject; and said classifying includes cropping from said image an image portion that includes said first region and does not include said second region nor said third region, and determining said vector based on an image obtained by cropping said image portion.

15. The method as recited in claim 13, wherein each of said plurality of images includes a first region representing a cross-section of a torso of a subject, a second region representing a cross-section of a left arm of said subject, and a third region representing a cross-section of a right arm of said subject; and said classifying includes cropping from said image an image portion that includes a central portion of said first region and does not include a body surface portion of said first region, said second region, nor said third region, and determining said vector based on an image obtained by cropping said image portion.

16. The method as recited in claim 9, further comprising detecting a position of said blood vessel from said search region.

17. A magnetic resonance imaging (MRI) apparatus comprising:

a scanning section for performing a scan on a body part to be imaged containing a blood vessel; and

a processing apparatus configured to: produce a plurality of images in a plurality of slices defined in the body part to be imaged containing the blood vessel; classify said plurality of images into a plurality of classes based on which portion in said imaged body part each of said plurality of images represents; and define a search region for searching for said blood vessel from within said image based on within which of said plurality of classes said image falls.

18. The MRI apparatus as recited in claim 17, wherein said plurality of classes includes:

a first class corresponding to images having a large cross-sectional area of lungs;

a second class corresponding to images intersecting a vicinity of a border between the lungs and a liver;

a third class corresponding to images having a large cross-sectional area of the liver; and

a fourth class corresponding to images intersecting the liver and kidneys.

19. The MRI apparatus as recited in claim 17, wherein said classifying includes classifying said plurality of images into a plurality of classes based on a map for identifying which portion of said imaged body part each of said plurality of images represents.

20. The MRI apparatus as recited in claim 19, wherein said classifying includes:

determining a vector whose elements are pixel values of pixels in said image;

mapping a point corresponding to said image onto said map based on said vector; and

determining within which of said plurality of classes said image falls based on a position at which said point is mapped in said map.