Image creation device, image creation method, and method for creating learned model

文档序号:572761 发布日期:2021-05-18 浏览:24次 中文

阅读说明:本技术 图像制作装置、图像制作方法以及学习完毕模型的制作方法 (Image creation device, image creation method, and method for creating learned model ) 是由 押川翔太 高桥涉 于 2019-09-27 设计创作,主要内容包括:本图像制作装置具备:学习图像生成部(1),其基于三维数据来生成教师输入图像和教师输出图像;噪声施加部(2),其对教师输入图像和教师输出图像施加相同的噪声;学习部(3),其基于被施加了所述噪声的教师输入图像和被施加了所述噪声的教师输出图像来进行机器学习,由此对用于提取或除去特定部分的学习模型进行学习;以及图像制作部(4),其使用学习完毕的学习模型来制作提取出或除去了特定部分的图像。(The image forming apparatus includes: a learning image generation unit (1) that generates a teacher input image and a teacher output image on the basis of the three-dimensional data; a noise application unit (2) that applies the same noise to the teacher input image and the teacher output image; a learning unit (3) that performs machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for extracting or removing a specific portion; and an image creation unit (4) that creates an image from which the specific portion has been extracted or removed, using the learned model.)

1. An image forming apparatus includes:

a learning image generation unit that generates, based on three-dimensional data having three-dimensional pixel value data, a teacher input image that is a two-dimensional image of a region including a specific portion of a subject, and a teacher output image that is an image showing the specific portion or an image excluding the specific portion at the same angle of view as the teacher input image;

a noise application unit that applies the same noise to the teacher input image and the teacher output image;

a learning section that performs machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for extracting or removing the specific portion; and

and an image creating unit that creates an image from which the specific portion is extracted or removed, using the learned learning model, based on the captured image including the specific portion.

2. The image forming apparatus according to claim 1,

the learning image generation unit is configured to: the three-dimensional data is acquired based on a computed tomography image obtained by imaging an object, and the teacher input image and the teacher output image are generated by digital reconstruction simulation.

3. The image forming apparatus according to claim 1,

the noise applying unit is configured to: noise adjusted based on noise of a result of image capturing obtained by image capturing is applied to the teacher input image and the teacher output image.

4. The image forming apparatus according to claim 3,

the noise applying unit is configured to apply gaussian noise, and is configured to apply gaussian noise adjusted to a standard deviation corresponding to a standard deviation of noise of a photographing result obtained by photographing an image to the teacher input image and the teacher output image.

5. The image forming apparatus according to claim 1,

the specific part is a bone, and the specific part is a bone,

the learning image generation unit is configured to: generating the teacher input image including a two-dimensional region of a bone of the subject and the teacher output image excluding the bone of the subject,

the learning unit is configured to: performing machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning the learning model for removing bone,

the image creating unit is configured to: and creating an image from which the bone is removed, using the learned model, based on the captured image including the bone.

6. The image forming apparatus according to claim 1,

the specific portion is a blood vessel having a blood vessel,

the learning image generation unit is configured to: generating the teacher input image including a two-dimensional region of a blood vessel of the subject and the teacher output image showing the blood vessel of the subject,

the learning unit is configured to: performing machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning the learning model for extracting a blood vessel,

the image creating unit is configured to: and extracting the blood vessel and making an image by using the learned learning model according to the shot image comprising the blood vessel.

7. A method for producing an image, wherein,

generating a two-dimensional teacher input image including a region of a specific portion of the subject based on three-dimensional data having three-dimensional pixel value data,

generating a teacher output image that is an image showing the specific portion or an image excluding the specific portion in the same angle of view as the teacher input image,

applying the same noise to the teacher input image and the teacher output image,

performing machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for extracting or removing the specific portion,

and creating an image from which the specific portion is extracted or removed, using the learned learning model, based on the captured image including the specific portion.

8. A method for making a model after learning, wherein,

generating a two-dimensional teacher input image including a region of a specific portion of the subject based on three-dimensional data having three-dimensional pixel value data,

generating a teacher output image that is an image showing the specific portion or an image excluding the specific portion in the same angle of view as the teacher input image,

applying the same noise to the teacher input image and the teacher output image,

performing machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for extracting or removing the specific portion.

Technical Field

The present invention relates to an image creating device, an image creating method, and a method of creating a learned model.

Background

Conventionally, an image forming apparatus is known. Such a device is disclosed, for example, in international publication No. 2007/029467.

The above-mentioned international publication No. 2007/029467 discloses an image processing apparatus (image creating apparatus) that performs machine learning based on a teacher input image and a teacher output image, and outputs an enhanced image from a processing target image using a learning result. In the image processing apparatus, in order to learn a large number of patterns, a plurality of image processes are performed on a teacher input image to generate a plurality of teacher input images. For example, the processing is performed so as to deteriorate the image quality of the teacher input image. In this case, noise is included in the teacher input image, and a learning result with reduced noise is obtained by machine learning.

Documents of the prior art

Patent document

Patent document 1: international publication No. 2007/029467

Disclosure of Invention

Problems to be solved by the invention

However, in the image processing apparatus of international publication No. 2007/029467, since machine learning is performed to reduce noise of a teacher input image, it is possible to learn a filter that extracts a low frequency region of an image such as smoothing noise. Therefore, when an image is created using the learning result, there is a problem that the image is smoothed, and the output image is blurred.

The present invention has been made to solve the above-described problems, and an object of the present invention is to provide an image creating device, an image creating method, and a learned model creating method that can suppress blurring when creating an image using a learned model learned by machine learning.

Means for solving the problems

In order to achieve the above object, an image forming apparatus according to a first aspect of the present invention includes: a learning image generation unit that generates a teacher input image that is a two-dimensional image of a region including a specific portion of a subject and a teacher output image that is an image showing the specific portion or an image excluding the specific portion at the same angle of view as the teacher input image, based on three-dimensional data having three-dimensional pixel value data; a noise application unit that applies the same noise to the teacher input image and the teacher output image; a learning section that performs machine learning based on a teacher input image to which the noise is applied and a teacher output image to which the noise is applied, thereby learning a learning model for extracting or removing a specific portion; and an image creating unit that creates an image from which the specific portion has been extracted or removed, using the learned learning model, based on the captured image including the specific portion.

In the image forming apparatus according to the first aspect of the present invention, since the same noise is applied to both the teacher input image and the teacher output image, it is possible to learn image processing such as extracting or removing a specific portion of the teacher input image and a specific portion of the teacher output image, the specific portion being a difference other than a noise component. This prevents the filter for smoothing noise from being learned, and thus can suppress the blur of the created image.

In the image forming apparatus according to the first aspect, it is preferable that the learning image generating unit is configured to: three-dimensional data is acquired based on a computed tomography image obtained by imaging an object, and a teacher input image and a teacher output image are generated by digital reconstruction simulation. With such a configuration, a plurality of sets of teacher input images and teacher output images can be easily generated by digital reconstruction simulation from the computed tomography image, and therefore machine learning can be performed using a plurality of sets of image groups. This can improve the learning accuracy of the learning model.

In the image forming apparatus according to the first aspect, it is preferable that the noise applying unit is configured to: the teacher input image and the teacher output image are subjected to noise adjusted based on noise of a photographing result obtained by photographing the images. With such a configuration, since the teacher input image and the teacher output image can be subjected to machine learning by applying noise corresponding to noise in the case of actually capturing images, it is possible to effectively suppress the influence of noise and extract or remove a specific portion in the case of creating an image from a captured image including the specific portion using a learning model having been learned.

In this case, it is preferable that the noise applying unit is configured to apply gaussian noise, and to the teacher input image and the teacher output image, gaussian noise adjusted to a standard deviation corresponding to a standard deviation of noise of a result of photographing obtained by photographing the image is applied. With such a configuration, the standard deviation of the applied gaussian noise can be matched with the noise of the image actually captured, and therefore, the noise can be more effectively suppressed from affecting the image creation, and the specific portion can be extracted or removed.

In the image creating device according to the first aspect, it is preferable that the specific portion is a bone, and the learning image generating unit is configured to: the teacher input image generation unit generates a two-dimensional teacher input image including a region of a bone of a subject and a teacher output image excluding the bone of the subject, and the learning unit is configured to: learning a learning model for removing bones by performing machine learning based on a teacher input image to which noise is applied and a teacher output image to which noise is applied, the image creating unit being configured to: from the captured image including the bone, an image from which the bone is removed is created using the learned learning model. With such a configuration, since a learning model in which only a bone is removed without removing a noise component as much as possible can be learned, an image can be created using the learning model, and a bone-suppressed image in which blurring is suppressed can be output.

In the image creating device according to the first aspect, it is preferable that the specific portion is a blood vessel, and the learning image generating unit is configured to: the teacher input image generation unit generates a two-dimensional teacher input image including a region of a blood vessel of a subject and a teacher output image showing the blood vessel of the subject, and the learning unit is configured to: the image creating unit performs machine learning based on a teacher input image to which noise is applied and a teacher output image to which noise is applied, thereby learning a learning model for extracting a blood vessel, and is configured to: based on the captured image including the blood vessel, the learned learning model is used to extract the blood vessel and create an image. With such a configuration, since a learning model in which a blood vessel is extracted without removing a noise component as much as possible can be learned, an image can be created using the learning model, and a blood vessel image with blur suppressed can be output.

An image creating method according to a second aspect of the present invention creates a two-dimensional teacher input image of a region including a specific portion of a subject based on three-dimensional data having three-dimensional pixel value data, creates a teacher output image which is an image showing the specific portion or an image from which the specific portion is removed from the teacher input image at the same angle of view, applies the same noise to the teacher input image and the teacher output image, performs machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learns a learning model for extracting or removing the specific portion, and creates an image from which the specific portion is extracted or removed using the learned learning model from the captured image including the specific portion.

In the image creating method according to the second aspect of the present invention, since the same noise is applied to both the teacher input image and the teacher output image by the configuration described above, it is possible to learn image processing such as extracting or removing a specific portion of the teacher input image and the teacher output image, the specific portion being a difference other than the noise component. This eliminates learning of a filter for smoothing noise, and thus can suppress blurring of a created image.

A learning model creation method according to a third aspect of the present invention creates a two-dimensional teacher input image of a region including a specific portion of a subject based on three-dimensional data having three-dimensional pixel value data, creates a teacher output image which is an image showing the specific portion or an image from which the specific portion is removed from the teacher input image at the same angle of view as the teacher input image, applies the same noise to the teacher input image and the teacher output image, and performs machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for extracting or removing the specific portion.

In the learning-completed model creation method according to the third aspect of the present invention, since the same noise is applied to both the teacher input image and the teacher output image by the above-described configuration, it is possible to perform learning by image processing such as extracting or removing a specific portion of the teacher input image and the teacher output image, the specific portion being a difference other than the noise component. This prevents the filter for smoothing noise from being learned, and thus can suppress the blur of the created image.

ADVANTAGEOUS EFFECTS OF INVENTION

According to the present invention, as described above, it is possible to suppress blurring when an image is created using a learning model obtained by machine learning.

Drawings

Fig. 1 is a block diagram showing an image forming apparatus according to an embodiment.

Fig. 2 is a diagram showing an X-ray imaging apparatus that captures an image for creating an image by the image creating apparatus according to the embodiment.

Fig. 3 is a diagram for explaining a learning method and an image creating method according to an embodiment.

Fig. 4 is a diagram for explaining a learning image creating method according to an embodiment.

Fig. 5 is a diagram for explaining a learning method in the case of removing a bone according to an embodiment.

Fig. 6 is a diagram for explaining a learning method in the case of extracting a blood vessel according to an embodiment.

Fig. 7 is a flowchart for explaining a learning process of a learning model according to an embodiment.

Fig. 8 is a flowchart for explaining the image creating process according to the embodiment.

Fig. 9 is a block diagram showing an image forming apparatus according to a modification of the embodiment.

Detailed Description

Hereinafter, embodiments embodying the present invention will be described based on the drawings.

The configuration of an image creating apparatus 100 according to an embodiment of the present invention, an image creating method according to an embodiment, and a learned model creating method according to an embodiment will be described with reference to fig. 1 to 8.

(configuration of image processing apparatus)

First, the configuration of the image forming apparatus 100 according to the first embodiment will be described with reference to fig. 1.

The image creating apparatus 100 is configured to: an image from which a specific portion of an image obtained by X-ray imaging is extracted or removed is created using a learning model that has been learned through machine learning. As shown in fig. 1, the image creating apparatus 100 includes a learning image generating unit 1, a noise applying unit 2, a learning unit 3, and an image creating unit 4. Further, the image forming apparatus 100 is configured to: a CT image can be acquired from the CT imaging apparatus 10. Further, the image forming apparatus 100 is configured to: the radiographic image can be acquired from the radiographic apparatus 20. Further, the image forming apparatus 100 is configured to: the created image can be transmitted to the display device 30.

The image creating apparatus 100 includes a communication unit for acquiring data from the CT imaging apparatus 10 or the X-ray imaging apparatus 20 and transmitting the data to the display apparatus 30. The communication unit includes, for example, an input/output Interface such as an HDMI (registered trademark) (High-Definition Multimedia Interface), a Display Port, and a USB Port.

The image creating apparatus 100 includes, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like as hardware components. The image creating apparatus 100 includes an HDD (Hard Disk Drive) or a nonvolatile memory as a storage unit.

As shown in fig. 2, the X-ray imaging apparatus 20 includes a top plate 21, an X-ray irradiation unit 22, and an FPD (Flat Panel Detector) 23. The top plate 21 is configured to support a subject O (human). The X-ray irradiation unit 22 is configured to irradiate X-rays toward the subject O. The FPD 23 is configured to detect X-rays irradiated from the X-ray irradiation unit 22 and transmitted through the object O.

The learning image generation unit 1 of the image creating apparatus 100 is configured to: a teacher input image and a teacher output image for learning the learning model by the learning unit 3 are generated. Specifically, the learning image generation unit 1 generates, based on three-dimensional data having three-dimensional pixel value data, a teacher input image which is a two-dimensional image of a region including a specific portion of the subject O, and a teacher output image which is an image showing the specific portion or an image excluding the specific portion at the same angle of view as the teacher input image. Specifically, the learning image generator 1 acquires three-dimensional data (data having three-dimensional voxel values) based on a computed tomography image (CT image) obtained by imaging the subject O. The learning image generation unit 1 is configured to: a teacher input image and a teacher output image are generated by digital reconstruction simulation (DRR) based on the three-dimensional data.

The learning image generation unit 1 is configured to: sets of teacher input images and teacher output images are generated from one three-dimensional data (CT image data). For example, the learning image generation unit 1 is configured to: approximately 10 ten thousand sets of teacher input images and teacher output images are generated from one three-dimensional data. The learning image generation unit 1 is configured to: the projection angle, projection coordinates, parameters for generating DRR images, contrast, edge enhancement, and the like of the digital reconstruction simulation are made different from each other, and a plurality of sets of teacher input images and teacher output images are generated.

As shown in fig. 4, when generating an image by digital reconstruction simulation, the learning image generating unit 1 virtually arranges the three-dimensional data 11, the X-ray source 12, and the projection surface 13, and virtually projects the three-dimensional data 11. Then, the learning image generation unit 1 performs calculation based on the three-dimensional data 11 to generate an image 14 projected onto the projection surface 13.

When generating the teacher input image, the learning image generating unit 1 virtually irradiates a portion including a specific portion of the three-dimensional data 11 with X-rays from the X-ray source 12, and adds the CT values (voxel values) that have passed through to calculate each pixel value of the two-dimensional image 14.

When generating the teacher output image, the learning image generating unit 1 virtually irradiates a portion including a specific portion of the three-dimensional data 11 with X-rays from the X-ray source 12, and calculates each pixel value of the two-dimensional image 14 by adding CT values (voxel values) of only the specific portion or by adding CT values excluding the CT values (voxel values) of the specific portion. For example, when a specific portion is extracted, each pixel value of the two-dimensional image 14 is calculated by adding only CT values (voxel values) in a range corresponding to the specific portion. In addition, when extracting a specific part on CT, a known 3D segmentation technique may be used, or a specific part region may be acquired from treatment planning data created by a treatment planning apparatus. When a specific portion is excluded, the CT values excluding the CT values (voxel values) in the range corresponding to the specific portion are added to calculate each pixel value of the two-dimensional image 14.

The learning image generation unit 1 is configured to: when the motion (pulsation) of a blood vessel or the like is set as a specific portion, a teacher input image and a teacher output image are generated based on four-dimensional data including time in addition to a three-dimensional space. This enables highly accurate learning even for an object that moves with the passage of time.

The noise applying unit 2 is configured to: noise is applied to the teacher input image and the teacher output image generated by the learning image generation unit 1. Here, in the present embodiment, the noise applying unit 2 is configured to: the same noise is applied to the teacher input image and the teacher output image. Further, the noise applying unit 2 is configured to: the teacher input image and the teacher output image are subjected to noise adjusted based on noise of a photographing result obtained by photographing the images. That is, the noise applying unit 2 is configured to: noise adjusted based on noise included when the X-ray imaging apparatus 20 performs X-ray imaging is applied. The noise of the X-ray imaging performed by the X-ray imaging apparatus 20 is measured by performing X-ray imaging by placing a uniform substance (a substance having substantially the same X-ray transmittance (for example, air, water, or the like)) between the X-ray irradiation unit 22 and the FPD 23, for example.

The noise applying unit 2 is configured to apply gaussian noise. Further, the noise applying unit 2 is configured to: gaussian noise adjusted to a standard deviation sigma corresponding to a standard deviation sigma of noise of a photographic result obtained by photographing an image is applied to a teacher input image and a teacher output image. That is, the noise applying unit 2 is configured to: gaussian noise adjusted to an equivalent standard deviation σ based on noise included when X-ray imaging is performed by the X-ray imaging apparatus 20 is applied.

Further, the noise applying unit 2 is configured to: the same noise is applied to a group of the teacher input image and the teacher output image generated by the same condition. On the other hand, the noise applying unit 2 is configured to: different noise is applied to each group of the generated teacher input image and teacher output image. For example, the noise applying unit 2 is configured to: noise is applied to each group of the generated teacher input image and teacher output image in such a manner that the parameters are randomly changed.

In the present embodiment, as shown in fig. 3, the learning unit 3 is configured to: machine learning is performed based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, whereby a learning model for extracting or removing a specific portion is learned. The learning unit 3 is configured to: image analysis is performed based on the teacher input image to which noise is applied and the teacher output image to which noise is applied by AI (artificial intelligence). The learning unit 3 is configured to perform learning using a convolutional layer. The learning unit 3 is configured to perform learning by deep learning. The learning unit 3 is configured to perform learning using a neural network.

The learning of the learning model by the learning unit 3 may be performed for each subject O or for each X-ray imaging apparatus 20. In addition, the learning may be performed periodically. Further, a learning model in which the learning data of different subjects O are collected can be created.

In the present embodiment, as shown in fig. 3, the image creating unit 4 is configured to: an image from which the specific portion is extracted or removed is created using a learned learning model from the captured image including the specific portion. That is, the image creating unit 4 is configured to: from the two-dimensional X-ray image including the specific portion captured by the X-ray imaging apparatus 20, a two-dimensional image in which the specific portion is extracted or removed is created using the learning model learned by the learning unit 3. Further, the image creating unit 4 is configured to: an image from which a specific portion is extracted or removed is created from a real-time two-dimensional image captured by the X-ray imaging apparatus 20. For example, during treatment of the subject O or the like, the image creating unit 4 creates an image from which a bone is removed, a blood vessel is extracted, or a tool such as a stent is extracted, from the image which is seen through by the X-ray imaging apparatus 20. The image creating unit 4 transmits the created image to the display device 30. This causes the display device 30 to display the image from which the specific portion is extracted or removed.

An example in which the specific portion is a bone will be described with reference to fig. 5. As shown in fig. 5, the learning image generation unit 1 generates a two-dimensional teacher input image including a region of a bone of the subject O as a specific portion. Further, the learning image generation unit 1 generates a two-dimensional teacher output image from which a bone of the subject O, which is a specific portion, is removed. The noise applying unit 2 applies the same noise to the teacher input image and the teacher output image in two dimensions. The learning unit 3 learns a learning model for removing bones by machine learning from the teacher input image and the teacher output image to which the same noise is applied. Then, using the learning model for removing the bone, an image in which the bone is removed from the X-ray image of the subject O captured by the X-ray imaging apparatus 20 is created by the image creating unit 4.

An example in which the specific portion is a blood vessel will be described with reference to fig. 6. As shown in fig. 6, a two-dimensional teacher input image of a region including a blood vessel of the subject O as a specific portion is generated by the learning image generation unit 1. Further, the learning image generation unit 1 generates a two-dimensional teacher output image in which the blood vessels of the subject O, which is a specific portion, are extracted. The noise applying unit 2 applies the same noise to the teacher input image and the teacher output image in two dimensions. The learning unit 3 learns a learning model for extracting a blood vessel by machine learning from a teacher input image and a teacher output image to which the same noise is applied. Then, using the learning model for extracting the blood vessel, the image creating unit 4 creates an image in which the blood vessel is extracted from the X-ray image of the subject O captured by the X-ray imaging apparatus 20.

(learning model learning process)

Next, a learning model learning process performed by the image creating apparatus 100 will be described with reference to fig. 7.

In step S1, a CT image (three-dimensional data) created by the CT imaging apparatus 10 is acquired. In step S2, a plurality of sets of teacher input images and teacher output images are generated from the CT images.

In step S3, the same noise is applied to the generated teacher input image and teacher output image of the same group. In step S4, the learning model is learned. Thereby, a learned model for extracting or removing a specific portion is generated.

(image creation processing)

Next, an image creating process performed by the image creating apparatus 100 will be described with reference to fig. 8.

In step S11, the imaging result (two-dimensional X-ray image) of the object O obtained by the X-ray imaging apparatus 20 is acquired. In step S12, an image from which a specific portion has been extracted or removed is created using the learned model from the acquired X-ray image.

(Effect of the present embodiment)

In the present embodiment, the following effects can be obtained.

In the present embodiment, as described above, the following are provided: a noise application unit 2 that applies the same noise to the teacher input image and the teacher output image; and a learning unit 3 that performs machine learning based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, and thereby learns a learning model for extracting or removing a specific portion. Thus, since the same noise is applied to both the teacher input image and the teacher output image, it is possible to learn image processing such as extracting or removing a specific portion of the teacher input image and the teacher output image, which is a difference other than the noise component. This prevents the filter for smoothing noise from being learned, and thus can suppress the blur of the created image.

In the present embodiment, as described above, the learning image generation unit 1 is configured to: three-dimensional data is acquired based on a computed tomography image obtained by imaging the object O, and a teacher input image and a teacher output image are generated by digital reconstruction simulation. Thus, a plurality of sets of teacher input images and teacher output images can be easily generated by digital reconstruction simulation from the computed tomography image, and machine learning can be performed using the plurality of sets of images. This can improve the learning accuracy of the learning model.

In the present embodiment, as described above, the noise applying unit 2 is configured to: the teacher input image and the teacher output image are subjected to noise adjusted based on noise of a photographing result obtained by photographing the images. Accordingly, since machine learning can be performed by applying noise corresponding to noise in the case of actually photographing to the teacher input image and the teacher output image, it is possible to effectively suppress the influence of noise and extract or remove a specific portion in the case of creating an image from a photographed image including the specific portion by using a learning model having been learned.

In the present embodiment, as described above, the noise applying unit 2 is configured to apply gaussian noise, and is configured to: gaussian noise adjusted to a standard deviation corresponding to a standard deviation of noise of a photographing result obtained by photographing an image is applied to a teacher input image and a teacher output image. This makes it possible to match the standard deviation of the applied gaussian noise with the noise of the image actually captured, and therefore, it is possible to more effectively suppress the noise from affecting the image creation and extract or remove a specific portion.

In the present embodiment, as described above, the learning image generation unit 1 is configured to: a two-dimensional teacher input image including a region of bone of the subject O and a teacher output image excluding the bone of the subject are generated. Further, the learning unit 3 is configured to: machine learning is performed based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for removing bones. Further, the image creating unit 4 is configured to: from the captured image including the bone, an image from which the bone is removed is created using the learned learning model. Thus, since the learning model in which only the bone is removed without removing the noise component as much as possible can be learned, the image can be created using the learning model, and the bone-suppressed image in which the blur is suppressed can be output.

In the present embodiment, as described above, the learning image generation unit 1 is configured to: a two-dimensional teacher input image of a region including a blood vessel of the subject O and a teacher output image showing the blood vessel of the subject are generated. Further, the learning unit 3 is configured to: machine learning is performed based on the teacher input image to which the noise is applied and the teacher output image to which the noise is applied, thereby learning a learning model for extracting a blood vessel. Further, the image creating unit 4 is configured to: based on the captured image including the blood vessel, the learned learning model is used to extract the blood vessel and create an image. Accordingly, since a learning model in which a blood vessel is extracted without removing a noise component as much as possible can be learned, an image can be created using the learning model, and a blood vessel image in which blurring is suppressed can be output.

(modification example)

The embodiments disclosed herein are considered to be illustrative in all respects, rather than restrictive. The scope of the present invention is defined by the claims rather than the description of the above embodiments, and includes all modifications (variations) within the meaning and scope equivalent to the claims.

For example, in the above-described embodiment, an example is shown in which the apparatus for performing machine learning and the apparatus for performing image creation are the same apparatus, but the present invention is not limited to this. In the present invention, the device for performing machine learning and the device for performing image creation may be different devices. For example, as shown in fig. 9, machine learning can be performed by an image learning apparatus 200 provided with a learning image generation unit 1, a noise application unit 2, and a learning unit 3. Further, the image creation may be performed by the image creation device 300 provided with the image creation unit 4. In this case, the image learning apparatus 200 may be disposed at a location separate from the image creating apparatus 300. The image learning apparatus 200 may be configured by a server computer installed in the cloud.

In the above-described embodiment, an example of a configuration in which a learning model for extracting or removing a specific portion from the imaging result of the X-ray imaging apparatus is learned is shown, but the present invention is not limited to this. In the present invention, a learning model for extracting or removing a specific portion from imaging results of an imaging apparatus other than the X-ray imaging apparatus may be learned. In addition, the present invention can be widely applied to a case where a teacher output image at the time of learning is not a label image but an image of the same manner as a teacher input image.

In the above-described embodiment, an example of a configuration in which an image in which a specific portion is extracted or removed from a captured image is created is shown, but the present invention is not limited to this. In the present invention, an image from which noise is further removed from the image from which the specific portion is extracted or removed may be created.

In the above-described embodiment, the example in which the subject is a human is shown, but the present invention is not limited to this. In the present invention, the subject may be a living body other than a human being, or may be a non-living body.

In addition, although the above embodiment shows an example in which the specific portion is a blood vessel or a bone, the present invention is not limited thereto. In the present invention, the specific portion may be a portion other than blood vessels and bones.

In the above-described embodiment, the example in which the noise applied to the teacher input image and the teacher output image is gaussian noise is described, but the present invention is not limited to this. In the present invention, the noise applied to the teacher input image and the teacher output image may be noise other than gaussian noise.

In the above-described embodiment, for convenience of explanation, an example of processing in the image forming apparatus is described using a flow-driven flowchart in which processing is sequentially performed according to a process flow, but the present invention is not limited to this. In the present invention, the processing of the image forming apparatus may be performed by event-driven (event-drive) processing in which the processing is executed in event units. In this case, the processing may be performed based on a complete event-driven type, or may be performed by combining event-driven and flow-driven types.

Description of the reference numerals

1: a learning image generation unit; 2: a noise applying section; 3: a learning unit; 4: an image creation unit; 100: an image producing device; o: a subject is examined.

16页详细技术资料下载
上一篇:一种医用注射器针头装配设备
下一篇:用于编辑全景射线照相图像的设备和方法

网友询问留言

已有0条留言

还没有人留言评论。精彩留言会获得点赞!

精彩留言,会给你点赞!