JPH0773291A - Image processor - Google Patents

Abstract

PURPOSE:To improve the area recognizing accuracy of an entire two-dimensional pattern by acquiring the information on a reference area through the preprocessing that is carried out before an area which is hard to extract out of the two-dimensional pattern is recognized. CONSTITUTION:An image processor is provided with a reference tissue image extracting part 4 which extracts the two-dimensional pattern of a reference tissue through several preprocessings and with the image data on a single tomographic surface used as the input, and a recognition processing part 6 which carries out the recognition of the two-dimensional pattern through several preprocessings and based on the reference tissue image data received from the part 4 and the knowledge stored in a knowledge file 5 with the image data on the tomographic surface received from a continuous tomographic surface image data file 2-1 (2-2) used as the input. Then the part 6 also recognizes the two-dimensional pattern of the contour of a three-dimensional image for each tomographic surface, and the output of this recognition processing result shows the position of the contour of a relevant slice surface with high precision. Then the two-dimensional pattern of a relevant area is obtained from the recognition processing result.

Description

Detailed Description of the Invention

[0001]

The present invention relates to MRI-CT and X-ray C
T, an image processing apparatus for recognizing a two-dimensional pattern of a contour of a three-dimensional image based on image data of a plurality of continuous tomographic planes obtained from an ultrasonic diagnostic apparatus, etc. The present invention relates to an image processing apparatus capable of efficiently extracting feature pattern information representing the contour of the image at high speed.

In the diagnosis of organs such as the heart and the liver, it is strongly desired to create a three-dimensional contour image of the organ from an image of a living body obtained from a diagnostic device in order to improve diagnosis accuracy. The present invention provides an image processing device that automatically and effectively creates an effective three-dimensional image from image data of each tomographic plane with high accuracy.

[0003]

2. Description of the Related Art Recently, a technique for making a diagnosis by making image data of a continuous tomographic plane of a living body using an MRI-CT, an X-ray CT and an ultrasonic diagnostic apparatus has been widely used. . In a conventional image processing apparatus for processing these data, a two-dimensional contour pattern of an organ or a tissue is recognized on each tomographic plane, and a recognition result for each tomographic plane is combined to create a three-dimensional image. However, it was not easy to automatically extract the three-dimensional organ contour with high accuracy and speed from the image data of the actual continuous tomographic plane. This is because in image processing performed without considering the anatomical form of the tissue, there are many contour patterns that are erroneously recognized or unrecognizable. For this reason, conventionally, a specialist doctor inputs the two-dimensional organ contour read for each tomographic image into a computer and extracts the three-dimensional organ contour.

[0004]

A conventional image processing apparatus for creating a three-dimensional contour image from images of a large number of continuous tomographic planes in which the absolute position of the tomographic plane in the vertical direction is regular (for example, at regular intervals). Then, when a contour image of an organ of a living body is targeted, it is difficult to obtain a practical level with the ability of the contour recognition processing function of a normal image processing apparatus. Therefore, it has been necessary to manually read a continuous tomographic image by a specialist (to be referred to as a manual method), identify the contour of the intended organ, and input the image data again.

However, in the method of inputting the image data of the organ contour in the image of the continuous tomographic plane by the manual method, there is a problem that the reading and decoding of the image and the input processing take time, and therefore the number of data to be processed is limited. It was

The present invention uses 3 from the image data of the continuous tomographic plane.
It is an object of the present invention to improve the accuracy of the processing for automatically creating a three-dimensional contour image and realize an image processing apparatus capable of high-speed operation at a practical level.

[0007]

According to the present invention, in a recognition process of a two-dimensional pattern of a region of interest such as an organ contour for image data of each continuous tomographic plane, a certain region in the two-dimensional pattern is a luminance region. Since all or any of the shapes and positions have characteristic properties, the area (hereinafter referred to as a reference area) can be easily extracted from the two-dimensional pattern by a simple process, or the area can be easily extracted. When the information about the reference area can be determined to some extent, the information about the reference area is acquired by the pre-processing before the recognition processing of the area that is difficult to extract in the two-dimensional pattern, and the position information of the reference image is extracted from the original area that is difficult to extract. By adding it to the luminance / morphology information, the area recognition accuracy of the entire two-dimensional pattern is improved.

FIG. 1 is a block diagram of the principle of the present invention, which is shown using an exemplary method. In FIG. 1, 1
Is a tomographic apparatus capable of generating image data of a continuous tomographic plane such as MRI-CT.

Reference numeral 2-1 is a continuous tomographic plane image data file for storing image data of each continuous tomographic plane output from the tomographic apparatus 1. Reference numeral 2-2 is a continuous tomographic plane image data file that stores image data of each continuous tomographic plane output from the tomographic apparatus 1 at the same position under different imaging conditions.

3-1 to 3-N are continuous tomographic image data stored in the file. Image data is
The density level of the image is expressed in pixel units. Four
Is a reference tissue image extraction unit that inputs image data of one tomographic plane and extracts a two-dimensional pattern of the reference tissue through some preprocessing.

The reference numeral 5 indicates a recognition processing unit 6 which will be described later.
It is a knowledge file that defines a method of recognizing input image data. 6 is a continuous tomographic image data file 2-
It is a recognition processing unit that performs recognition based on the knowledge of the knowledge file 5 together with the reference tissue image data from the reference tissue image extraction unit 4 through some pre-processing with the image data of the tomographic plane from 1 or 2-2 as input. .

Multivalued image data of a plurality of (N) continuous tomographic planes output from the tomographic apparatus 1 at regular intervals are stored in the continuous tomographic plane image data files 2-1 and 2-2. It is stored separately as indicated by 1 to 3-N. The tomographic plane image data 3-1 to 3-N of the continuous tomographic plane image data file 2-2 is the differential processing for emphasizing the edge, the distance distribution from the continuous pixel portion of the contour lines in the same tomographic plane and the adjacent tomographic plane. Is input to the recognition processing unit 6 after preprocessing such as processing for obtaining

The tomographic plane image data 3-1 to 3-N of the continuous tomographic plane image data file 2-1 are the same as the tomographic plane image data 3-1 to 3-N of the continuous tomographic plane image data file 2-2. An image captured under the image capturing condition or a different image capturing condition and in which the reference tissue is easily extracted is used. Tomographic plane image data 3-1 to 3-N of the data file 2-1
Is input to the post-reference tissue image extraction unit 4 which has been preprocessed,
The reference tissue region is extracted by performing binarization with an appropriate threshold value.

When X-ray CT is used in the tomographic apparatus, the tomographic plane image data 3-1 to 3-1 to extract the reference tissue region
As 3-N, one having the same imaging condition as that directly input to the recognition processing unit 6 is selected.

The reference tissue image extraction unit 4 determines an area occupied by the reference tissue, creates binarized image data depending on whether the pixel is inside or outside the reference tissue in the image data. Image data in which the distance distribution from the contour portion is calculated for each pixel is obtained. These two image data are input to the recognition processing unit 6 as the reference image data.

The recognition processing section 6 has a function of recognizing a two-dimensional pattern of the contour of a three-dimensional image for each tomographic plane.
The hardware is configured to perform parallel processing by multiplexing for each tomographic plane and for each recognition target pattern of the item of interest, or configured to perform sequential processing using a smaller number of hardware. May be.

The recognition processing result output from the recognition processing unit 6 represents a highly accurate position of the contour line on the slice plane, and a two-dimensional pattern of the region of interest is obtained from the result. In the configuration of the example of FIG. 1, two types of continuous tomographic plane screen file groups 2-1 and 2-2 are used, but if one type can extract the reference tissue, it is input to the recognition processing unit 6. Only one may be used.

The two-dimensional pattern output from the recognition processing unit 6 may be input to the reference tissue image extraction unit 4 for the adjacent surface and used for extracting the reference tissue of the adjacent surface.

[0019]

For example, when it is desired to know the three-dimensional contour image of the liver in the diagnosis of the abdomen of a living body, the image data of the continuous tomographic planes of the MRI-CT output includes not only the liver but also the abdominal wall, stomach, vertebral body (vertebral body). ), Image information of abdominal aorta and other organs and tissues are included. FIGS. 2A and 2B show examples of images of such continuous tomographic planes. FIG. 2A shows the nth tomographic plane, and FIG. 2B shows the (n + 1) th tomographic plane. It represents. In the (n) th slice of (a), the abdominal wall, vertebral body, and abdominal aorta regions can be clearly recognized in addition to the liver of interest, and (n) th slice of the (n) th slice is further recognized. The presence of the gastric region is visible.

Even in a group of images of tissues or organs that are difficult to read, certain specific tissues and organs are located at positions of their occupied areas.
The shape or brightness is characteristic, and it is often possible to predict the occupied area in a certain range between different images. In the present invention, by utilizing such a property, the region extraction is performed for other regions such as soft tissue whose shape and position are difficult to predict.

Focusing on the liver as an item of interest, for example, the characteristics of the brightness and shape shown by the liver in the image are not clear enough to completely discriminate the adjacent peripheral tissues.
For areas where automatic recognition of such soft tissues is difficult,
Conventionally, it relied on the work of interpretation by an expert, but in the present invention, in the image information of organs and tissues other than the item of interest, one that can easily perform contour extraction, and if information about the existing position is easily obtained, The image position information of the area is acquired in advance and used as information when extracting the area that is difficult to recognize as the reference tissue.

FIG. 3 shows an example of the recognition processing according to the present invention, in which the abdominal wall shown by hatching,
The vertebral body and the abdominal aorta are respectively acquired as reference tissues in the recognition processing target area in advance, the existence possibility of other region extraction difficult regions based on the existence region of the reference tissue, whether or not they are included in the reference tissue region, and The distance distribution from the edge of the reference tissue region is represented and set on the input image.

Note that, for example, as shown in FIG. 2A, an image of a tissue such as the abdominal wall, vertebral body, or abdominal aorta is easy to discriminate from other organs or tissues, and these tomographic planes can be recognized with high accuracy. When the above is obtained, the recognition result can be applied to another tomographic plane as a reference image on the tomographic plane to facilitate the recognition process of the difficult-to-recognize portion. This is because in the case of the abdominal wall, vertebral body, and abdominal aorta, the position and size do not change so much depending on the slice plane.

[0024]

Embodiments of the image processing apparatus according to the present invention will be described below. FIG. 4 is a basic configuration diagram of hardware of the embodiment of the present invention for extracting a multi-slice contour image of a three-dimensional image of interest such as an organ using a fuzzy element. The components denoted by 7 to 11 in the figure are the same as those in a normal personal computer, 7 is a CPU, 8 is a main memory MS.
U and 9 are file control units FCU, 10 is a work file device, and 11 is a system bus. Further, 12 to 15 are components specially added for image processing, 12 is a frame memory board, 13 is a fuzzy inference board, 14 is a fuzzy element, 15 is a memory, and 1 is a memory.
Reference numeral 6 is an original image file device, and 17 is a video signal output device.

The image data of each continuous tomographic plane output from the tomographic device or the like is recorded in the original image file device 16 of FIG. 4 in an analog video signal format. The image data of each tomographic plane read from the original image file device 16 via the video signal output device 17 for creating a three-dimensional image is
It is input to the frame memory board 12, A / D-converted, stored in the frame memory in units of tomographic planes in a multivalued digital signal format, and further stored in the work file device 10 via the FCU 9. F1 to F7 shown in the work file device 10 are files created for work.

The fuzzy inference board 13 realizes the function of the recognition processing section 6 in FIG. 1 by the fuzzy inference function using the knowledge file 5, and a large number of fuzzy elements 14 are arranged for each tomographic plane and each item of interest. It is a thing.
The memory 15 is also used to store inference rules, membership functions, operation data, etc. for performing fuzzy operations. Inference rules and membership functions are CP
It is created on the U7 side and supplied to the fuzzy control board 13.

The details of the structure of the files F1 to F7 shown in the work file device 10 are shown in FIG. The contents of each file are as follows. F
Reference numeral 1 is an original image file that holds the original image data from the frame memory board 12.

F2 is an original image file of a copy of F1. If the value of F2 is a high value or a low value in the pixel of interest, the inference rule and the membership function are created so that the possibility of the contour candidate point in that column is increased. Whether the value is the high value or the low value depends on the type of image and region of interest used.

F3 is a differential image file that stores image data in which the edge of the image is emphasized by performing differential processing with the original image data of F2. If the value of F3 is a high value at the pixel of interest, the inference rule and the membership function are created so as to increase the possibility of the contour candidate point in the column.

F4 is an intra-layer continuity file that stores image data showing the result of detecting the continuity of the region of interest in the same tomographic plane, and has the information of the distance distribution from the contour candidate points in the processed column. With.

F5 is an inter-layer continuity file that stores image data showing the result of detecting the continuity of the region of interest with the adjacent tomographic plane, and the distance distribution from the contour candidate points on the processed adjacent plane. With information that has.

If the values of the files F4 and F5 described above are low in the pixel of interest, that is, the continuity is high, the inference rule and the membership function are created so as to increase the possibility of the contour candidate points in the column.

F6 stores the reference image data in which the contour line and the occupied area of the reference tissue image which are easy to extract are extracted from the original image file F1 and binary values are given to the pixels inside and outside thereof. This is a reference tissue contour file. If the value of F6 is a numerical value given in the reference tissue region in the pixel of interest, the inference rule and the membership function are created so that the possibility of being a contour candidate point in that column becomes low.

F7 is a reference tissue region file that represents the distance distribution from the reference tissue edge by giving the distance values from the edge of the occupied area of the reference image to all the tomographic plane pixels. The inference rule and the membership function regarding F7 vary depending on the values of F2 to F6. However, when the pixel of interest is determined to be within the reference tissue in F6, the value of F7 is large, that is, the pixel of interest is the center of the reference tissue. When it is close to, the inference rule and the membership function are created so that the pixel of interest is unlikely to be a contour candidate point in the column.

F8 is an output file for storing the image data of the fuzzy inference result. Here, as a specific example, FIG. 6 shows a processing flow when the X-ray-CT is used in a tomographic apparatus. The processing flow of FIG. 6 is the same as the original image file F1.
Read the original image data from the
7 shows the process of creating 7 and transferring it to the fuzzy inference board 13 of FIG. 4 for inference processing. The configuration of the embodiment of FIG. 4 will be referred to as necessary. For convenience of explanation, the case where the region of interest is the liver in the abdominal tomographic (transverse cross section) image will be described. On the tomographic image of this region, the feature is that the brightness of the hard tissue in the tomographic image is high.

Therefore, in this continuous tomographic image, vertebral bodies and ribs are considered as reference tissues as hard tissues, and regions are extracted by threshold processing. Further, the reference tissue is isolated and smoothed. Then, after removing the noise of the isolated points on the tomographic image screen, the obtained image is used as the image of the reference tissue. Then, the numerical value 1 is given to the pixels within the contour of the reference tissue (including the contour line), that is, within the area, and the numerical value 0 is given to the pixels outside the contour, that is, outside the area, and the area is binarized with the contour line of the reference tissue. . Let the file be F6.

Next, the distance from the contour line of the reference tissue to the target pixel is calculated, and the distance distribution is given numerical values inside and outside the reference tissue. The distance distribution file is set to F7.

The files F2 to F7 are the files F1.
4 is executed by the CPU 7 of FIG. 4 executing the corresponding image processing program. The image data of the files F2 to F7 of the work file device 10 are processed by the fuzzy inference board 13 before the fuzzy inference processing is started.
Transferred to. Similarly, a file (not shown) of inference rules and membership functions is transferred from the work file device 10 to the fuzzy inference board 13.

The CPU 7 side controls each fuzzy element 14 of the fuzzy inference board 13 to execute the fuzzy inference. The result of the fuzzy inference that has been performed up to the defuzzify processing is returned to the MSU 8. From the accuracy of the output target pixel as a contour candidate point, the CPU 7 determines the contour line candidate point for each line of the screen. By performing this series of processing over the entire screen, image data of a two-dimensional pattern representing the contour of the region of interest in which the candidate points are continuous can be obtained. The image data of this two-dimensional pattern is stored in the output file F8, and then used to create a three-dimensional image.

Next, as another specific example, FIG.
The processing flow of an example when CT is used for a tomography apparatus is shown.
The case where the region of interest is the liver in the abdominal tomographic (transverse) image will be described. An image obtained by a method of emphasizing a blood vessel at the same position as that obtained by an ordinary method in MRI is prepared.
The continuous tomographic image in the blood vessel enhancement has a feature that the brightness of the abdominal aorta and the inferior vena cava is high.
Therefore, in the blood vessel-enhanced image, the abdominal aorta and inferior vena cava are considered as reference tissues, and region extraction by threshold processing is performed. Further, the reference tissue is isolated and smoothed. Then, after removing the noise of the isolated points on the tomographic image screen, the obtained image is used as the image of the reference tissue. Then, a numerical value 1 is given to pixels within the contour (including on the contour line) of the reference tissue, that is, within the area, and a numerical value 0 is given to pixels outside the contour, that is, outside the area.
And the region is binarized by the contour line of the reference tissue. Let the file be F6.

Next, the distance from the contour line of the reference tissue to the target pixel is calculated, and the distance distribution is given numerical values inside and outside the reference tissue. The distance distribution file is set to F7.

The subsequent processing is the same as in the case of X-ray-CT in FIG. When creating the image data of the reference tissue for each tomographic plane, the creation process can be simplified by utilizing the characteristics of the image of the reference tissue. For example, as in the case of the vertebral body shown in FIG. 8A, when the position of the image of the reference tissue in each tomographic plane is almost unchanged,
The image position of the vertebral body extracted on a specific tomographic plane can be recorded and used at the time of image recognition processing on each tomographic plane. When the tissue has a linear shape as in the example of the abdominal aorta shown in FIG. 8B, the abdomen is formed by twelve tomographic planes j and k (for example, the tomographic planes at both ends) that are separated at regular intervals. The image area of the abdominal aorta can be obtained by measuring the position and size of the aorta and interpolating the values on the intermediate tomographic plane from these values. Furthermore, if it is known that the image of a certain tissue has higher brightness than the images of other parts,
It is possible to scan the image data of each tomographic plane using an appropriate threshold value in advance, extract only the information of the image for each tomographic plane, and use it.

FIG. 9 partially shows an example of a fuzzy inference rule used in the embodiment of the present invention.
These rules are the characteristics of the pixel at the point of interest in the processing of certain tomographic image data, such as high and low brightness, high and low difference values, high and low intra-layer continuity, high and low inter-layer continuity, inside and outside the reference image area, We show conditions of fuzzy inference that takes each value of distance from the edge of the reference image and determines the probability of that point being a contour candidate point based on these values. The fuzzy inference rules are optimally set according to the characteristics of the image that requires interpretation.

[0044]

The present invention has the following advantages over the prior art. (1) It is possible to substitute the interpretation work by the doctor by causing the computer to perform the interpretation.

(2) Since the doctor does not read the image, the continuous tomographic image output from the medical image device can be processed at any time. (3) High-speed automatic contour extraction of target tissue, which is an essential requirement for performing automatic three-dimensional image reconstruction, because the continuity of organs can be grasped at appropriate slice intervals in various continuous tomographic images of the human body. It will be one method. Further, for a tissue whose shape is not easy to be extracted, more accurate extraction can be performed by considering the general anatomical form of the tissue.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is an explanatory diagram of an image example of a tomographic plane.

FIG. 3 is an explanatory diagram of an example of recognition processing by a reference organization according to the present invention.

FIG. 4 is a hardware configuration diagram of an embodiment of the present invention.

FIG. 5 is a flow chart of file input processing according to an embodiment of the present invention.

FIG. 6 is a processing flow chart of an example in which X-ray CT is used in a tomographic apparatus.

FIG. 7 is a process flow diagram of an example in which MRI-CT is used in a tomographic apparatus.

FIG. 8 is an explanatory diagram of an example of creating two-dimensional pattern information using the characteristics of the image of the reference tissue according to the first embodiment of the present invention.

FIG. 9 is an explanatory diagram of an example of a fuzzy inference rule used in the present invention.

[Explanation of symbols]

 1 tomographic device 2-1 and 2-2 continuous tomographic image data file 3-1 to 3-N continuous tomographic image data 4 reference tissue image extraction unit 5 knowledge file 6 recognition processing unit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G06T 9/20 15/00 8420-5L G06F 15/66 B 7459-5L 15/70 335 Z 9192- 5L 15/72 450 K

Claims (2)

[Claims] 1. An image processing apparatus for performing image processing for inputting image data of a plurality of continuous tomographic planes and extracting a two-dimensional pattern forming a contour of a three-dimensional image for each tomographic plane. A recognition processing unit that performs a process of identifying a two-dimensional pattern forming a contour of a three-dimensional image of a predetermined type for each tomographic plane with respect to the data, and a predetermined to which the image data of the tomographic plane is input. And a reference tissue image extraction unit that extracts a region of a two-dimensional pattern of the reference tissue, wherein the recognition processing unit recognizes a two-dimensional pattern forming a contour of a three-dimensional image of a predetermined type for each tomographic plane. In this case, the image processing apparatus is characterized by including, as a recognition condition, a relative positional relationship based on the two-dimensional pattern region of the reference tissue output from the reference tissue image extraction unit. 2. The image processing apparatus according to claim 1, wherein at least the recognition processing unit is realized by using a fuzzy element.