Denoising method, denoising program and medical image diagnostic device

Abstract

To perform faster denoising.SOLUTION: A denoising method according to an embodiment denoises a plurality of input images. The denoising method includes: grouping the plurality of input images into a plurality of groups; performing, for each group, a first sequence of T1 denoising sampling steps using a probabilistic model, starting with the initial representative image for the group, to generate a corresponding sequence of representative images for the group; and performing, for each input image in the group in relation to each group of the plurality of groups, a second sequence of T2 denoising sampling steps using the acquired probabilistic model, starting with the representative image denoised by the final denoising sampling step of the first sequence, to generate a final image that is a restored image corresponding to the input image, where T=T1+T2.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The embodiments disclosed in this specification and the drawings relate to a noise removal method, a noise removal program, and a medical imaging diagnosis apparatus.

Background Art

[0002] Deep learning-based image restoration models typically use a series of steps to modify an input image to a desired degree. When a series of images are acquired, the deep learning model can process the sequence by repeating a series of steps for each image within the series. The number of steps used to process the sequence can affect the processing time and capacity required by the deep learning model.

[0003] The foregoing description of the "Background Art" is for generally presenting the context of the present disclosure. The inventor's research within the scope described in this Background Art section, as well as aspects of the description that may not be recognized as prior art at the time of filing, are not expressly or implicitly recognized as prior art to the present disclosure.

[0004] The foregoing paragraphs are provided as an introduction to the overview and are not intended to limit the following claims. The described embodiments will be best understood by reference to the following detailed description, along with the accompanying drawings, which will together provide further advantages.

Prior Art Documents

Non-Patent Documents

[0005]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to perform faster noise removal. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problems. The problems corresponding to the respective effects of each configuration shown in the embodiments described later can also be positioned as other problems.

Means for Solving the Problems

[0007] The noise removal method according to the embodiment removes noise from a plurality of input images. The noise removal method obtains a diffusion-based probability model learned to perform noise removal over T (where T is an integer of 2 or more) steps using at least one target image and at least one conditional image, groups the plurality of input images into a plurality of groups, determines an initial representative image for each group of the plurality of groups, and for each group of the plurality of groups, uses the obtained probability model to perform a first sequence of T1 (where T1 is an integer of 1 or more) noise removal sampling steps starting from the initial representative image of the group to generate a corresponding sequence of the representative image of the group, and for each group of the plurality of groups, uses the obtained probability model to perform a second sequence of T2 (where T2 is an integer of 1 or more) noise removal sampling steps starting from the representative image noise-removed by the last noise removal sampling step of the first sequence to generate a final image that is a restored image corresponding to the input image for each input image within the group. Also, T = T1 + T2.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5A

Figure 5B

Figure 5C

Figure 6

Figure 7

DETAILED DESCRIPTION OF THE INVENTION

[0009] Referring now to the drawings, like reference numerals indicate the same or corresponding parts throughout the several views.

[0010] Throughout this specification, references to "one embodiment", "specific embodiment", "embodiment", "implementation", "example", or similar terms mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Thus, appearances of such phrases or various places throughout this specification are not necessarily all referring to the same embodiment. Further, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments without limitation.

[0011] In one embodiment, the present disclosure relates to a system and method for image restoration using a deep learning-based model. Image restoration techniques include, but are not limited to, noise removal, blur removal, resolution improvement (e.g., super-resolution imaging), and image / signal reconstruction (e.g., compressive sensing). These techniques can be used independently or in combination, respectively, to improve the visibility of features within an image. Image restoration has important applications in medical imaging modalities such as Computed Tomography (CT) scanning and Magnetic Resonance Imaging (MRI), which are often affected by noise due to physical interactions within the imaging system. It will be understood that the systems and methods described herein are not limited to medical imaging applications and can be used in various types and technologies of imaging. In particular, the method of the present disclosure is useful for processing any volume of image data (e.g., a series of images or image slices) acquired over a spatial or temporal span.

[0012] A generative deep learning-based model can be used to reduce noise and similar artifacts in the acquired image and generate a restored image of higher quality than the acquired image. In one embodiment, an image can be denoised by using the generative model to transform a first data distribution (noisy image data) into a second data distribution (restored image data). In one embodiment, the generative model can be a Denoising Diffusion Probabilistic Model (DDPM). The DDPM is described herein as an exemplary example of a class of generative models, and it can be understood that other types of probabilistic models, particularly diffusion-based probabilistic models for image restoration, are also suitable for the method of the present disclosure.

[0013] DDPM can be used to denoise images in a series of diffusion steps. DDPM can be learned to denoise images in an iterative process, and DDPM generates an increasingly denoised image (a more denoised image) at each diffusion step. Note that in this specification, for a model such as DDPM, "the model is learned" is synonymous with "the model is trained". DDPM can be learned to denoise images by converting a first probability distribution corresponding to an input image (e.g., a noisy image) into a second probability distribution corresponding to an output image (e.g., a denoised image). In one example, the first probability distribution can be a normal distribution corresponding to the normal (Gaussian) noise present in the acquired image. DDPM can be learned to remove noise by converting the normal probability distribution into a predicted distribution corresponding to the denoised and restored image.

[0014] In one embodiment, the DDPM can be trained using a set or sequence of training images (also referred to as training images or training pictures). The set of training images can include a target image that is a clean image or a denoised image (denoised picture), and a noisy image generated from the target image. In one embodiment, the noisy training image can be generated by applying modeled noise (e.g., Gaussian noise) to the target image in one or more steps. The set of training images can include images with increasing amounts of noise. The set of training images can further include a pure noise image generated from the target image. In one embodiment, the modeled noise can be similar to or based on the type of noise expected in the acquired image used by the DDPM for restoration. In one embodiment, the target image can be similar to or based on the type of image used by the DDPM for recovery. Each training image can be input to the DDPM. The DDPM can be trained to denoise the input training image and output a restored image at each step of a series of diffusion steps. The series of diffusion steps can correspond to one or more steps used to apply noise to the target image. In this way, the DDPM can be trained to "reverse" the step-by-step process of applying noise to the image in order to remove the noise from the image.

[0015] In each diffusion step, the DDPM can be trained to minimize a loss function, which corresponds to the difference between the predicted output image (predicted output) for a given diffusion step and the noise of the training image. Therefore, the DDPM can be trained to accurately predict and model the noise difference between each input image and output image at each diffusion step. In one embodiment, the training of the DDPM can include setting one or more weights of the model. One or more weights of the model can vary for each diffusion step within a series of diffusion steps or for at least one diffusion step within that series of steps. In one embodiment, a conditional image can be input into the DDPM during the learning process to guide the generation of the output image. In one embodiment, the conditional image can be a target image. The target image used for the training of the DDPM can be at least one target image and can also include a plurality of target images. For example, at least one target image can include a low-resolution medical image (e.g., a CT image) and an edge-detected medical image (e.g., a CT image) or a medical image processed by other methods. Similarly, the conditional image used for the training of the DDPM can be at least one conditional image and can also include a plurality of conditional images. In one example, at least one conditional image can include a low-resolution medical image (e.g., a CT image) and an edge-detected medical image (e.g., a CT image), or a medical image processed by other methods. In one example, at least one conditional image or at least one target image can include three consecutive conditional images for a multi-dimensional (e.g., 2.5D) process.

[0016] In one embodiment, the input to the learned DDPM can be a noisy image, a conditional image, and a diffusion step (also referred to as a time step or a sampling step). The DDPM can predict a second probability distribution corresponding to the restored image, given the conditional image as a known condition. In one embodiment, the DDPM can denoise a pure noise image in a series of diffusion steps to generate the final restored image. The pure noise image can be the first input image to the DDPM. The DDPM can output a denoised image (also referred to herein as a restored image) for each diffusion step. The denoised images output from each diffusion step within the series of diffusion steps can be input to the next diffusion step to iteratively denoise the pure noise image. The DDPM can include one or more learned weights used to output the restored image, and the values of the one or more learned weights can depend on the diffusion step. Further details regarding the learning and use of the DDPM are provided in Ho, J. et al., (2020). “Denoising diffusion probabilistic models.” Advances in neural information processing systems, 33, 6840 - 6851, and in Xia, W. et al., (2022). “Low-Dose CT Using Denoising Diffusion Probabilistic Model for 20x times Speedup.” arXiv preprint arXiv:2209.15136, each of which is hereby incorporated by reference in its entirety for all purposes.

[0017] In one embodiment, the DDPM can be used for denoising a series of images. The series of images can be sequentially acquired over a temporal or spatial dimension. For example, to evaluate the function of the kidneys, a series of images can be acquired over time by a kidney scan. In the denoising process (also called the inference process or sampling process), the DDPM can denoise in the initial denoising (diffusion) step and identify larger or more generalized features. These larger features are usually consistent across a series of images. For example, the general shape and position of the kidneys, as well as the structures within them, can be first identified in the scan images and are less likely to vary within a single kidney scan. Next, in later diffusion steps, the DDPM can denoise and identify smaller features or details. In the example of a kidney scan, the DDPM can identify finer details of the shape and size of the kidney structures, as well as the position of the contrast agent within the kidneys. These details may vary across a series of images as the kidney system processes the body's fluids. The changes across a series of images are likely to be gradual and continuous over time. Therefore, adjacent images within a series of images may be similar to each other.

[0018] In another example, a series of images can be acquired by scanning one or more cross-sections of the body along one or more directions. In a similar manner, the DDPM can first denoise and identify larger features such as the general shape of the body cross-section and the organs within it. The DDPM can then denoise and identify smaller features and / or finer details of the organs. Two adjacent images within a series of images also depict a portion of the body that is close to each other. Therefore, as the scan progresses along the body, adjacent images within the series of images are also likely to be similar to each other and share larger features.

[0019] Figure 1 shows a series of N images (Z0, Z1, Z2, Z3, Z4, Z5, ···, Z N) is a schematic diagram of a process for noise removal. A series of N images can be collected over a period of time or along a spatial direction. For each image in the series of images, a pure noise image x T can be input into the DDPM. The pure noise image x T can be generated using a probability distribution model such as a Gaussian distribution. A conditional image can also be input into the DDPM for noise removal processing. In one embodiment, the conditional image can be an image to be denoised (one of a series of images Z0, Z1, Z2, Z3, Z4, Z5, ···, Z N ). The conditional image can include a plurality of conditional images such as an edge detection image or other processed images.

[0020] The DDPM can be used to remove noise from each image (Z0, Z1, Z2, Z3, Z4, Z5, ···, Z N ) in a series of T diffusion steps. The DDPM removes noise from the pure noise image x T in a series of T diffusion steps, and for each image in the series, a sequence of restored images x T-1 , x T-2 , etc. can be generated. At each diffusion step, the image restored from the previous diffusion step can be input into the DDPM together with the conditional image and the time step (T-1, T-2, etc.). For example, the DDPM can remove noise from the pure noise image x T based on the conditional image Z0 and generate the restored image x T-1 at time step T-1. The DDPM can then remove noise from the image x T-1 based on the conditional image Z0 and generate a further restored image x T-2 at time step T-2. At time step T, the DDPM can output the image x0, which is the denoised version of the conditional image Z0. For each image (Z0, Z1, Z2, Z3, Z4, Z5, ···, Z NAfter T time steps for , DDPM can output the corresponding denoised images (x0, x1, x2, x3, x4, x5, ···, x N ). In this way, DDPM performs N * T diffusion steps and denoises all images in a series of images. Here, "*" is an operator indicating multiplication.

[0021] In one embodiment, the present disclosure relates to a bulk diffusion method for image noise removal that utilizes the similarity between adjacent images in a series of images to achieve faster noise removal of the series of images using a diffusion-based probabilistic model. The method described herein can reduce the number of diffusion steps required to denoise a series of images while maintaining the inference accuracy. By reducing the number of diffusion steps, it is possible to achieve not only faster noise removal but also a reduction in the computational power usage.

[0022] In one embodiment, the method can include grouping the images in a series of images and performing initial batch noise removal on the image groups (groups of images) using a learned DDPM. The image groups can be subsets of adjacent images in a series of images collected over a certain period or in the spatial direction. The series of images can be divided into one or more groups, and each group can include one or more images. Each of the one or more image groups can have the same or different numbers of images. For example, the first image group G0 can include the first n images in a series of images, the second image group G1 can include the subsequent n + 1 to n + m images in a series of images, and so on. The number of images in a group may sometimes be referred to as the thickness of the group in this specification. In one embodiment, the thickness of the group can be adjusted based on the type of image acquisition or the subject imaged in the series of images. For example, in each image of the group, the number of images in the group can be set such that one or more features or types of features (e.g., features of a certain size) are constant.

[0023] In one embodiment, the method can include determining a representative image of a group of n images. In one example, the representative image can be generated by calculating the average value of each pixel across the n images. In one embodiment, the representative image can be an image of the image group (e.g., the first image, the nth image, the n / 2th image). Note that " / " is an operator indicating division. In one embodiment, the representative image can be a pre-processed image, such as an image processed and weighted in the frequency domain. The representative image can be generated by any combination of image calculations and processing and is not limited to the examples provided herein. The representative image can be input into the DDPM as a conditional image at each diffusion step of a series of initial diffusion steps. The number of initial diffusion steps can be represented by the quantity T1. The DDPM can use the representative image as a conditional image to denoise a pure noise image over T1 initial diffusion steps. The features identified by the DDPM in the initial T1 diffusion steps are likely to be consistent across each image within the image group. Therefore, denoising conditioned on the representative image rather than each image within the group is sufficient for any image within the group at the initial diffusion steps. The number of initial diffusion steps in which the representative image is used as a conditional image can be adjusted based on factors such as the total number of diffusion steps (T), the type of scan, the features predicted in a series of images, etc. For example, the number of initial diffusion steps can be set such that the DDPM can identify the features present in each image within the image group during the initial diffusion steps.

[0024] The DDPM can output a representative restored image after the last step (the T1-th step) of a series of initial diffusion steps. The representative restored image can be generated by the DDPM from a pure noise image using the representative image of the image group as a conditional image. In one embodiment, the representative restored image can include one or more features shared across each image within the image group. After the T1 initial diffusion steps, the appearance (look) of each image within the image group may diverge. For example, the appearance of finer details and smaller features may vary across each image within the image group. These features may not be distinguishable by the DDPM until the T1 initial diffusion steps are completed. The T1 initial diffusion steps can result in a denoised, visualized representative restored image of the larger, less detailed features common to the series of images. Thus, the T1 initial diffusion steps can be shared across each image within the image group and need not be repeated by the DDPM for each image.

[0025] In one embodiment, the representative restored image generated from the initial diffusion step of T1 times can be used as the input image for further diffusion steps conditioned on each image within the image group. DDPM can use each image of the group of n images as a conditional image to denoise the representative restored image and generate a restored image for each image within the group of n images. The denoising of the representative restored image conditioned on each image can be performed over T2 diffusion steps. Thereafter, DDPM can output n final images corresponding to the n images within the group. In this way, DDPM uses the representative image as a conditional image, executes a series of T1 initial diffusion steps once, and generates a restored representative image, thereby reducing the number of diffusion steps required to denoise each image within the group. Thereafter, DDPM can denoise the restored representative image with a series of T2 diffusion steps for n times and generate n output images. By this method, it is no longer necessary to repeat the initial diffusion step of T1 times for each of the n images within the image group. The remaining T2 diffusion steps can be used to identify and denoise the features specific to each image within the image group or the distinct features within the image.

[0026] Figure 2 is a schematic diagram of the bulk diffusion method according to an embodiment of the present disclosure. The learned DDPM can be used to denoise a series of N images (Z0, Z1, Z2, Z3, Z4, Z5, ···, Z N ). The series of N images can be collected over a period of time or along a spatial direction. In one embodiment, the series of N images can be grouped into one or more groups (G0, G1, ···, G M ), and each group includes a subset of the N images. For example, as shown in Figure 2, group G0 can include images Z0 to Z5, and group G1 can include images Z6 to Z 11 and so on.

[0027]

Number

[0028]

Number

[0029] That is, the total number of diffusion steps is T(t + T1). The t subsequent diffusion steps may be referred to herein as the quantity T2. In this way, DDPM can denoise each image in a group of N images using T1+(T2*N) diffusion steps. Here, T1 + T2 = T. By this method, the number of diffusion steps required is reduced compared to the method shown in FIG. 1 that uses N*T steps to diffuse a series of N images.

[0030] The process of generating a representative restored image for each group of images and denoising the representative restored image conditioned on each image within the image group can be repeated for each group in a series of images. Each group can include the same or different numbers of images. The number T1 of initial diffusion steps and the number T2 of subsequent diffusion steps may be different for each group or the same for each group. In one embodiment, the number T1 of initial diffusion steps can be referred to as the length of the initial diffusion process using the representative image. The length of the initial diffusion process can be adjusted based on the predicted content of the images within the group. For example, the length of the initial diffusion process can be adjusted based on the predicted size of one or more features. In one embodiment, the length of the initial diffusion process and / or the total number of diffusion steps (T) may depend on the learning of DDPM.

[0031] In one embodiment, the image can be downsampled before the diffusion step. For example, the representative image of the image group can be downsampled before the initial diffusion step so that the downsampled representative image is smaller in size than the original images of the group. By downsampling the representative image, the processing time or capacity required to remove noise from the representative image in the initial diffusion step can be reduced. The representative image can be downsampled without affecting the appearance of the features identified (noise removed) in the initial diffusion step.

[0032]

Number

[0033] FIG. 3 is an explanatory diagram of parameters that can be adjusted by the bulk diffusion method. Each of the parameters can be set individually. The parameters may be shared among image groups or may be different for one or more image groups. In one embodiment, the thickness of the group is shown as the number of images in the group. The representative image of the group can be downscaled before the first T1 initial diffusion steps and then upscaled before the remaining T2 diffusion steps. The number T1 of initial diffusion steps, or the length of the initial diffusion process, can also be adjusted. The length may be adjusted as a part of the total number (T) of diffusion steps or may be an absolute number of steps.

[0034] FIG. 4 shows a process for implementing the bulk diffusion method described herein according to one embodiment of the present disclosure. The process of FIG. 4 can be implemented by a computer via computer-readable instructions. In one embodiment, a series of images can be grouped into one or more groups G0, G1, ···, G M For each group, the DDPM can generate a series of restored representative images at the initial diffusion step (T1, or from T to t) using the following equation (1).

[0035]

Number

[0036]

Number

[0037]

Number

[0038] The subsequent diffusion steps are performed using the same model and equation (1) as for generating the output image in the initial diffusion step. However, in the subsequent diffusion steps, instead of the representative image, an image from the group of images is used as the conditional image. After the subsequent diffusion steps of t, DDPM can output the final restored image x n for each input image Z0, Z1, ···, Z n in the group. This process can be repeated for each group of images.

[0039] Figures 5A to 5C show the results of the bulk diffusion method described herein according to an embodiment of the present disclosure. Figure 5A is an explanatory diagram of a target image, for example, a medical image of the brain. In this example, the input images to the DDPM model are a series of 12 medical images. The DDPM can denoise each of the series of 12 medical images. Figure 5B is an explanatory diagram of the denoised medical images from a series of denoised medical images generated by the DDPM using the method described with reference to Figure 1. The series of 12 medical images are denoised by the DDPM using 200 diffusion steps for each image, resulting in a total of 2400 diffusion steps. Figure 5C is an explanatory diagram of the denoised images from a series of denoised medical images generated by the DDPM using the bulk diffusion method described with reference to Figure 2. The series of medical images can be denoised by the DDPM of the bulk diffusion method, where T1 = 190 and T2 = 10. Thus, a series of 12 medical images can be denoised using 190+(10*12)=310 diffusion steps. The denoised images in Figures 5B and 5C have the same quality and recovery level.

[0040] The systems and methods described herein are compatible with other methods for reducing the processing time of diffusion-based image denoising models. Such methods include, but are not limited to, using a denoising diffusion implicit model, reducing the number of diffusion steps T, implementing early stopping of the diffusion process, using a fast ordinary differential equation solver, pre-segmentation diffusion sampling, and using a high-frequency spatial diffusion model.

[0041] Next, the hardware of device 601 according to an exemplary embodiment will be described with reference to FIG. 6. In FIG. 6, device 601 includes a processing circuit. Device 601 can be used to execute any of the methods described herein related to obtaining a DDPM, learning a DDPM, receiving an acquired image, and / or denoising an image using a DDPM. In one embodiment, device 601 can be a server, a computer, etc. In one embodiment, device 601 can communicate with an image acquisition device such as the CT device illustrated in FIG. 7 or can be incorporated into the image acquisition device. In one embodiment, the methods described herein can be distributed across one or more devices, and the one or more devices include at least a portion of the elements of device 601. The processing circuit includes one or more of the elements described next with reference to FIG. 6. Process data and instructions can be stored in memory 602. These processes and instructions may also be stored on a storage medium disk 604 such as a hard drive disk (HDD) or a portable storage medium, or may be stored remotely. Further, the processes and instructions are not limited by the form of the computer-readable medium in which they are stored. For example, the instructions may be stored on a CD, DVD, flash memory, RAM, ROM, PROM, EPROM, EEPROM, hard disk, or another information processing device such as a server or a computer with which device 601 communicates.

[0042] Furthermore, the claimed processes may be provided by a utility application, a background daemon, or a component of an operating system, or a combination thereof, and executed in cooperation with a CPU 600 and an operating system such as Microsoft Windows®, UNIX®, Solaris, LINUX®, Apple MAC-OS, and other systems known to those skilled in the art.

[0043] The hardware elements for realizing device 601 can be realized by various circuit elements known to those skilled in the art. For example, CPU 600 may be an Intel Xeon or Core processor of Intel Corporation in the United States, or an Opteron processor of AMD Corporation in the United States, or other processor types recognized by those skilled in the art. Alternatively, CPU 600 may be implemented using an FPGA, ASIC, PLD, or discrete logic circuit as recognized by those skilled in the art. Further, CPU 600 may be implemented as a plurality of processors that cooperate in parallel to execute the instructions of the above-described process.

[0044] The device 601 in FIG. 6 also includes a network controller 606, such as an Intel Ethernet (registered trademark) PRO network interface card of Intel Corporation in the United States, for interfacing with the network 650 and communicating with other devices. As can be understood, the network 650 can be a public network such as the Internet, a private network such as a LAN or WAN network, or any combination thereof, and can also include a PSTN or ISDN subnetwork. The network 650 can be wired, such as an Ethernet network, or wireless, such as a cellular network including EDGE, 3G, 4G, and 5G wireless cellular systems. The wireless network can also be WiFi, Bluetooth (registered trademark), or any other known wireless communication format.

[0045] The device 601 further includes a display controller 608, such as an NVIDIA GeForce GTX or Quadro graphics adapter of NVIDIA Corporation in the United States, for interfacing with a display 610 such as an LCD monitor. The general-purpose I / O interface 612 interfaces with a keyboard and / or mouse 614, as well as a touch screen panel 616 on or separate from the display 610. The general-purpose I / O interface also connects to various peripheral devices 618 including printers and scanners.

[0046] The sound controller 620 is also provided in the device 601 and interfaces with the speaker / microphone 622 to provide sound and / or music.

[0047] The general-purpose storage controller 624 connects the storage medium disk 604 to a communication bus 626 which may be ISA, EISA, VESA, PCI, or the like, in order to interconnect all components of the device 601. Descriptions of the general characteristics and functions of the display 610, the keyboard and / or mouse 614, and the display controller 608, the storage controller 624, the network controller 606, the sound controller 620, and the general-purpose I / O interface 612 are omitted in this specification for the sake of brevity since these characteristics are known.

[0048] In one embodiment, the image processed using the bulk diffusion method described herein can be a CT image acquired by a CT device or scanner. FIG. 7 shows an implementation of a radiation gantry included in a CT device or CT scanner. As shown in FIG. 7, the radiation gantry 9900 is shown from the side and further includes an X-ray tube 9901, an annular frame 9902, and a plurality of rows or a two-dimensional array type X-ray detector 9903. The X-ray tube 9901 and the X-ray detector 9903 are mounted radially on the annular frame 9902 rotatably supported about the rotation axis RA, sandwiching a subject such as a patient. The rotation unit 9907 rotates the annular frame 9902 at a high speed, for example, 0.4 seconds / rotation, while the subject moves along the axis RA in the depth or front-back direction of the page shown in the figure.

[0049] Embodiments of an X-ray computed tomography (CT) apparatus according to the present disclosure will be described below with reference to the accompanying drawings. Note that the X-ray CT apparatus includes various types of apparatuses such as a rotation / rotation type apparatus in which, for example, an X-ray tube and an X-ray detector both rotate around a subject to be examined, and a fixed / rotation type apparatus in which many detector elements are arranged in an annular or planar shape and only the X-ray tube rotates around the subject to be examined. The present disclosure can be applied to either type. Here, the currently mainstream rotation / rotation type is exemplified. The X-ray CT apparatus is an example of a medical image diagnostic apparatus.

[0050] The multi-slice X-ray CT apparatus further includes a high-voltage generator 9909 that generates a tube voltage applied to the X-ray tube 9901 through a slip ring 9908 so that the X-ray tube 9901 generates X-rays. The X-ray detector 9903 is located on the opposite side from the X-ray tube 9901 across the subject to detect the irradiated X-rays that have propagated through the subject. The X-ray detector 9903 is, for example, a photon-counting type detector. The X-ray detector, or the photon-counting type detector 9903, further includes individual detector elements or units such as a processing circuit.

[0051] The CT apparatus further includes another device for processing the detection signal from the X-ray detector 9903. A data acquisition circuit or a data acquisition system (DAS) 9904 converts the signal output from the X-ray detector 9903 of each channel into a voltage signal, amplifies the signal, and further converts the signal into a digital signal. The X-ray detector 9903 and the DAS 9904 are configured to manage a predetermined total number of projections per rotation (TPPR).

[0052] The above data is sent via the contactless data transmitter 9905 to the preprocessing device 9906 housed in a console outside the X-ray imaging gantry 9900. The preprocessing device 9906 performs specific corrections. The memory 9912 stores the resulting data, also called projection data, at a stage just before the reconstruction process. The memory 9912 is connected to the system controller 9910 via the data / control bus 9911 together with the reconstruction device 9914, the input device 9915, and the display 9916. The system controller 9910 controls a current regulator 9913 that limits the current to a level sufficient to drive the CT system.

[0053] The detector is rotated and / or fixed with respect to a scanned subject such as a patient in various generations of CT scanner systems. In one embodiment, the above-described CT system can be an example in which a third-generation geometry system and a fourth-generation geometry system are combined. In the third-generation system, the X-ray tube 9901 and the X-ray detector 9903 are radially attached to the annular frame 9902, and the annular frame 9902 rotates around the rotation axis RA to rotate around the subject. In the fourth-generation geometry system, the detector is fixedly arranged around the patient, and the X-ray tube 9901 rotates around the patient. In an alternative embodiment, the X-ray imaging gantry 9900 has a number of detectors arranged on an annular frame 9902 supported by a C-arm and a stand.

[0054] The post-reconstruction process executed by the reconstruction device 9914 can optionally include image filtering and smoothing, volume rendering processing, and image difference processing. The image reconstruction process can implement various CT image reconstruction methods. The reconstruction device 9914 can use the memory to store, for example, projection data, reconstructed images, calibration data and parameters, and computer programs.

[0055] This specification includes many specific implementation details, but these should not be construed as limitations on the scope of what can be claimed, but rather as descriptions of features that may be specific to particular embodiments.

[0056] The specific features described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, the various features described in the context of a single embodiment can also be implemented separately in multiple embodiments, or in any suitable sub-combination. Furthermore, although features may be described above as acting in a particular combination and may even initially be claimed as such, in some instances, one or more features of the claimed combination can also be deleted from that combination. Also, the claimed combination can be directed to a sub-combination, or a variant of a sub-combination.

[0057] Similarly, operations are shown in the drawings in a particular order, but this should not be understood as requiring that such operations be performed in the particular order shown or in a sequential order, or that all of the operations shown be performed, in order to achieve a desirable result. In some situations, multitasking and parallel processing may be advantageous. Moreover, the separation of the various system modules and components in the embodiments described above should not be understood as being required in every embodiment, and it is understood that the described program components and systems may generally be integrated together into a single software product, or packaged into multiple software products.

[0058] Specific embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the operations recited in the claims can be performed in a different order and still achieve desirable results. As one example, the processes depicted in the accompanying figures do not necessarily require the particular order or sequential order shown to achieve desirable results. In some cases, multitasking and parallel processing may be advantageous.

[0059] Also, embodiments of the present disclosure may be as described in the following appendices.

[0060] (1) A noise removal method for removing noise from a plurality of input images, comprising: obtaining a diffusion-based probability model learned to perform noise removal over T (where T is an integer greater than or equal to 2) steps using at least one target image and at least one conditional image; grouping the plurality of input images into a plurality of groups; for each group of the plurality of groups, determining an initial representative image of the group; for each group, using the obtained probability model to perform a first sequence of T1 (where T1 is an integer greater than or equal to 1) noise removal sampling steps starting from the initial representative image of the group to generate a corresponding sequence of the representative image of the group; for each group of the plurality of groups, for each input image within the group, using the obtained probability model to perform a second sequence of T2 (where T2 is an integer greater than or equal to 1) noise removal sampling steps starting from the representative image denoised by the last noise removal sampling step of the first sequence to generate a final image that is a restored image corresponding to the input image, wherein T = T1 + T2.

[0061] (2) The noise removal method according to (1), wherein for a specific group among the plurality of groups, the initial representative image is the average of the plurality of input images of the specific group.

[0062] (3) For a specific group among the plurality of groups, the initial representative image is one of the input images of the specific group in the noise removal method described in (1) to (2).

[0063] (4) For each group, performing the first sequence includes, for each noise removal sampling step in the first sequence, inputting into the probability model, as inputs to the probability model, the initial representative image of the group, the one preceding the first sequence among the representative images of the group, and a value indicating the noise removal sampling step in the noise removal method described in (1) to (3).

[0064] (5) Grouping is the noise removal method described in (1) to (4), which groups a plurality of input images into a plurality of groups such that all groups do not necessarily have the same size.

[0065] (6) The noise removal method described in (1) to (5) further includes obtaining a plurality of input images as a time sequence of a plurality of reconstructed medical images.

[0066] (7) The noise removal method described in (1) to (6) further includes obtaining a plurality of input images as a spatial sequence of a plurality of reconstructed medical images.

[0067] (8) The noise removal method described in (1) to (7) further includes downsampling the initial representative image of each group.

[0068] (9) The method described in (1) to (8) further includes upsampling the representative image denoised by the last noise removal sampling step of the first sequence for each group of the plurality of groups.

[0069] (10) A noise removal program for causing a computer to remove noise from a plurality of input images, the program including: a process of obtaining a diffusion-based probability model learned to perform noise removal over T (where T is an integer of 2 or more) steps using at least one target image and at least one conditional image; a process of grouping the plurality of input images into a plurality of groups; for each group of the plurality of groups, a process of determining an initial representative image of the group; for each group, a process of performing a first sequence of T1 (where T1 is an integer of 1 or more) noise removal sampling steps starting from the initial representative image of the group using the obtained probability model to generate a corresponding sequence of the representative image of the group; for each group of the plurality of groups, for each input image in the group, a process of performing a second sequence of T2 (where T2 is an integer of 1 or more) noise removal sampling steps starting from the representative image noise-removed by the last noise removal sampling step of the first sequence using the obtained probability model to generate a final image that is a restored image corresponding to the input image, where T = T1 + T2.

[0070] (11) The noise removal program according to (10), wherein for a specific group among the plurality of groups, the initial representative image is the average of the plurality of input images of the specific group.

[0071] (12) The noise removal program according to (10) to (11), wherein for a specific group among the plurality of groups, the initial representative image is one of the input images of the specific group.

[0072] (13) The noise removal program according to (10) to (12), wherein for each group, the process of performing the first sequence includes, for each noise removal sampling step in the first sequence, a process of inputting, as inputs to the probability model, the initial representative image of the group, the one preceding the first sequence of the representative images of the group, and a value indicating the noise removal sampling step to the probability model.

[0073] (14) The grouping process is the noise removal program described in (10) to (13) that groups a plurality of input images into a plurality of groups such that all of the groups do not necessarily have the same size.

[0074] (15) The method further includes obtaining a plurality of input images as a time sequence of a plurality of reconstructed medical images, the noise removal program described in (10) to (14).

[0075] (16) The method further includes obtaining a plurality of input images as a spatial sequence of a plurality of reconstructed medical images, the noise removal program described in (10) to (15).

[0076] (17) The noise removal program described in (10) to (16) further includes downsampling an initial representative image of each group among the plurality of groups.

[0077] (18) For each group of the plurality of groups, the noise removal program described in (10) to (17) further includes upsampling a representative image that has been noise-removed by a last noise removal sampling step of a first sequence.

[0078] (19) A medical image diagnostic apparatus for removing noise from a plurality of input images, which acquires a diffusion-based probability model learned to perform noise removal over T (where T is an integer of 2 or more) steps using at least one target image and at least one conditional image, groups the plurality of input images into a plurality of groups, determines an initial representative image for each group of the plurality of groups, and for each group, uses the acquired probability model to perform a first sequence of T1 (where T1 is an integer of 1 or more) noise removal sampling steps starting from the initial representative image of the group to generate a corresponding sequence of representative images of the group, and for each group of the plurality of groups, for each input image within the group, uses the acquired probability model to perform a second sequence of T2 (where T2 is an integer of 1 or more) noise removal sampling steps starting from the representative image noise-removed by the last noise removal sampling step of the first sequence to generate a final image that is a restored image corresponding to the input image, and T = T1 + T2, the medical image diagnostic apparatus.

[0079] (20) For a specific group among the plurality of groups, the initial representative image is the average of the plurality of input images of the specific group, or an input image of the specific group, the medical image diagnostic apparatus according to (19).

[0080] Obviously, in light of the above teachings, numerous modifications and variations are possible. Therefore, it should be understood that within the scope of the claims, the present invention can be implemented in ways different from the methods specifically described herein.

[0081] According to at least one of the embodiments or variations described above, faster noise removal can be performed.

[0082] Although some embodiments have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, as well as in the invention described in the claims and the scope of equivalents thereof.

Description of Reference Numerals

[0083] 600 CPU 601 Device

Claims

1. A noise removal method for removing noise from a plurality of input images, comprising: obtaining a diffusion-based probability model learned to perform noise removal over T (where T is an integer greater than or equal to 2) steps using at least one target image and at least one conditional image; grouping the plurality of input images into a plurality of groups; for each group of the plurality of groups, determining an initial representative image of the group; for each group, performing a first sequence of T1 (where T1 is an integer greater than or equal to 1) noise removal sampling steps starting from the initial representative image of the group using the obtained probability model to generate a corresponding sequence of the representative image of the group; for each group of the plurality of groups, for each input image within the group, performing a second sequence of T2 (where T2 is an integer greater than or equal to 1) noise removal sampling steps starting from the representative image denoised by the last noise removal sampling step of the first sequence using the obtained probability model to generate a final image that is a restored image corresponding to the input image; wherein, T = T1 + T2, the noise removal method.

2. The noise removal method according to claim 1, wherein for a specific group of the plurality of groups, the initial representative image is an average of the plurality of input images of the specific group.

3. The noise removal method according to claim 1, wherein for a specific group of the plurality of groups, the initial representative image is one of the plurality of input images of the specific group.

4. For each group, performing the first sequence includes, for each noise removal sampling step in the first sequence, inputting into the probability model the initial representative image of the group, the one preceding in the first sequence of the representative images of the group, and a value indicating the noise removal sampling step. The noise removal method according to claim 1.

5. The noise removal method according to claim 1, wherein the grouping groups the plurality of input images into the plurality of groups such that all of the groups do not necessarily have the same size.

6. The noise removal method according to claim 1, further comprising obtaining the plurality of input images as a temporal sequence of a plurality of reconstructed medical images.

7. The noise removal method according to claim 1, further comprising obtaining the plurality of input images as a spatial sequence of a plurality of reconstructed medical images.

8. The noise removal method according to claim 1, further comprising downsampling the initial representative images of each group.

9. The noise removal method according to claim 8, further comprising, for each group of the plurality of groups, upsampling the representative image denoised by the last noise removal sampling step of the first sequence.

10. A noise removal program for causing a computer to remove noise from a plurality of input images, a process of obtaining a diffusion-based probability model learned to perform noise removal over T (T is an integer of 2 or more) steps using at least one target image and at least one conditional image; a process of grouping the plurality of input images into a plurality of groups; a process of determining an initial representative image of each group for each of the plurality of groups; a process of performing, for each group, a first sequence of T1 (T1 is an integer of 1 or more) noise removal sampling steps starting from the initial representative image of the group using the obtained probability model to generate a corresponding sequence of the representative image of the group; for each group of the plurality of groups, for each input image within the group, a process of performing, for generating a final image that is a restored image corresponding to the input image, a second sequence of T2 (T2 is an integer of 1 or more) noise removal sampling steps starting from the representative image denoised by the last noise removal sampling step of the first sequence using the obtained probability model; causing the computer to execute, a noise removal program, where T = T1 + T2.

11. A medical image diagnostic apparatus for removing noise from a plurality of input images, obtaining a diffusion-based probability model learned to perform noise removal over T (T is an integer of 2 or more) steps using at least one target image and at least one conditional image, group the plurality of input images into a plurality of groups, for each group of the plurality of groups, determine an initial representative image of the group, for each group, use the obtained probability model to execute a first sequence of T1 (T1 is an integer greater than or equal to 1) noise removal sampling steps starting from the initial representative image of the group to generate a corresponding sequence of representative images of the group, for each group of the plurality of groups, for each input image within the group, a processing circuit that uses the obtained probability model to execute a second sequence of T2 (T2 is an integer greater than or equal to 1) noise removal sampling steps starting from the representative image denoised by the last noise removal sampling step of the first sequence to generate a final image that is a restored image corresponding to the input image comprising, a medical image diagnostic apparatus, where T = T1 + T2.

Images