JP2020185031A - Image processing program, image processing device, and image processing method - Google Patents

Abstract

To accurately detect an upper end of a lung field area based on a medical image.SOLUTION: An image processing device 10 includes a contour line extraction part 11 and an area determination part 12. The contour line extraction part 11 extracts a contour line 2 from a binary image 1a indicating a coronal cross section of a chest, generated based on one or more cross sectional images taken by capturing the inside of the chest of a human body. The area determination part 12 searches convex apexes 4a and 4b in the contour line 2 by tracing the contour line 2 in an upper direction from the outside of both right and left sides in the human body. The area determination part 12 determines the positions of the convex apexes 4a and 4b that have been searched as the upper ends of a lung field area.SELECTED DRAWING: Figure 1

Description

The present invention relates to an image processing program, an image processing apparatus, and an image processing method.

Medical images such as CT (Computed Tomography) images are widely used for diagnosing lung diseases. In diagnostic imaging using medical images, the doctor has to interpret a large number of images, which imposes a heavy burden on the doctor. Therefore, there is a demand for a technique for supporting a doctor's diagnostic work in some way by a computer.

For example, in order to identify the region to be read from the medical image or narrow down the medical image to be read, a technique for specifying the lung field region based on the medical image has been proposed. As an example, a technique has been proposed in which a projected image in which the internal state of the chest is projected to the front by a minimum value projection method based on a plurality of CT images is generated, and the lung field region is extracted from the projected image. In addition, as another example, a technique for identifying an image of the axial surface at the upper end of the diaphragm necessary for diagnosing emphysema by detecting the position of the apex on the head side in the bottom region of the lung field based on the pixel value of the three-dimensional medical image. Has been proposed.

Further, as a related technique, a technique for identifying the end portion of the lung field region in this image has been proposed by using a total value profile based on a chest X-ray image.

Japanese Unexamined Patent Publication No. 2012-096025 JP-A-2017-104277

Toshihiro Tomita, 7 outsiders, "Extraction of lung field region of chest CT image by model information and minimum value projection", MEDICAL IMAGING TECHNOLOGY Vol.15 No.2, March 1997, pp. 164-174

By the way, near the upper end of the lung field region, a trachea exists between the right lung and the left lung. In the above-mentioned projected image using the coronal cross-sectional image and the minimum value projection method, it is difficult to distinguish the tracheal region from the lung field region, so that it is difficult to accurately detect the upper end of the lung field region. is there.

In one aspect, it is an object of the present invention to provide an image processing program, an image processing apparatus and an image processing method capable of detecting the upper end of a lung field region with high accuracy based on a medical image.

One idea is to extract contour lines from a binarized image showing the coronal section of the chest, which is generated by a computer based on one or more cross-sectional images of the inside of the chest of the human body, and both left and right in the human body. An image processing program is provided that searches for convex vertices in the contour line by tracing the contour line from the outside to the upward direction, and identifies the position of the searched convex apex as the upper end of the lung field region. To.

Further, in one plan, a contour line extraction unit that extracts a contour line from a binarized image showing a coronal cross section of the chest, which is generated based on one or more cross-sectional images of the inside of the chest of the human body, It has a region identification part that searches for a convex apex in the contour line by tracing the contour line upward from both the left and right sides of the human body and specifies the position of the searched convex apex as the upper end of the lung field region. An image processing device is provided.

Further, one proposal provides an image processing method in which a computer executes a process similar to the process based on the image processing program.

In one aspect, the upper end of the lung field region can be detected with high accuracy based on medical images.

It is a figure which shows the configuration example and the processing example of the image processing apparatus which concerns on 1st Embodiment. It is a figure which shows the structural example of the image processing system which concerns on 2nd Embodiment. It is a figure which shows the organ arrangement of a chest. It is a block diagram which shows the structural example of the processing function provided in the image processing apparatus. It is a figure for demonstrating the generation process of the front projection image. It is a figure which shows the internal structure example of the convex vertex detection part. It is the first figure which shows the processing example of the region upper end extraction part. It is a 2nd figure which shows the processing example of the region upper end extraction part. It is a 3rd figure which shows the processing example of the region upper end extraction part. This is an example of a flowchart showing the entire processing by the image processing device. This is an example of a flowchart showing the process of extracting the upper end of the lung field region. This is an example of a flowchart showing the vertex detection process. It is a figure which shows the internal structure example of the convex vertex detection part in the modification. It is a figure which shows the processing example of the maximum value detection part. This is an example of a flowchart showing the vertex detection process in the modified example.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
FIG. 1 is a diagram showing a configuration example and a processing example of the image processing apparatus according to the first embodiment. The image processing device 10 shown in FIG. 1 includes a contour line extraction unit 11 and a region identification unit 12. The processing of the contour line extraction unit 11 and the area identification unit 12 is realized, for example, by executing a predetermined program by a processor (not shown) included in the image processing device 10.

A binarized image 1a showing a coronal cross section of the chest of the human body is input to the contour line extraction unit 11. The binarized image 1a is an image generated based on one or more cross-sectional images taken of the inside of the chest. For example, as such a cross-sectional image, a plurality of axial cross-sectional images (for example, CT slice images of the axial cross-section) in the chest region are taken. In this case, for example, a projected image is generated by projecting the minimum value of the pixel group in the front-rear direction in each of these axial cross-sectional images to the front, and a binarized image 1a is generated by binarizing the projected image. Will be done. The projected image in this case is not an image showing the state of the coronal cross section at a specific position of the human body, but an image showing the maximum region of the lungs and trachea in the coronal cross section.

Alternatively, a plurality of coronal cross-sectional images are generated based on a plurality of axial cross-sectional images taken, and the image selected from the generated coronal cross-sectional images is binarized to generate a binarized image 1a. May be done. In this case, for example, the image having the maximum width or area of the air region is selected from the coronal cross-sectional images and binarized. Further, the binarized image 1a may be generated by binarizing the image selected by the same procedure from the coronal cross-sectional images of the chest region taken in a plurality of images.

The contour line extraction unit 11 extracts the contour line 2 from the binarized image 1a. FIG. 1 shows an image 1b in which the contour line 2 in the binarized image 1a is displayed in an easy-to-understand manner. In the image 1b, the contour line 2 is shown by a thick line.

The region identification unit 12 identifies the position of the upper end of the lung field region based on the extracted contour line 2. In this identification, the region specifying unit 12 searches for the convex vertices on the contour line 2 by tracing the contour line 2 from the outside of both the left and right sides of the human body upward. The region identification unit 12 identifies the searched convex apex as the upper end of the lung field region.

In the image 1c shown in FIG. 1, of the contour lines 2 in the image 1b, only the contour lines located outside both the left and right sides are shown by thick lines. The region specifying unit 12 detects the convex apex 4a by following the contour line 2 as shown by an arrow 3a from the outer position on the left side of the contour line 2, for example. The region identification unit 12 identifies the position of the detected convex apex 4a as the upper end of the region of the right lung. Further, the region specifying unit 12 detects the convex apex 4b by tracing the contour line 2 from the position on the outer right side of the contour line 2 as shown by the arrow 3b. The region specifying portion 12 identifies the position of the detected convex apex 4b as the upper end of the region of the left lung.

Here, the upper ends of the right lung and the left lung each have a convex shape like a convex portion in the bell shape. In the upper end region of the lung field region, the trachea is arranged between the convex upper end of the right lung and the convex upper end of the left lung. Due to such an anatomical structure, when contour line 2 is traced upward from the left outside, it reaches the convex upper end of the right lung before reaching the area of the trachea. Similarly, when contour line 2 is traced upward from the outside on the right, it reaches the convex upper end of the left lung before reaching the area of the trachea.

Therefore, the region specifying portion 12 can accurately detect the upper end of the right lung by tracing the contour line 2 upward from the left outer side and searching for the convex apex 4a. Further, the region specifying portion 12 can accurately detect the upper end of the left lung by tracing the contour line 2 upward from the outside on the right and searching for the convex apex 4b. As described above, the image processing apparatus 10 according to the present embodiment can detect the upper end of the lung field region with high accuracy based on the medical image.

[Second Embodiment]
FIG. 2 is a diagram showing a configuration example of the image processing system according to the second embodiment. The image processing system shown in FIG. 2 includes a CT device 50 and an image processing device 100. The image processing device 100 is an example of the image processing device 10 shown in FIG.

The CT device 50 captures an X-ray CT image of the human body. In the present embodiment, the CT apparatus 50 captures a predetermined number of CT slice images of the axial cross section in the chest region while changing the position with respect to the height direction of the human body (the direction perpendicular to the axial cross section) at predetermined intervals.

The image processing device 100 extracts an image of the lung field region based on a plurality of CT slice images taken by the CT device 50. Thereby, the image processing apparatus 100 supports the doctor's image diagnosis work using the CT slice image, particularly the lung disease diagnosis work.

In this extraction process, the image processing apparatus 100 detects the upper end and the lower end of the lung field region based on the plurality of CT slice images, and based on the detection result, the lung field region is captured from the plurality of CT slice images. Narrow down the crowded CT slice images. Then, the image processing apparatus 100 identifies the lung field region by machine learning using the narrowed-down CT slice image, and extracts an image of the specified lung field region.

The image processing device 100 has, for example, the following hardware configuration. As shown in FIG. 2, the image processing apparatus 100 includes a processor 101, a RAM (Random Access Memory) 102, an HDD (Hard Disk Drive) 103, a graphic interface (I / F) 104, and an input interface (I / F) 105. It has a reading device 106 and a communication interface (I / F) 107.

The processor 101 comprehensively controls the entire image processing device 100. The processor 101 is, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), or a PLD (Programmable Logic Device). Further, the processor 101 may be a combination of two or more elements of the CPU, MPU, DSP, ASIC, and PLD.

The RAM 102 is used as the main storage device of the image processing device 100. At least a part of an OS (Operating System) program or an application program to be executed by the processor 101 is temporarily stored in the RAM 102. Further, the RAM 102 stores various data necessary for processing by the processor 101.

The HDD 103 is used as an auxiliary storage device for the image processing device 100. The OS program, application program, and various data are stored in the HDD 103. As the auxiliary storage device, other types of non-volatile storage devices such as SSD (Solid State Drive) can also be used.

A display device 104a is connected to the graphic interface 104. The graphic interface 104 causes the display device 104a to display an image according to an instruction from the processor 101. Display devices include liquid crystal displays and organic EL (Electroluminescence) displays.

An input device 105a is connected to the input interface 105. The input interface 105 transmits a signal output from the input device 105a to the processor 101. The input device 105a includes a keyboard, a pointing device, and the like. Pointing devices include mice, touch panels, tablets, touchpads, trackballs, and the like.

A portable recording medium 106a is attached to and detached from the reading device 106. The reading device 106 reads the data recorded on the portable recording medium 106a and transmits it to the processor 101. Examples of the portable recording medium 106a include an optical disk, a magneto-optical disk, and a semiconductor memory.

The communication interface 107 transmits / receives data to / from another device such as the CT device 50 via the network.
With the above hardware configuration, the processing function of the image processing apparatus 100 can be realized.

By the way, in image diagnosis using CT images, the number of images to be read is increasing due to the sophistication of the imaging apparatus, so that the burden on the doctor is increasing. In particular, in a group of diseases called diffuse lung disease, lesions are distributed over a wide area of the lung. For this reason, the number of images to be read by a doctor is extremely large, and abundant knowledge and experience are required for the diagnosis, which makes the diagnosis difficult. When diagnosing such cases, doctors narrow down the candidates for disease names by referring to similar cases in the past for which the diagnosis results have been confirmed. However, since it takes time and effort to search for similar cases in the past, a technique for searching for such similar cases is required.

In cases where lesions are widely distributed, such as diffuse lung disease, even if one CT slice image of the current patient and one CT slice image of the past patient are similar, When viewed three-dimensionally, the distribution of lesions is not always similar. Therefore, for accurate diagnosis, it is important to judge the similarity of the distribution of lesions throughout the lung. Therefore, in order to improve the search accuracy of similar cases, it is necessary to judge the similarity by using CT slice images of the entire lung region instead of a partial region.

Therefore, there is a method of identifying the lung field region by machine learning using all CT slice images (axial cross-sectional images) of the chest region and recognizing the abnormal shadow corresponding to the lesion from the identified lung field region. It is being developed. However, in this method, a CT slice image in which the lung field region is not shown may be mistakenly recognized as an image in which the lung field region is shown, and an abnormal shadow may be mistakenly recognized from the CT slice image. Therefore, if only the CT slice image obtained by capturing the region between the upper end and the lower end of the lung field region can be specified from the CT slice images and machine learning can be performed using the specified CT image, the lungs It is considered that the recognition accuracy of the field area and the recognition accuracy of abnormal shadows in the lung field area can be improved.

Against this background, the image processing device 100 of the present embodiment extracts the lung field region based on the captured CT slice image in the stage before performing machine learning, and the CT in which the lung field region is shown is captured. A slice image is specified, and machine learning is performed using only the specified CT slice image.

Here, FIG. 3 is a diagram showing the arrangement of organs in the chest.
The image processing device 100 generates a binarized image showing a coronal cross section of the chest based on a plurality of CT slice images (axial cross section images) obtained by photographing the chest region. Specifically, as will be described later, the image processing device 100 generates a front projection image by the minimum value projection method based on each CT slice image, and binarizes the front projection image to obtain a binarized image. Generate. Then, the image processing device 100 extracts the lung field region based on such a binarized image.

However, when such a binarized image is used, there is a problem that it is difficult to accurately identify the upper end of the lung field region. This is because, as shown in FIG. 3, the trachea 203 is adjacent to the upper region of the right lung 201 and the upper region of the left lung 202. In the CT slice image, the difference between the CT concentration value in the lung field region and the CT concentration value in the trachea 203 region is small. Therefore, in the binarized image, each region of the right lung 201, the left lung 202, and the trachea 203 has the same pixel value, and since these cannot be distinguished, the upper ends of the right lung 201 and the left lung 202 are each upper ends. Can no longer be identified accurately.

Here, as can be seen from FIG. 3, the vicinity of each upper end of the right lung 201 and the left lung 202 has a convex shape like a convex portion in the bell shape, and the convex apex is the right lung 201. , Each upper end of the left lung 202 is formed. Therefore, the image processing device 100 of the present embodiment extracts the contour line of the organ from the above-mentioned binarized image, traces the contour line in order from the left and right outside to the upward direction, and detects the convex apex. , The position of each upper end of the right lung 201 and the left lung 202 is accurately specified.

FIG. 4 is a block diagram showing a configuration example of a processing function included in the image processing apparatus. As shown in FIG. 4, the image processing device 100 includes an image input unit 111, a projected image generation unit 112, a region upper end extraction unit 113, a region lower end extraction unit 114, a slice selection unit 115, and a lung field region extraction unit 116. .. For the processing of the image input unit 111, the projected image generation unit 112, the region upper end extraction unit 113, the region lower end extraction unit 114, the slice selection unit 115, and the lung field region extraction unit 116, for example, the processor 101 executes a predetermined program. It is realized by.

The image input unit 111 receives a predetermined number of CT slice images captured by the CT device 50 and outputs them to the projection image generation unit 112 and the region lower end extraction unit 114. As mentioned above, these CT slice images are axial cross-sectional images in the chest. Further, since these CT slice images are images taken at regular intervals in the height direction of the human body, each CT slice image is added to the coordinates (X coordinate, Y coordinate) on the image. It also has coordinates in the height direction (Z coordinates).

The projection image generation unit 112 generates a front projection image in which the internal state of the chest is projected to the front by the minimum value projection method based on a predetermined number of CT slice images from the image input unit 111.
The region upper end extraction unit 113 extracts the upper end of the lung field region based on the generated front projection image. The region upper end extraction unit 113 includes a contour line extraction unit 121 and a convex vertex detection unit 122. The contour line extraction unit 121 binarizes the front projection image to generate a binarized image, and extracts the contour line from the binarized image. The convex apex detection unit 122 detects the upper end positions of the right lung and the left lung by detecting the convex apex from the contour line.

The region lower end extraction unit 114 extracts the lower end of the lung field region (lower end of the right lung, lower end of the left lung) based on a predetermined number of CT slice images from the image input unit 111.
The slice selection unit 115 includes the upper end and the lower end from the CT slice images received by the image input unit 111 based on the detection results of the upper end and the lower end of the lung field region by the region upper end extraction unit 113 and the region lower end extraction unit 114. Select a CT slice image whose imaging target is the area between.

The lung field region extraction unit 116 extracts an image region of the lung field region from the CT slice image selected by the slice selection unit 115 based on the learning result by machine learning, and outputs the extraction result. Further, the lung field region extraction unit 116 may recognize an abnormal shadow from the extracted image region based on the learning result by machine learning. Further, the lung field region extraction unit 116 outputs the CT slice image selected by the slice selection unit 115 as training data for machine learning, and adds this learning data to the existing learning data to obtain the lung field region. Machine learning may be performed for extraction and recognition of abnormal shadows.

Next, the upper end extraction process of the lung field region will be described with reference to FIGS. 5 to 9.
First, FIG. 5 is a diagram for explaining a process of generating a front projection image. The CT slice image group 211 shown in FIG. 5 includes a predetermined number of CT slice images output from the image input unit 111 to the projection image generation unit 112. In FIG. 5, the left-right direction of the human body is the X-axis, the depth direction is the Y-axis, and the height direction is the Z-axis. Each CT slice image shows an image of the inside of the human body in the XY plane (ie, axial cross section).

The projection image generation unit 112 generates a front projection image by using the following minimum value projection method. The projection image generation unit 112 executes the following processing for each CT slice image. The projection image generation unit 112 selects a pixel array along the Y-axis from the CT slice image, and projects the minimum value of the CT density value in the pixel array onto the X-axis and the projection plane (X-Z plane) along the Z-axis. To do. The projection image generation unit 112 projects the minimum value of the CT density value for each pixel sequence of the CT slice image. As a result, a pixel sequence along the X axis is formed from one CT slice image on the projection plane. Then, by executing such a projection process for each CT slice image, a pixel sequence corresponding to the number of CT slice images is formed on the projection plane.

The projection image generation unit 112 generates a front projection image 212 by performing an interpolation calculation in the Z-axis direction based on, for example, a pixel sequence formed on the projection plane along the X-axis. By such a minimum value projection method, the largest region of the organ regions (at least the lung and tracheal regions) on the coronal image generated from the CT slice image group 211 is projected on the front projection image 212. ..

In addition to the CT slice image group 211 described above, when a plurality of coronal cross-sectional images of the chest are taken by the CT device 50, the projected images may be generated using these coronal cross-sectional images. In this case, the front projection image is generated by projecting the minimum value of the CT density value of the same pixel in each coronal cross-sectional image onto the projection plane.

Next, the processing of the region upper end extraction unit 113 will be described.
First, FIG. 6 is a diagram showing an example of the internal configuration of the convex vertex detection unit. As shown in FIG. 6, the convex vertex detection unit 122 includes a contour line dividing unit 131, a turning point detection unit 132, and a vertex detection unit 133.

The contour line dividing unit 131 divides the binarized image obtained by binarizing the front projection image into blocks of the same size. As a result, the contour line extracted from the binarized image is divided into blocks.

The turning point detection unit 132 calculates the slope of a straight line connecting both ends of the contour line for the block including the contour line. The turning point detection unit 132 traces the contour line upward from the left end, and detects a turning point in which the positive and negative of the slope of the straight line are reversed between adjacent blocks including the contour line. Further, the turning point detection unit 132 traces the contour line upward from the right end, and detects a turning point in which the positive and negative of the slope of the straight line are reversed between the adjacent blocks including the contour line.

The vertex detection unit 133 subdivides the blocks on both sides of the detected turning point into smaller blocks, and calculates the slope of the straight line connecting both ends of the contour line for the block including the contour line among the subdivided blocks. To do. The apex detection unit 133 detects the convex apex of the right lung and the convex apex of the left lung based on the position where the positive and negative of the slope of the straight line is reversed between the blocks, and determines the position of each detected vertex as the right lung. Identify as the top of each region of the left lung.

FIG. 7 is a first diagram showing a processing example of the region upper end extraction unit.
The contour line extraction unit 121 of the region upper end extraction unit 113 generates the binarized image 221 by binarizing the front projection image generated by the projection image generation unit 112. For example, assuming that the minimum value of the CT density value is -1000 HU and the maximum value is 1000 HU, the contour line extraction unit 121 sets 0 (black) for pixels with a CT density value of less than 500 HU and 1 (black) for pixels with a CT density value of 500 HU or more in the front projection image. White) to generate a binarized image 221.

Next, the contour line extraction unit 121 extracts the contour line 222a from the generated binarized image 221. The contour line 222a is a boundary line of regions having different pixel values. However, in the binarized image 221 it is presumed that the lung field region exists in the central region. Therefore, for example, the contour line extraction unit 121 extracts only the contour line in the central region and cancels the extraction of another contour line existing outside the contour line. In FIG. 7, the image 222 shows a state in which the contour line 222a shown by the thick line is extracted from the binarized image 221.

Next, the contour line dividing unit 131 of the convex vertex detecting unit 122 divides the image 222 from which the contour line 222a is extracted into blocks of the same size. Here, the contour line dividing unit 131 divides the image 222 by dividing the image 222 by dividing lines arranged at equal intervals in the horizontal direction and the vertical direction. In FIG. 7, image 223 shows a state in which image 222 is divided into blocks.

It is desirable that the height and width of the divided blocks are smaller than the width of the trachea. This condition increases the possibility of accurately detecting the convex apex of the right and left lungs.
Next, the turning point detection unit 132 of the convex vertex detection unit 122 calculates the slope of the straight line connecting both ends of the contour line for the block including the contour line. The turning point detection unit 132 detects a turning point where the positive and negative of the slope are reversed based on the slope of the straight line. The apex detection unit 133 of the convex apex detection unit 122 extracts the upper end of the lung field region by verifying the blocks on both sides of the detected turning point. Hereinafter, processing examples of the turning point detection unit 132 and the vertex detection unit 133 will be described with reference to FIGS. 8 and 9.

FIG. 8 is a second diagram showing a processing example of the region upper end extraction unit. FIG. 8 shows an example of processing for detecting the convex apex of the right lung.
The turning point detection unit 132 traces the contour line on the image 223 upward from the left end, and detects a turning point in which the positive and negative of the slope of the straight line are reversed between the adjacent blocks including the contour line. In the example of FIG. 8, the block B1 includes the left end of the contour line. The turning point detection unit 132 first calculates the slope of the straight line connecting both ends of the contour line in the block B1. Next, the turning point detection unit 132 traces the contour line upward and calculates the slope of the straight line connecting both ends of the contour line for the block B2 adjacent to the block B1. Then, the turning point detection unit 132 determines whether the positive or negative of the slope of the straight line is reversed between the block B1 and the block B2. In the example of FIG. 8, since the slope of the straight line remains positive, the turning point detection unit 132 then traces the contour line upward, and for the block B3 adjacent to the block B2, the slope of the straight line connecting both ends of the contour line. Is calculated. Then, the turning point detection unit 132 determines whether the positive or negative of the slope of the straight line is reversed between the block B2 and the block B3.

The turning point detection unit 132 repeats such a determination in order while following the contour line. Then, in the example of FIG. 8, it is assumed that the positive and negative of the slope of the straight line are reversed between the block B11 and the block B12. That is, it is assumed that the slope of the straight line L1 of the block B11 is positive, but the slope of the straight line L2 of the block B12 is negative. In this case, the point P1 through which the contour line passes at the boundary between the block B11 and the block B12 is detected as a turning point. This turning point is presumed to be located near the convex apex located at the upper end of the right lung. Therefore, from the blocks in the image 223, the blocks B11 and B12 on both sides sandwiching the detected turning point are narrowed down as candidates for the block in which the upper end of the right lung exists.

Next, the vertex detection unit 133 subdivides the blocks B11 and B12 into smaller blocks. Then, the vertex detection unit 133 calculates the slope of the straight line connecting both ends of the contour line for the block including the contour line among the subdivided blocks. The vertex detection unit 133 detects a position where the positive and negative of the slope of the straight line are reversed between adjacent blocks. In the example of FIG. 8, it is assumed that the positive and negative of the slope of the straight line are reversed between the block B12a and the block B12b. That is, it is assumed that the slope of the straight line L2a of the block B12a is positive, but the slope of the straight line L2b of the block B12b is negative. In this case, the point P2 through which the contour line passes at the boundary between the block B12a and the block B12b is detected as a turning point. When a plurality of turning points are detected in the block B12, the vertex detection unit 133 selects the turning point having the maximum coordinate in the height direction and continues the process.

The vertex detection unit 133 subdivides the blocks B12a and B12b on both sides of the detected turning point into smaller blocks, and detects the turning point where the positive and negative of the slope of the straight line are reversed by the same procedure as described above. The vertex detection unit 133 repeats such a process until the turning point is no longer detected. Then, the apex detection unit 133 detects the last detected turning point as a convex apex located at the upper end of the right lung. The height-wise coordinates (Z-coordinates) of the detected convex vertices indicate the coordinates of the upper end of the region of the right lung.

By such a process, the upper end of the right lung can be extracted with high accuracy. That is, as shown in FIG. 3, the upper end of the right lung has a convex shape like a convex portion in a bell shape, and the trachea exists on the right side of the position of this convex shape. Due to such a structure, the contour line is traced upward from the left end as shown in FIG. 8 to reach the convex upper end of the right lung before reaching the area of the trachea. Therefore, the position of the upper end of the right lung can be extracted accurately.

FIG. 9 is a third diagram showing a processing example of the region upper end extraction unit. FIG. 9 shows an example of processing for detecting the convex apex of the left lung.
The turning point detection unit 132 and the apex detection unit 133 also execute the process described with reference to FIG. 8 in the left-right reversal of the left lung. That is, the turning point detection unit 132 traces the contour line on the image 223 upward from the right end, and detects a turning point in which the positive and negative of the slope of the straight line are reversed between adjacent blocks including the contour line. In the example of FIG. 9, the block B21 includes the right end of the contour line. The turning point detection unit 132 first calculates the slope of the straight line connecting both ends of the contour line in the block B21, and then calculates the slope of the straight line connecting both ends of the contour line in the block B22. Then, the turning point detection unit 132 determines whether the positive or negative of the slope of the straight line is reversed between the block B21 and the block B22. In the example of FIG. 9, the slope of the straight line remains positive, so the turning point detection unit 132 then calculates the slope of the straight line connecting both ends of the contour line in the block B23, and between the block B22 and the block B23. It is determined whether the positive or negative of the slope of the straight line is reversed.

In this way, in FIG. 9, it is assumed that the positive and negative of the slope of the straight line are reversed between the block B31 and the block B32. That is, it is assumed that the slope of the straight line L11 of the block B31 is negative, but the slope of the straight line L12 of the block B32 is positive. In this case, the point P11 through which the contour line passes at the boundary between the block B31 and the block B32 is detected as a turning point.

Next, the vertex detection unit 133 further subdivides the blocks B31 and B32 on both sides of the turning point (point P11) into blocks, and detects a turning point in which the positive and negative of the slope of the straight line are reversed by the same procedure as in FIG. To do. In the example of FIG. 9, it is assumed that the positive and negative of the slope of the straight line are reversed between the block B32a and the block B32b. That is, it is assumed that the slope of the straight line L12a of the block B32a is negative, but the slope of the straight line L12b of the block B32b is positive. In this case, the point P12 through which the contour line passes at the boundary between the block B32a and the block B32b is detected as a turning point.

The vertex detection unit 133 subdivides the blocks B32a and B32b on both sides of the detected turning point into smaller blocks, and detects the turning point where the positive and negative of the slope of the straight line are reversed by the same procedure as described above. The vertex detection unit 133 repeats such a process until the turning point is no longer detected. Then, the apex detection unit 133 detects the last detected turning point as a convex apex located at the upper end of the left lung. The height coordinate (Z coordinate) of the detected convex apex indicates the coordinate of the upper end of the region of the left lung.

By such a process, the upper end of the left lung can be extracted with high accuracy. That is, as shown in FIG. 3, the upper end of the left lung has a convex shape like a convex portion in the bell shape, and the trachea exists on the left side from the position of this convex shape. Due to such a structure, by tracing the contour line upward from the right end as shown in FIG. 9, the convex upper end of the left lung is reached before reaching the area of the trachea. Therefore, the position of the upper end of the left lung can be extracted accurately.

Next, the processing of the image processing apparatus 100 will be described with reference to a flowchart.
FIG. 10 is an example of a flowchart showing the entire processing by the image processing apparatus.
[Step S11] The image input unit 111 receives a predetermined number of CT slice images captured by the CT device 50 and outputs them to the projection image generation unit 112 and the region lower end extraction unit 114.

[Step S12] Using the CT slice image output from the image input unit 111, the lung field region upper end extraction process is executed by the projection image generation unit 112 and the region upper end extraction unit 113. As a result, the upper end of the lung field region (the upper end of the right lung and the upper end of the left lung) is extracted.

[Step S13] The region lower end extraction unit 114 extracts the lower end of the lung field region (lower end of the right lung, lower end of the left lung) based on a predetermined number of CT slice images output from the image input unit 111.
[Step S14] The slice selection unit 115 excludes an image obtained by capturing a region above the upper end extracted in step S12 from the CT slice images received by the image input unit 111. Further, the slice selection unit 115 excludes an image obtained by capturing a region below the lower end extracted in step S13 from the remaining CT slice images after exclusion. As a result, the CT slice image obtained by capturing the lung field region is selected as the CT slice image to be extracted from the lung field region from the CT slice images received by the image input unit 111.

[Step S15] The lung field region extraction unit 116 extracts an image region of the lung field region from the CT slice image selected in step S14 based on the learning result by machine learning, and outputs the extraction result. By using only the CT slice image selected in step S14 as the extraction target, the lung field region can be extracted with high accuracy.

Further, the lung field region extraction unit 116 may recognize an abnormal shadow from the extracted image region based on the learning result by machine learning. In this process as well, by using only the CT slice image selected in step S14, erroneous detection of abnormal shadows can be reduced, and the detection accuracy can be improved.

FIG. 11 is an example of a flowchart showing the lung field region upper end extraction process. The process of FIG. 11 corresponds to the process of step S12 of FIG.
[Step S21] The projection image generation unit 112 generates a front projection image by the minimum value projection method based on a predetermined number of CT slice images output from the image input unit 111.

[Step S22] The contour line extraction unit 121 binarizes the generated front projection image to generate a binarized image.
[Step S23] The contour line extraction unit 121 extracts a contour line from the generated binarized image.

[Step S24] The contour line dividing unit 131 divides the contour line by dividing the image from which the contour line is extracted into blocks of the same size.
[Step S25] The left apex detection process for detecting the convex apex of the left lung (right lung) in the image is executed based on the divided contour line.

[Step S26] The right apex detection process for detecting the convex apex of the right lung (left lung) in the image is executed based on the divided contour line.
[Step S27] The vertex detection unit 133 outputs the upper end position (height) of the lung field region from the detection results of steps S25 and S26. That is, the height of the convex apex of the right lung (Z coordinate) is output as the upper end position of the right lung, and the height of the convex apex of the left lung (Z coordinate) is output as the upper end position of the left lung.

FIG. 12 is an example of a flowchart showing the vertex detection process. The process of FIG. 12 corresponds to the process of steps S25 and S26 of FIG. The processes of steps S25 and S26 are basically the same except that the left-right relationship is reversed. Therefore, here, the left apex detection process in step S25 will be mainly described, and the right apex detection process in step S26 will be supplemented as appropriate. Further, in the following description, the "division curve" refers to a contour line divided by blocks.

[Step S31] The turning point detection unit 132 identifies the leftmost dividing curve on the image. In this process, a block including the left end of the contour line is selected, and the contour line (division curve) included in the block is specified. The turning point detection unit 132 calculates the coordinates of both ends of the dividing curve. In the right vertex detection process, the rightmost dividing curve on the image is specified.

[Step S32] The turning point detection unit 132 calculates the slope of the straight line connecting both ends of the dividing curve based on the coordinates calculated in step S31.
[Step S33] The turning point detection unit 132 identifies the next dividing curve by tracing the contour line upward. If the contour line cannot be traced upward, the next dividing curve is specified by tracing the contour line to the right. The turning point detection unit 132 calculates the coordinates of both ends of the dividing curve. In the right vertex detection process, the next dividing curve is specified by tracing the contour line upward, and if it cannot be traced upward, the next dividing curve is obtained by tracing the contour line to the left. Be identified.

[Step S34] The turning point detection unit 132 calculates the slope of the straight line connecting both ends of the dividing curve based on the coordinates calculated in step S33.
[Step S35] The turning point detection unit 132 has the inclination of the straight line (first inclination) calculated in step S34 and the inclination of the straight line calculated in step S32 or calculated by the previous execution of step S34 (step S35). It is determined whether a positive / negative turning point of the slope is detected by comparing with the second slope). When both the angles of the first inclination and the second inclination are 0 or more, or when both angles are less than 0, the positive and negative of the inclination are not reversed and the turning point is not detected. On the other hand, when one of the first inclination and the second inclination is 0 or more and the other angle is less than 0, it is determined that the positive and negative of the inclination are reversed. In this case, the point of contact between the two straight lines (that is, the passing point of the contour line at the boundary between the two blocks including each straight line) is detected as a turning point.

If the turning point is not detected, the turning point detection unit 132 executes the process of step S33 again, and if the turning point is detected, executes the process of step S36.
[Step S36] At the time of the first execution of step S36, the vertex detection unit 133 records the coordinates of the turning point detected in step S35 in the RAM 102. At the time of the second and subsequent executions of step S36, the apex detection unit 133 records the coordinates of the turning point detected by the previous execution of step S40 in the RAM 102. In the latter case, the coordinates of the turning point recorded by the previous execution of step S36 are discarded.

[Step S37] The vertex detection unit 133 subdivides the blocks on both sides of the turning point on which the coordinates are recorded into blocks having a smaller size. As a result, the contour line in each block is subdivided into a shorter dividing curve.

[Step S38] The vertex detection unit 133 calculates the coordinates of both ends of each of the subdivided curves.
[Step S39] The vertex detection unit 133 calculates the slope of the straight line connecting both ends of each of the subdivided division curves based on the coordinates calculated in step S38.

[Step S40] The vertex detection unit 133 determines whether a positive / negative turning point of the slope has been detected based on the slope of each straight line calculated in step S39. Specifically, the vertex detection unit 133 compares the slope of each straight line for each combination of adjacent straight lines. Similar to the determination in step S35, when the angles of both inclinations are 0 or more or less than 0, the contact point of each straight line is not a turning point. On the other hand, when the angle of inclination of one is 0 or more and the angle of inclination of the other is less than 0, the contact point of each straight line is detected as a turning point.

If the turning point is not detected, the turning point detection unit 132 executes the process of step S41. On the other hand, when one or more turning points are detected, the turning point detection unit 132 re-executes the process of step S36. In the latter case, when a plurality of turning points are detected, the turning point having the maximum height (Z coordinate) is selected and the coordinates are recorded in step S36.

[Step S41] The vertex detection unit 133 reads out the coordinates of the turning point recorded in step S36, and outputs these coordinates as the coordinates of the convex vertex on the left side, that is, the coordinates of the upper end of the right lung. In the right apex detection process, the read coordinates are output as the coordinates of the right convex apex, that is, the coordinates of the upper end of the left lung.

According to the image processing apparatus 100 of the second embodiment described above, the upper end of the lung field region can be accurately extracted from the front projection image based on the CT slice image of the chest. Therefore, an image showing the lung field region can be accurately selected from the CT slice images, and the lung field region extraction process based on the learning result of machine learning can be accurately executed using the selected image. In addition, using the selected image, it is possible to accurately execute the recognition process of abnormal shadows based on the learning result of machine learning.

<Modified example of the second embodiment>
Next, a modified example in which a part of the processing in the second embodiment is changed will be described. In the description of the modified example, the same reference numerals are used for the same components and the same processing steps as in the second embodiment.

FIG. 13 is a diagram showing an example of the internal configuration of the convex vertex detection unit in the modified example. In the modified example, the image processing apparatus 100 includes the convex vertex detection unit 122a shown in FIG. 13 instead of the convex vertex detection unit 122 shown in FIG.

The convex apex detection unit 122a includes a contour line dividing unit 131 and a turning point detecting unit 132 shown in FIG. Further, the convex vertex detection unit 122a includes a maximum value detection unit 134 instead of the vertex detection unit 133 shown in FIG. The maximum value detection unit 134 detects the convex vertices by detecting the point at which the height (Z coordinate) on the contour line is maximum from the blocks on both sides of the conversion point detected by the conversion point detection unit 132. To do.

FIG. 14 is a diagram showing a processing example of the maximum value detection unit. FIG. 14 shows an image 223 similar to that of FIGS. 8 and 9. Further, the point P1 is detected as a turning point from the contour line on the left side by the process described in FIG. 8 by the turning point detecting unit 132. Further, the point P11 is detected as a turning point from the contour line on the right side by the process described with reference to FIG. 9 by the turning point detecting unit 132.

The maximum value detection unit 134 identifies the blocks B11 and B12 on both sides of the turning point (point P1). The maximum value detection unit 134 detects the maximum value of the height (Z coordinate) on the dividing curve (contour line) inside the blocks B11 and B12, and positions the pixel P21 where the maximum value is detected as a convex shape of the right lung. Detect as the apex, the upper end of the right lung.

Similarly, the maximum value detection unit 134 identifies the blocks B31 and B32 on both sides of the turning point (point P11). The maximum value detection unit 134 detects the maximum value of the height (Z coordinate) on the internal division curve (contour line) of the blocks B31 and B32, and positions the pixel P22 where the maximum value is detected in the convex shape of the left lung. Detect as the apex, the upper end of the left lung.

As described above, in the modified example, the blocks B11 and B12 in which the upper end of the right lung exists is narrowed down by tracing the contour line upward from the left end and the right end, respectively, as in the second embodiment, and the left lung. Blocks B31 and B32 having the upper end of the block B31 and B32 are narrowed down. After that, unlike the second embodiment, the convex apex of the right lung is detected based on the coordinates of the contour line in blocks B11 and B12, and the convexity of the left lung is detected based on the coordinates of the contour line in blocks B31 and B32. Shaped vertices are detected. As a result, the upper ends of each of the right lung and the left lung can be detected with high accuracy.

FIG. 15 is an example of a flowchart showing the vertex detection process in the modified example. The process of FIG. 15 corresponds to the process of steps S25 and S26 of FIG. Further, the processing steps having the same contents as those in FIG. 12 are shown with the same reference numerals, and the description thereof will be omitted.

As shown in FIG. 15, in the modified example, after steps S31 to S35 of FIG. 12 are executed, the processes of steps S51 to S53 are executed. Here, as in FIG. 12, the left-sided apex detection process in step S25 will be mainly described, and the right-sided apex detection process in step S26 will be supplemented as appropriate.

[Step S51] The maximum value detection unit 134 identifies the blocks on both sides of the turning point detected in step S35.
[Step S52] The maximum value detecting unit 134 detects the maximum value of the height (Z coordinate) from the dividing curve (contour line) included in the two specified blocks.

[Step S53] The maximum value detection unit 134 outputs the position of the pixel in which the maximum value is detected as the coordinates of the convex apex on the left side, that is, the coordinates of the upper end of the right lung. In the right apex detection process, the position of the pixel in which the maximum value is detected is output as the coordinates of the right convex apex, that is, the coordinates of the upper end of the left lung.

In addition, at least a part of the processing functions provided in the image processing apparatus 100 in the above second embodiment and its modification may be mounted on the CT apparatus 50. Further, the processing functions provided in the image processing device 100 may be distributed and mounted in a plurality of computer devices.

Further, the processing functions of the devices (for example, the image processing devices 10, 100) shown in the above embodiments can be realized by a computer. In that case, a program describing the processing content of the function that each device should have is provided, and the above processing function is realized on the computer by executing the program on the computer. The program describing the processing content can be recorded on a computer-readable recording medium. Computer-readable recording media include magnetic storage devices, optical disks, opto-magnetic recording media, semiconductor memories, and the like. Magnetic storage devices include hard disk devices (HDDs), magnetic tapes, and the like. Optical discs include CDs (Compact Discs), DVDs (Digital Versatile Discs), and Blu-ray Discs (Blu-ray Discs: BDs, registered trademarks). The magneto-optical recording medium includes MO (Magneto-Optical disk) and the like.

When a program is distributed, for example, a portable recording medium such as a DVD or a CD on which the program is recorded is sold. It is also possible to store the program in the storage device of the server computer and transfer the program from the server computer to another computer via the network.

The computer that executes the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its own storage device and executes the processing according to the program. The computer can also read the program directly from the portable recording medium and execute the processing according to the program. In addition, the computer can sequentially execute processing according to the received program each time the program is transferred from the server computer connected via the network.

1a Binarized image 1b, 1c image 2 Contour line 3a, 3b Arrow 4a, 4b Convex apex 10 Image processing device 11 Contour line extraction unit 12 Area identification unit

Claims (8)

On the computer
A contour line is extracted from a binarized image showing the coronal cross-section of the chest, which is generated based on one or more cross-sectional images of the inside of the chest of the human body.
By tracing the contour line upward from both the left and right sides of the human body, the convex apex in the contour line is searched, and the position of the searched convex apex is specified as the upper end of the lung field region.
An image processing program that executes processing. On the computer
A plurality of axial cross-sectional images showing the axial cross-section of the chest are used as the one or more cross-sectional images, and a projected image is generated by projecting the minimum value of the pixel group in the front-rear direction in each of the plurality of axial cross-sectional images to the front. Then, the binarized image is generated by binarizing the projected image.
The image processing program according to claim 1, wherein the processing is further executed. On the computer
Based on the identification result of the upper end of the lung field region, an image including the lung field region in the imaging range is specified from the plurality of axial cross-sectional images.
The image processing program according to claim 2, wherein the processing is further executed. In the search for the convex vertices, the binarized image is divided into a plurality of divided regions, and both ends of the contour line are connected for each of the first divided regions including the contour line among the plurality of divided regions. When the inclination of the first straight line is calculated and the first division region is sequentially traced upward from the outside of both the left and right sides of the contour line, the first division region is adjacent to the first division region. The convex apex is specified based on the position where the positive or negative of the slope of the straight line changes.
The image processing program according to any one of claims 1 to 3. In the search for the convex vertices, the two adjacent second division regions in which the positive and negative slopes of the first straight line have changed in the first division region are further divided into a plurality of third division regions. , The slope of the second straight line connecting both ends of the contour line is calculated for each of the fourth divided regions including the contour line in the third divided region, and the second straight line is calculated in the adjacent fourth divided region. The convex apex is specified based on the position where the positive / negative of the slope of the straight line of 2 changes.
The image processing program according to claim 4. In the search for the convex vertices, the said is based on the upper end position of the contour line in two adjacent second divided regions in which the positive and negative of the slope of the first straight line of the first divided region are changed. Identify convex vertices,
The image processing program according to claim 4. A contour line extraction unit that extracts a contour line from a binarized image showing a coronal cross section of the chest, which is generated based on one or more cross-sectional images of the inside of the chest of the human body.
A region specifying portion that searches for a convex apex in the contour line by tracing the contour line upward from both the left and right sides of the human body, and specifies the position of the searched convex apex as the upper end of the lung field region. When,
An image processing device having. The computer
A contour line is extracted from a binarized image showing the coronal cross-section of the chest, which is generated based on one or more cross-sectional images of the inside of the chest of the human body.
By tracing the contour line upward from both the left and right sides of the human body, the convex apex in the contour line is searched, and the position of the searched convex apex is specified as the upper end of the lung field region.
Image processing method.

Images