JP2021142132A - Image processing apparatus, x-ray imaging apparatus, and image processing method - Google Patents

Abstract

To provide a technology to reduce exposure in an X-ray examination, and predict a high dose equivalent image from a low dose image acquired by an X-ray imaging apparatus.SOLUTION: X-ray generation has randomness and follows Poisson distribution, which is characteristic of an X-ray. An image processing apparatus predicts a high dose image from a low dose image by using correction processing and a learned model specialized in the characteristics of an X-ray. In the correction processing, the image processing apparatus executes correction processing for adjusting noise levels according to brightness for the low dose image, applying the learned model to data in which the noise levels are adjusted, and predicts a high quality image of high dose equivalent. In an X-ray imaging apparatus, when a low dose imaging mode is set in the settings of an imaging mode, an automatic control unit executes control to emit an X-ray with a lower dose than a standard irradiation dose, and low dose image is acquired. An image processing unit predicts a high dose image from the low dose image.SELECTED DRAWING: Figure 1

Description

The present invention relates to a medical X-ray image pickup device such as a mammography device and a CT device, and particularly for a low-dose image containing a large amount of noise acquired by the X-ray image pickup device, a low-dose image to a high-dose image using a trained model. Regarding the technology to predict.

The medical X-ray imaging device is a device that irradiates an imaging target with X-rays, detects the X-ray dose that has passed through the imaging target, and makes an image by utilizing the difference in the amount of X-ray absorption of the tissue to be imaged. It makes it possible to observe the internal morphology of tissues and is used for disease detection and follow-up. In general, the higher the X-ray dose to be irradiated, the better the image quality of the diagnostic image. In particular, the X-rays generated from the X-ray generator are not uniform along the time axis and have randomness. Therefore, by controlling the X-ray irradiation time and increasing the dose, it is possible to increase the dose. The influence of the randomness of X-ray generation is mitigated, and a diagnostic image with good image quality can be obtained.

On the other hand, as the dose increases, the exposure to the human body increases, so there is a need for imaging using an appropriate dose that can obtain good image quality while suppressing exposure. For this reason, recent diagnostic devices are equipped with a function of automatically controlling an appropriate imaging dose while maintaining a certain image quality standard according to the patient and the observation site (Patent Document 1). etc).

In addition, various improvements have been made in the image processing system so that the presence / absence and state of lesions can be easily observed from medical images. For example, it has been proposed to apply a deep learning (DL) technique to improve image quality and determine a specific disease (Patent Document 2 and the like).

Japanese Unexamined Patent Publication No. 2013-233420 Japanese Unexamined Patent Publication No. 2019-181226

The technique disclosed in Patent Document 2 introduces a trained model into the image processing of an MR diagnostic apparatus and attempts to restore an image from an image with data loss. However, unlike the magnetic signal measured by the MR device, X-rays have randomness in the generation of X-rays, and the degree of variation in noise level varies depending on the intensity (dose). As a result, the captured image also has noise variation according to the brightness. Therefore, even if a model trained by simply using a low-dose image and a high-dose image as a set is used, a good learning effect cannot be obtained. Further, since the imaging voltage and current are controlled according to the thickness and hardness of each imaging target, if the trained model is applied as it is, insufficient prediction or over-prediction is likely to occur.

An object of the present invention is to provide an X-ray diagnostic apparatus capable of obtaining a low-exposure and high-quality diagnostic image. Specifically, it is an object to accurately predict an image equivalent to a high dose from a low dose image acquired by an X-ray medical imaging apparatus by image processing, thereby reducing exposure during an X-ray examination. ..

In order to solve the above problems, the present invention predicts a high-dose image from a low-dose image by using a correction process specialized for X-ray characteristics and a trained model. The correction process includes a process of aligning the noise level according to the brightness.

That is, the image processing device of the present invention inputs a correction processing unit that corrects the X-ray image acquired by the X-ray imaging device and an image after correction by the correction processing unit, and learns a low-dose image and a high-dose image. Using the trained model learned by using it as data, a high-dose image prediction unit that outputs an image that can be imaged at a higher dose than the dose at which the input image is imaged is provided, and the correction processing unit is the input. The difference in noise level due to the difference in dose between the low-dose image and the high-dose image is corrected for the image.
This trained model is a trained model trained using a low-dose image that has undergone the same correction processing as the input image as training data.

Further, the X-ray imaging apparatus of the present invention uses an imaging unit that collects image data of a subject, a control unit that controls the imaging unit according to an imaging mode in the imaging unit, and image data collected by the imaging unit. It includes an image processing unit that performs image processing, and the image processing unit includes the functions of the image processing apparatus of the present invention.

Specifically, the X-ray imaging apparatus of the present invention has an X-ray irradiation unit that irradiates X-rays and an X-ray detection unit that detects X-rays that are irradiated from the X-ray irradiation unit and transmitted through the subject. A line imaging unit, an image processing unit that creates an X-ray image using the X-ray data detected by the X-ray detection unit, and a mode setting unit that sets one of a plurality of imaging modes having different X-ray doses. The X-ray irradiation unit, the X-ray detection unit, and an automatic control unit that controls the image processing unit are provided according to the imaging mode set by the mode setting unit, and the image processing unit is the image processing device described above. Therefore, the automatic control unit controls the image processing unit to perform processing by the correction processing unit and the high-dose image prediction unit when the imaging mode is set to the low-dose imaging mode.

Further, the image processing method of the present invention is an X-ray image obtained by imaging a standard dose or a dose higher than the standard dose from an X-ray image captured at a dose lower than the dose set as a standard in the X-ray image pickup apparatus. This is an image processing method that predicts an image. After performing processing on the X-ray image to be processed to correct the difference in noise level due to the difference in dose between the low-dose image and the high-dose image, the low dose Using a trained model trained using an image and a high-dose image as training data, an X-ray image after the correction process is input, and an image that can be imaged at a higher dose than the dose at which the image was imaged is obtained. It is characterized by outputting.

According to the present invention, an image corresponding to a high dose can be predicted at a dose lower than the X dose in the current standard imaging mode, and the reduction of the exposure dose of the subject, the period diagnosis and the early detection can be expected. In addition, since dose control is often time-based, low-dose imaging can shorten the imaging time, alleviate pain caused by patient pressure during imaging, and improve the efficiency of examination technicians.

The figure which shows the whole structure of the X-ray image pickup apparatus The figure which shows the structure of the mammography apparatus as an example of an X-ray imaging apparatus. The figure which shows the structure of the image processing apparatus The figure which shows the structure of the main part of the image processing part of 1st Embodiment, and the flow of processing Flow chart showing the processing process of the X-ray imaging device Diagram showing the structure of the trained model (A) to (C) are diagrams for explaining the process of aligning the noise levels by the correction process. The figure explaining the processing of the correction processing part The figure which shows the processing of the modification of 1st Embodiment The figure explaining the detection timing of the X-ray detection part of the 2nd Embodiment The figure explaining an example of the correction and image quality improvement processing of 2nd Embodiment The figure explaining another example of the correction and image quality improvement processing of 2nd Embodiment The figure which shows the structure of the CT apparatus as an example of the X-ray imaging apparatus.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
<First Embodiment>
First, an embodiment of an X-ray imaging apparatus to which the image processing apparatus of the present invention is applied will be described. As the X-ray imaging device, any device such as a mammography, an X-ray CT device, an X-ray imaging device, or the like that images an image using X-rays transmitted through a subject can be applied.

As shown in FIG. 1, the X-ray imaging apparatus 10 includes an X-ray irradiation unit 11 that irradiates the subject 20 with X-rays, an X-ray detection unit 12 that detects X-rays that have passed through the subject 20, and X-ray irradiation. The irradiation control unit 13 and the detection control unit 14, which control the operations of the unit 11 and the X-ray detection unit 12, respectively, the mode setting unit 16 and the mode setting unit 16 for setting any of a plurality of imaging modes having different X-ray doses. It includes an automatic control unit 15 that controls the irradiation control unit 13 and the detection control unit 14 according to a set mode, and an image processing unit 30 that inputs an X-ray signal detected by the X-ray detection unit 12 and creates an X-ray image. ing. The irradiation unit 11, the X-ray detection unit 12, and their control units 13 and 14 are collectively referred to as an imaging unit.

The X-ray imaging apparatus 10 further includes a display unit 17 for displaying an X-ray image and an operation input unit 18 for a user to set desired imaging conditions and select a mode. Further, although not shown, a storage device for storing programs and data required for image processing may be provided.

The configuration of the imaging unit differs depending on the type of the X-ray imaging device 10, but is the same as the configuration of a known X-ray imaging device.

For example, when the X-ray imaging device 10 is a mammography device 10A, as shown in FIG. 2, the X-ray irradiation unit 11 X-rays due to collision of electron beams from the cathode 111 and the cathode 111 that generate electron beams. It is an X-ray generator equipped with an anode (target) 112 that generates rays, a filter 113 that controls an irradiation angle and an irradiation pattern, and the X-ray detection unit 12 is an FPD (Flat Panel) that combines a scintillator and a photoelectric conversion element. It is composed of Detector) and II (Image Intensifier). FIG. 2 illustrates an FPD having an automatic exposure control AEC (Automatic Exposure Control) function, but the present invention is not limited to this. Further, a grid 122 that limits the incident direction of X-rays is fixed on the subject (breast) side of the FPD. Further, a compression plate 121 that presses the breast as a subject toward the detector side is attached between the X-ray irradiation unit 11 and the X-ray detection unit 12 so as to be movable in the vertical direction in the drawing.

Mammography is an important method for breast cancer examination, and since it is necessary to diagnose the presence or absence of a mass or calcification in the intersecting mammary glands, the demand for image quality is very high. This embodiment is suitable for mammography that requires such high image quality.

The mode setting unit 16 can set an automatic low-dose imaging mode in addition to the standard imaging mode, for example, according to the instruction from the operation input unit 18. The standard imaging mode is a mode in which an appropriate dose is automatically determined according to the thickness of the subject 20 (thickness of the portion through which X-rays are transmitted), the size of the subject, the age, and the like, and imaging is performed. The dose determined in the standard imaging mode is defined as 100% dose here, and the dose in the low dose imaging mode is determined at a predetermined ratio (50%, 30%, etc.) to the dose. The low-dose imaging mode may be capable of setting a plurality of low-dose modes having different proportions. The low-dose imaging mode may be set by user operation or may be set by default.

The image processing unit 30 creates a diagnostic image showing the internal structure of the tissue that can be diagnosed by a doctor using the signal detected by the X-ray detection unit 12, and displays it on the display unit 17. Therefore, the image processing unit 30 includes a data configuration unit 31 that creates image data and the like using the detection signal of the X-ray detection unit 12, and a data storage unit 32 that stores the image data and the like created by the data configuration unit 31. Image data selection unit 33 that selects image data that needs to be displayed or further image processing from the image data saved in the data storage unit 32, and image data (low dose image data) sent from the image data selection unit 33 are input. According to user specifications from the high image quality processing unit 34 that performs processing for high image quality, the display conversion unit 35 that converts image data into display images, and the operation input unit 18, display conversion parameters, for example, display The adjustment processing unit 36 for adjusting the brightness, size, angle, etc., and the display form is provided. The configuration of the image processing unit 30 shown is an example, and some elements may be omitted or another element (not shown) may be added.

A part or all of the functions of the image processing unit 30 can be realized as software executed by the CPU. Further, a part related to image data creation and a part of the image processing unit 30 may be realized by hardware such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

The X-ray imaging apparatus 10 of the present embodiment includes a mode setting unit 16 for setting an automatic low-dose imaging mode, and a high-quality processing unit 34 of the image processing unit 30 for an image acquired in the low-dose imaging mode. It is characterized by predicting an image (high-dose image) that will be obtained in the standard imaging mode, and other configurations are the same as those of the known medical imaging apparatus 10 and the image processing unit 30 associated therewith.

Note that FIG. 1 shows a configuration in which the image processing unit 30 is incorporated in a part of the X-ray imaging device 10, but as shown in FIG. 3, the image processing unit 30 is independent of the X-ray imaging device 10. It may be an image processing device, and in that case, the X-ray imaging device 10 and the image processing device 30 may be connected via a network such as the Internet or an intranet or a dedicated communication means. In this case, the image processing device 30 may be connected to a dedicated user interface (UI) 50 or a storage device 40. The UI is provided with a display unit 51 for displaying the image processing result and an operation input unit 52 for inputting conditions and the like necessary for image processing. The storage device 40 stores various programs and learning models (including trained models) that can be used by the image processing device 30. Hereinafter, the function of the image processing unit 30 shown in FIG. 1 will be described, but unless otherwise specified, the function is the same as that of the image processing device shown in FIG. 3, and the display unit 17 and the operation input unit 18 in FIG. 1 can be referred to. , The display unit 51 and the operation input unit 52 in FIG. 3 are also included.

Next, the configuration of the image processing unit 30 including the details of the high image quality processing unit 34 will be described with reference to FIGS. 3 and 4. 3 and 4 show only the elements mainly related to the high image quality processing among the various functions of the image processing unit 30, and some of the elements shown in FIG. 1 are omitted.

As shown in FIGS. 3 and 4, the high image quality processing unit 34 introduces a correction processing unit 341 and a learning model that level the brightness-dependent noise peculiar to the X-ray image to the image selected by the image selection unit 33. It includes a model introduction unit 342 and a high image prediction unit 343 that predicts a high-dose image based on the output of the learning model introduced by the model introduction unit 342. The learning model is stored in the storage device 40 connected to the image processing unit 30. The storage device 40 may be an external storage device connected to a CPU constituting the image processing unit 30, an HDD, a portable medium, or a storage device provided in a cloud on the Internet. The details of the processing of the high image quality processing unit 34 will be described later.

Next, an outline of the operation of the X-ray imaging apparatus 10 in the low-dose imaging mode will be described with reference to the flow of FIG. When the automatic low dose mode is set by the mode setting unit 16 (S501), the automatic control unit 15 calculates an irradiation dose lower than the standard imaging mode according to the subject according to the setting mode and is involved in the irradiation dose. Parameters such as irradiation voltage, irradiation current, irradiation time, and filter type are passed to the irradiation control unit 13 (S502). Further, the automatic control unit 15 passes parameters such as the detection timing and the number of times to the detection control unit 14. The irradiation control unit 13 controls the X-ray irradiation unit 11 under the received imaging conditions to generate X-rays and irradiate the subject with X-rays. The X-ray detection unit 12 detects the X-rays that have passed through the subject under the control of the detection control unit 14.

The data composition unit 31 of the image processing unit 30 configures the data detected by the X-ray detection unit 12 into a digital image and stores it in the data storage unit 32. Along with this, the image selection unit 33 selects one or a plurality of images to be processed from the data storage unit 32, that is, images captured at a low dose, and passes them to the high image quality processing unit 33 (S503). The high image quality processing unit 33 reads out the learning model (S504), corrects the image received from the data selection unit 33 (S505), and predicts the high-dose image using the learning model (S506) to improve the image quality. ..

The display conversion unit 36 converts the high-quality image into an image that can be easily confirmed on the display monitor (S507). At this time, when the user operation for the operation input unit 18 is accepted, the adjustment processing unit 36 changes the display conversion parameters and the display form (S508, S509). The image processing unit 30 displays the diagnostic image that has undergone each of the above processes, if necessary, together with the image before processing and other incidental information.

Next, the details of the above processes S504 to S506, that is, the details of the processing of the high image quality processing unit 33 of the image processing unit 30 will be described.

[S504: Introduction of trained model]
The model introduction unit 341 reads the trained model from the storage device 40. The trained model is a model trained using a set of a low-dose image and a high-dose image as training data. The trained model is basically a combination of a convolution layer, a pooling layer, and the like, and the parameters of each layer are determined after training.

FIG. 6 shows an example of the structure of the learning model. As shown in the figure, a CNN-based architecture is used to construct the learning model. In CNN, the input image is divided into a large number of patches, processed for each patch, the features of the low-dose image composed of a plurality of dimensions are converted into the features of the high-dose image by non-linear mapping, and the high-dose image is output. The parameters of each layer of the CNN are determined so that the error between the output and the high-dose image of the training data is below a predetermined level. However, the image data used for the training data is data in which the difference in noise level and dispersion due to the difference in dose is corrected.

In the X-ray image, the variation in noise changes depending on the brightness, and the noise level and its dispersion differ between the low-dose image and the high-dose image. As shown in FIG. 7A, the variation in noise of the X-ray image changes depending on the luminance value when the luminance value is on the horizontal axis and the noise dispersion is on the vertical axis, and the higher the luminance value, the greater the variation. Even if a high-dose image is predicted using an image in which such noise variation differs depending on the brightness value, for example, using a trained model trained to predict a high-dose image from a low-dose image, the position in the image can be predicted. Overcorrection and undercorrection occur every time, and effective learning cannot be performed. In addition, the high-dose image, which is the true value used for learning, also has noise, which has a variation different from the noise of the low-dose image, so the relationship between the two changes randomly, and learning becomes difficult. It doesn't work.

In the present embodiment, first, when learning the training model, the X-ray image (low-dose image and high-dose image) is not used as the input image for direct prediction, but both images are subjected to two correction processes of blur correction and tilt correction. After aligning the noise of the image signal of, the input image is used. Similarly, when making a prediction using the trained model, the same blur correction and tilt correction are performed on the low-dose image, and then the input image is input to the trained model to predict the high-dose image.

As a result, it is possible to suppress excess or deficiency of prediction and to predict an image corresponding to a high-dose image. The correction processing for the training data is the same processing as the correction processing for the input data when the high-dose image is predicted using the trained model, and is the same processing as the correction processing for the input data. Alternatively, it may be performed by the correction processing unit 342 of the image processing unit 30 connected to the X-ray imaging apparatus 10, or may be performed by using a computer or the like different from such an image processing unit 30. Further, the low-dose image used for learning preferably has the same dose (% with respect to the dose in the normal mode) as the low-dose image selected by the image selection unit 33.

The trained model read from the storage device 40 by the model introduction unit 341 is used in the high-quality image prediction unit 343 when predicting a high-quality image.

[S505: Correction process]
When the correction processing unit 342 receives the image acquired by the low-dose imaging from the image selection unit 33, the correction processing unit 342 performs the same correction processing as the correction processing of the learning data on the low-dose image.

As described above, the process in the correction processing unit 342 is a process for aligning the difference in noise level between the low-dose image and the high-dose image. Specifically, as shown in FIG. 8, the low-dose image 80 is subjected to the blurring process S801 using a smoothing filter such as a low-pass filter to form a blurred image 81 of the low-dose image. Next, the original low-dose image 80 is divided by the blurred image 81 (S802) to obtain an image (feature image) 82 from which noise and edges are extracted. As shown in FIG. 7B, the noise characteristics of the image after such blur correction have more variation in noise at low luminance than before processing (FIG. 7A), but at high luminance. The noise variation is greatly improved, and the noise variation depending on the brightness is eliminated as a whole.

Further, it is known that the signal value of the X-ray image has a slope according to the dose, and the slope is statistically obtained. The correction processing unit 342 corrects the inclination of the signal value according to the dose (S803), and prevents the prediction accuracy from being lowered due to the inconsistency of the inclination between the input image and the predicted high-dose image. To correct the slope, the inverse function of the statistically obtained slope function is applied to the feature image (image with noise and edges extracted) 82 extracted after the blurring process, and the slope is reversed around the center value of the signal. It is a process to be. The noise characteristics (luminance value dispersion value) of the image after the tilt correction are as shown in FIG. 7 (C).
The above steps S801 to S803 are the correction processing S505 performed by the correction processing unit 342 on the low-dose image.

[S506: Prediction of high-dose images]
First, the high-dose image prediction unit 343 outputs a prediction image 83 from which noise has been removed by using the trained model 45 read out in step S503 described above as an input image after correction processing (S804). The predicted image is a high-dose feature image in which only the edges are extracted. By multiplying this prediction result by the blur image 81 (DC signal component) obtained by the blur correction S801 (S805), a high-quality prediction image (high dose prediction image) 84 with reduced noise can be obtained.

[S507, S508: Image display]
The image processing unit 30 causes the display unit 17 to display the high dose prediction image 84 created by the high image quality processing unit 34. The mode of display is not particularly limited, but for example, only the high-dose prediction image may be displayed, or the low-dose image used for the prediction may be displayed side by side. Information other than images, such as noise and its luminance characteristics, may be displayed. Further, a user change or the like for the displayed image may be accepted via the operation input unit 18. In that case, the adjustment processing unit 36 calculates the parameters to be adjusted according to the command input from the operation input unit 18, and passes them to the display conversion unit 37, and the display conversion unit 37 changes the display.

According to this embodiment, since image prediction using a trained model is performed using only images of features excluding differences in noise levels between images captured at different doses, high learning in the construction of a learning model is performed. The effect can be obtained. By using such a trained model, the accuracy of prediction can be improved.

Further, the inclination of the signal value of the X-ray image differs depending on the dose, but according to the present embodiment, by performing the inclination correction in the correction process, the correction is excessive or insufficient even if the image is captured under conditions other than the standard dose. The accuracy of prediction can be improved without causing the above. By applying this embodiment to mammography, it is possible to meet the demand for high image quality and provide information useful for diagnosing the presence or absence of a mass or calcification in the mammary gland.

<Modification example>
In the first embodiment, the processing target of the high image quality processing unit 34 is the image data after the signal detected by the X-ray detection unit is corrected by the data configuration unit 31, such as scattered ray correction and beam hardening correction. However, it may be unprocessed image data (Raw data) that has not been subjected to such correction processing. In this case, as shown in FIG. 9, the trained model 45B trained using the Raw data is used, and the high dose prediction Raw image which is the output of the high image quality processing unit 34 using the trained model 45B is used. , The display conversion unit 35 performs a process of converting Raw data into corrected image data. Alternatively, the correction itself can be incorporated into the trained model. That is, a predicted high-dose image can be obtained by using a set of Raw data and a corrected high-dose image as learning data.

<Second embodiment>
In the first embodiment, the high-dose image is predicted from one low-dose image, but in the present embodiment, the high-dose image is predicted by using a plurality of low-dose images having different doses. The configuration of the device is the same as the configuration shown in FIGS. 1 and 3, and the present embodiment will be described below focusing on the differences from the first embodiment.

In the present embodiment, when the low-dose imaging mode is set in the mode setting unit 16, the automatic control unit 15 controls the X-ray detection unit 12 via the detection control unit 14, and the X-ray detection unit 12 is different in plurality. The detection time is controlled so that the X-ray signal is detected at the timing. For example, as shown in FIG. 10, if the time point at which a low dose of 50% with respect to the standard dose is detected is tun, it is accumulated in the X-ray detector at a plurality of time points t1, t2 ... The data is read out, and the intermediate accumulated data d1, d2 ... Are acquired. The intermediate accumulated data has a lower dose than the final accumulated data (50%), for example, the dose is 10%, 20%, 30%, and so on.

The data configuration unit 31 creates each image using these intermediate accumulated data and passes it to the data storage unit 32. The image selection unit 33 selects a plurality of low-dose images having different doses and passes them to the high-quality processing unit 34. At this time, the low-dose image passed by the image selection unit 33 to the high-quality processing unit 34 may be all of the plurality of low-dose images stored in the data storage unit 32, or may be a combination of predetermined low-dose images. For example, a combination of 20% and 30% low-dose images, a combination of 20%, 30% and 50% low-dose images, and the like are selected. The selection criteria may be set in advance by default, or when the user selects the low-dose imaging mode, the selection can be selected from a plurality of combinations, can be arbitrarily selected, or can be set by the user. May be good.

The trained model introduced by the model introduction unit 341 is a trained model similar to the first embodiment, and may be a plurality of trained models in which a plurality of low-dose images having different doses are trained as input training data. Alternatively, a trained model (a learning model having a plurality of input channels) may be used in which a plurality of low-dose images are trained to output one high-dose image as input learning data. Such a trained model 45A is stored in the storage device 40, and the model introduction unit 341 reads the trained model 45A into the high image quality processing unit 34 (high dose image prediction unit 343).

When the high image quality processing unit 34 receives a plurality of images from the image selection unit 33, as shown in FIG. 11, the correction processing unit 341 first corrects the images I1, I2, ... In, respectively. Run. That is, as shown in FIG. 8, blurring processing and tilt correction are performed to form a low-dose feature image containing only edges and noise. The high image prediction unit 345 inputs a plurality of corrected image data (low dose feature image) to the trained model 45A, outputs image data (high dose feature image) in which only edges are extracted, and blurs the image data (high dose feature image). A high dose prediction image is obtained by multiplying the blurred image obtained by the processing.

When a trained model trained for each low-dose image having a different dose is prepared, the high-quality processing unit 34 uses the corrected low-dose images I1, I2, as shown in FIG.・ ・ In is input to the corresponding trained models 45a, 45b, ..., Respectively, the outputs of the plurality of trained models are fused, and then, for example, a blurred image obtained from the image having the highest dose is applied to make the final result. Predictive high-dose image. Alternatively, after applying each blurred image to the edge image that is the output of each trained model, they are fused to obtain a final predicted high-dose image.

The image processing unit 30 causes the display unit 17 to display the predicted high-dose image. The display mode is the same as that of the first embodiment, but the predicted high-dose image is displayed together with the low-dose image used for the prediction, and the percentage of the low-dose image used for the prediction is displayed. You may do it. Upon seeing this result, the user can change the low-dose image to be combined and repeat only the image processing so that the image with the best image quality can be obtained. That is, according to the present embodiment, it is possible to achieve high image quality only by image processing in which the combination of images is different without performing imaging again.

Note that FIG. 10 shows an example of acquiring a plurality of images having different doses at different detection points, but the control of acquiring a plurality of images having different doses is not limited to the control of the detection time, for example, the irradiation control. Control of the irradiation amount via the unit 13, control of the combination of temporary blocking by the filter (113: FIG. 2) and the detection control unit 14, and control of subdividing the detected dose continuously recorded within the imaging time by digital processing. Etc. are also possible.

Further, as a combination of low-dose images, a case of combining different low-dose images has been described, but one low-dose image (after correction processing) may be used, or a plurality of same-dose images and other dose images may be used. May be combined. In the case of the trained model 45A of a plurality of input channels, such a combination can be performed at the time of input to the trained model 45A, and as shown in FIG. 12, a different trained model is used for each dose. In this case, it can be realized by appropriately weighting the output when synthesizing the output.

According to this embodiment, the accuracy of prediction can be further improved by using low-dose image data having a plurality of different doses. Further, the image quality can be improved by repeating only the image processing according to the prediction result.

<Third embodiment>
In the first embodiment, the X-ray imaging apparatus to which the present invention is applied has been described mainly by taking mamonmography as an example, but the present embodiment is an embodiment applied to a CT apparatus.

As shown in FIG. 13, the CT apparatus 10C comprises an X-ray generator (X-ray irradiation unit 11) 11C and an X-ray detector (X-ray irradiation unit 12) 12C in which detection elements are arranged in an arc shape. Relative between the rotating plate 19 which is arranged so as to face each other and has an opening for inserting the subject 20 in the center, the mechanism for rotating the rotating plate 19 (not shown), the bed 21 on which the subject 20 is laid down, and the rotating plate 19. It is equipped with a moving mechanism (not shown) that changes the position.

In the CT apparatus 10C, the rotating plate 19 is rotated to irradiate X-rays while gradually changing the imaging range of the subject 20, and the obtained transmitted X-ray data is fused by using a calculation such as convolution. An image (CT) image of a cross section of a sample can be taken. Further, by moving the bed 21 back and forth with the rotation of the rotating plate 19, it is possible to take an image in the slide direction, and three-dimensional image data can be obtained.

In such a CT device 10C, in addition to the irradiation amount from the X-ray generator 11C, the X-ray to be irradiated to the subject 20 is controlled by controlling the number of rotations (rotational speed) per hour and the moving speed of the bed 21. The dose of the line can be controlled. In the low dose mode, the automatic control unit 15 sets, for example, the bed moving speed faster than in the normal mode. The image quality of the CT image obtained as a result is deteriorated because the density of the cross section in the slide direction is low.

As the learning model, a low-quality CT image and a high-quality CT image in the normal mode are used as training data to be trained. Other processing is the same as that of the first embodiment, and after performing correction processing for leveling the noise level due to the difference in dose, high image quality is performed using the learning model. It should be noted that it is the same as in the first embodiment that the learning data to which the same correction processing is performed is used.

It is also possible to process the projection data before reconstructing it into a CT image instead of the CT image. In this case, after increasing the dose of the low-dose projection data, the CT image is reconstructed by convolution or the like.

10 X-ray imaging device 11 X-ray irradiation unit 12 X-ray detection unit 13 Irradiation control unit 14 Detection control unit 15 Automatic control unit 16 Mode setting unit 17 Display unit 16 Operation input unit 20 Subject 30 Image processing unit (image processing device)
31 Data selection unit 32 Data storage unit 33 Image selection unit 34 High image quality processing unit 341 Model introduction unit 342 Correction processing unit 343 High-dose image prediction unit 35 Display conversion unit 36 Adjustment processing unit 40 Storage device 45 Learned model 50 UI
51 Display unit 52 Operation input unit

Claims (13)

A correction processing unit that corrects the X-ray image acquired by the X-ray imaging device, and
Using the trained model that was trained by inputting the image corrected by the correction processing unit and using the low-dose image and the high-dose image as training data, the image was imaged at a higher dose than the dose when the input image was imaged. Equipped with a high-dose image prediction unit that outputs possible images,
The correction processing unit is an image processing device that corrects a difference in noise level due to a difference in dose between a low-dose image and a high-dose image with respect to the input image. The image processing apparatus according to claim 1.
The trained model is an image processing apparatus characterized in that the trained model is a trained model trained using a low-dose image subjected to the same correction processing as the input image as the training data. The image processing apparatus according to claim 1.
The processing performed by the correction processing unit includes a process of creating a blurred image by performing a blurring process on the input image, and a feature of extracting edges and noise of the input image using the blurred image and the input image. Including the process of creating an image
The high-dose image prediction unit inputs the feature image into the trained model, obtains a high-dose feature image, and creates a high-dose prediction image using the high-dose feature image and the blurred image. An image processing device characterized by. The image processing apparatus according to claim 1.
The high-dose image prediction unit is an image processing device characterized in that one high-dose prediction image is created by using a plurality of images having different doses at the time of imaging as input images. The image processing apparatus according to claim 4.
The trained model is an image processing device characterized in that it is a trained model in which a plurality of low-dose images and one high-dose image are trained as training data. The image processing apparatus according to claim 4.
The high-dose image prediction unit uses a plurality of trained models learned from each of the low-dose images having different doses as the training data, and processes the plurality of input images with the corresponding trained models. An image processing device characterized in that one high-dose prediction image is created by fusing the processing results. The image processing apparatus according to claim 1.
An image processing apparatus characterized in that the X-ray image is either raw image data acquired by the X-ray imaging apparatus or image data corrected by the X-ray imaging apparatus. The image processing apparatus according to claim 1.
An image processing device characterized in that the X-ray imaging device is a mammography device. An X-ray irradiation unit that irradiates X-rays, an X-ray imaging unit that has an X-ray detection unit that detects X-rays that are irradiated from the X-ray irradiation unit and transmitted through the subject, and an X-ray imaging unit.
An image processing unit that creates an X-ray image using the X-ray data detected by the X-ray detection unit, and an image processing unit.
A mode setting unit that sets one of multiple imaging modes with different X-ray doses,
An automatic control unit that controls the X-ray irradiation unit, the X-ray detection unit, and the image processing unit according to the imaging mode set by the mode setting unit.
It is an X-ray imaging apparatus equipped with
The image processing unit is the image processing apparatus according to any one of claims 1 to 5, and the automatic control unit performs the correction processing when the imaging mode is set to the low-dose imaging mode. An X-ray imaging apparatus characterized in that the image processing unit is controlled so as to perform processing by the unit and the high-dose image prediction unit. The X-ray imaging apparatus according to claim 9.
When the imaging mode is set to the low-dose imaging mode, the automatic control unit collects intermediate data at a plurality of different timings before the X-ray detector collects the set low-dose image data. Control the output and
The image processing unit is an X-ray imaging apparatus characterized in that it performs processing for predicting a high-dose image by using at least two or more data of the low-dose image data and the intermediate data. The X-ray imaging apparatus according to claim 9.
The image processing unit is an X-ray imaging apparatus further comprising a display conversion unit that converts a high-dose prediction image created by the high-dose image prediction unit into a display image. It is an image processing method that predicts an X-ray image obtained by imaging with a standard dose or a dose higher than the standard dose from an X-ray image captured at a dose lower than the dose set as a standard in an X-ray imaging device. hand,
After processing the X-ray image to be processed to correct the difference in noise level due to the difference in dose between the low-dose image and the high-dose image,
Using a trained model trained using a low-dose image and a high-dose image as training data, the corrected X-ray image can be input and imaged at a higher dose than when the image was imaged. An image processing method characterized by outputting an image. The image processing method according to claim 12.
In the correction process, the input image is blurred to create a blurred image, and the blurred image and the input image are used to create an extracted image obtained by extracting edges and noise of the input image. An image processing method characterized by including processing to be performed.

Images