Imaging apparatus, time-resolved medical imaging method, storage medium and program product

Abstract

Embodiments of the present disclosure relate to an imaging apparatus, a time resolved medical imaging method, a storage medium and a program product. The present disclosure relates to an imaging apparatus comprising: a data acquisition module configured to capture and receive a plurality of time ordered data sets used in an image generation process of an entity in at least one predetermined time period, wherein the plurality of time ordered data sets comprises spatial domain information and temporal domain information; at least one denoising module coupled to the data acquisition module and comprising a settable / adjustable bilateral filter, wherein the at least one denoising module is configured to set / adjust filter parameters of the settable / adjustable bilateral filter using the received spatial domain information and temporal domain information, to denoise the plurality of time ordered data sets using the settable / adjustable bilateral filter, and to obtain a plurality of denoised time ordered data sets.

Description

Technical Field

[0001] This disclosure relates to an imaging apparatus. It also relates to methods, storage media, and computer program products used in the imaging apparatus for time-resolved medical imaging. Background Technology

[0002] The clinical applications of medical imaging are crucial in human medicine and veterinary medicine. From two-dimensional (2D) medical imaging to three-dimensional (3D) medical imaging, where 2D medical imaging provides a planar, linear view limited by length and width, and 3D medical imaging adds a third dimension of depth, efforts to increase the dimensions of medical images to provide a more comprehensive presentation of medical imaging have never ceased.

[0003] In recent years, time-resolved medical imaging has provided a more advanced form of imaging. Time-resolved medical imaging is a medical imaging technique that treats time as an additional dimension and aims to represent the motion of a region of interest over time. With time-resolved medical imaging, the motion of the region of interest can be observed more effectively.

[0004] Furthermore, when ionizing radiation is involved in medical imaging, the higher the radiation dose contributing to the image, the more pronounced the image noise and the more easily low-contrast structures are perceived. However, the potential harm to a human or animal is proportional to the absorbed dose of ionizing radiation to the body's organs. Because time-resolved medical imaging typically involves capturing multiple static medical images at predetermined time points, the number of repeated scans a patient or animal can receive to continuously adapt to a radiotherapy treatment plan may be limited.

[0005] Therefore, experts have been working to improve imaging devices in time-resolved medical imaging to limit the radiation dose required to acquire high-quality medical images and to provide improved views of motion results in the region of interest.

[0006] Previous solutions for denoising temporally resolved imaging typically involved applying 2D or 3D spatial imaging denoising methods to temporally resolved medical imaging, where the temporal information of the temporally resolved medical imaging was not involved in the image denoising process. Meanwhile, bilateral filters are frequently used in 2D or 3D spatial imaging denoising methods to filter out image noise because they are successful in smoothing images while preserving edges.

[0007] In scientific publications, AMMendrik et al.'s "TIPS Bilateral Denoising for High-Quality Cerebral Blood Flow Maps in 4D CT Perfusion Scans," Phys. Medicine. Biology. 56, 3857–3872 (2011), explored the use of four-dimensional (4D) bilateral filters for denoising medical images of brain scans, which included measurements of voxel similarity in the temporal-intensity distribution in addition to voxel values ​​in the 3D spatial domain. However, fine-tuning the 4D bilateral filter to specific noise characteristics can be challenging. Furthermore, another scientific publication, Wagner, Fabian et al.'s "Ultra-Low Parameter Denoising: Trainable Bilateral Filter Layers in Computed Tomography," Medical Physics 49.8, 5107–5120 (2022), demonstrated that bilateral filters achieve strong denoising performance on 3D spatial image data while reducing computational load. However, in this paper, temporally resolved image data was not used to train the bilateral filter. Most image generation or denoising algorithms are built on deep artificial neural networks with multiple layers and include hundreds of thousands of trainable parameters, meaning that training denoising algorithms requires intensive computation. While training with deep artificial neural networks facilitates the extraction of complex features from data, it is impossible to fully understand the data processing and may produce illusions.

[0008] In view of this background, the problem addressed by this disclosure is to provide an imaging apparatus that implements a novel method for time-resolved medical imaging. Summary of the Invention

[0009] According to this disclosure, the problem is solved by an imaging apparatus having the features of claim 1 and / or by a method for time-resolved medical imaging having the features of claim 10. Furthermore, a computer program product having the features of claim 14 and a non-transitory computer-readable storage medium having the features of claim 15 are provided.

[0010] Based on this, the following is provided: An imaging apparatus comprising: a data acquisition module configured to capture and receive multiple time-ordered datasets used in an image generation process for an entity over at least a predetermined time period, wherein the multiple time-ordered datasets include spatial domain information and temporal domain information; at least one denoising module coupled to the data acquisition module and including a settable / adjustable bilateral filter, wherein the at least one denoising module is configured to set / adjust filter parameters of the settable / adjustable bilateral filter using the received spatial domain information and temporal domain information, thereby denoising the multiple time-ordered datasets using the settable / adjustable bilateral filter and acquiring multiple denoised time-ordered datasets; and an output module configured to output at least one time-resolved medical image for the entity using the multiple denoised time-ordered datasets. - A method for time-resolved medical imaging, particularly by employing an imaging apparatus according to the present disclosure. The present disclosure relates to an imaging apparatus, the method comprising the steps of: capturing and receiving multiple time-ordered datasets used in the image generation process of an entity within at least one predetermined time period, wherein the multiple time-ordered datasets include spatial domain information and temporal domain information; using the received spatial domain information and temporal domain information to set / adjust filter parameters of a settable / adjustable bilateral filter, thereby denoising the multiple time-ordered datasets using the settable / adjustable bilateral filter, and acquiring multiple denoised time-ordered datasets; and using the multiple denoised time-ordered datasets to output at least one time-resolved medical image of the entity. - A computer program product comprising a computer program that, when the program portion of the computer program is executed by a computer, causes an imaging device to perform the method according to the present disclosure. - A non-transitory computer-readable storage medium on which a computer stores a computer program according to the present disclosure. Attached Figure Description

[0011] The present disclosure is described in more detail below with reference to the embodiments shown in the accompanying drawings, wherein:

[0012] Figure 1 A block diagram illustrating an example of an imaging device is shown;

[0013] Figure 2 A block diagram of another example of an imaging device is shown;

[0014] Figure 3 A graph showing an example of multiple time-sorted datasets captured within a time period;

[0015] Figure 4 A schematic diagram illustrating an example of the filtering process for an exemplary image using a configurable / adjustable bilateral filter;

[0016] Figure 5 A diagram illustrating an example of a method for time-resolved medical imaging is shown.

[0017] The accompanying drawings are intended to provide a further understanding of embodiments of the present disclosure. They illustrate embodiments and, together with the description, help to explain the principles and concepts of the present disclosure. Other embodiments and the many advantages mentioned become clear with reference to the drawings. Elements in the drawings are not necessarily shown to scale.

[0018] In the accompanying drawings, the same, functionally equivalent, and identical operating elements, features, and components have the same reference numerals in each case, unless otherwise stated. Detailed Implementation

[0019] The method proposed in this disclosure focuses on a novel approach to denoise medical images by employing a configurable / adjustable bilateral filter, utilizing both spatial and temporal information from the data used during image generation. The configurable / adjustable bilateral filter is one whose filter parameters and configuration can be automatically set or adjusted, and is therefore optimizable. By employing this novel method, the amount of deposited radiation dose required to acquire medical images is reduced when capturing data used during image generation involves ionizing radiation or long acquisition times, without compromising image quality such as contrast, temporal resolution, and geometric accuracy. The current application aims to combine the benefits of the configurable / adjustable bilateral filter with the use of temporal information from the data used during image generation to achieve significantly improved medical images.

[0020] This disclosure is based on the idea of ​​providing an imaging apparatus for implementing a novel method for time-resolved medical imaging. This method utilizes spatial and temporal domain information from the data used in the image generation process to denoise the medical image using a configurable / adjustable bilateral filter. In this disclosure, for clarity, the processing from the input of data used in the image generation process for the entity to the output of a time-resolved image can be conceived as a processing pipeline having inputs, outputs, and channels between the inputs and outputs.

[0021] The input to the processing pipeline consists of multiple temporally ordered datasets used in the image generation process for entities captured and received within at least one predetermined time period. Therefore, these temporally ordered datasets include both spatial and temporal information. Entities can refer to regions of interest of patients, sick animals, healthy humans, or animals, providing comparative data for research or other purposes through medical imaging. Entities can also refer to non-living objects, such as imaging phantoms designed to simulate human or animal anatomy or tissue characteristics.

[0022] Processing multiple time-series datasets involves one or more configurable / adjustable bilateral filters in a processing pipeline. The configurable / adjustable bilateral filter is set / adjusted by the imaging apparatus's denoising module using spatial and temporal information from multiple time-series datasets. During the setup / adjustment of the configurable / adjustable bilateral filter, spatial and temporal information from multiple time-series datasets is used. Spatial information typically includes two or three spatial parameters and an intensity range parameter for the configurable / adjustable bilateral filter. Temporal information typically involves the time-intensity distribution of the parameters involved in these parameters. In time-resolved medical imaging, each scan generates images of regions of interest including the same anatomical structures at different predetermined time points. These different predetermined time points can be different time points within the same time period or different time points within different time periods. The time period can be a time cycle. These scans may be slightly distorted due to patient movement. If a pixel or voxel value differs significantly from surrounding pixel or voxel values ​​recorded from the same predetermined time point at different time cycles, it can be assumed to be noise. When these pixel or voxel values ​​differ significantly from those of surrounding pixels or voxels from scans that include similar anatomical structures but were re-encoded at different predetermined time points within a time period or time cycle, the pixel or voxel value can also be assumed to be noise. Therefore, the more scans performed on the region of interest, the easier it is to filter noise.

[0023] The output of the processing pipeline is a time-resolved medical image. The output module uses multiple denoised, time-ordered datasets to output at least one time-resolved medical image for the entity.

[0024] An imaging device employing a novel approach for time-resolved medical imaging improves the signal-to-noise ratio (SNR) of reconstructed medical images, better representing damage in low-contrast organs. It allows for dose reduction in time-resolved medical imaging when ionizing radiation is involved in the imaging process. It also allows for reduced acquisition time when extended acquisition times are involved in the imaging process. Furthermore, it avoids tube cooling conflicts in time-resolved tomography acquisition.

[0025] Advantageous configurations and developments become apparent from the description in the further dependent claims and with reference to the accompanying drawings.

[0026] In a possible embodiment of the imaging apparatus, the data acquisition module is configured to capture and receive multiple time-ordered datasets comprising raw detector data described by multiple sinograms. In some medical imaging processes, sinograms are used as 2D transition images between raw detector data and the final medical image. The proposed imaging apparatus allows for direct filtering of the sinograms. Taking a region of interest (ROI) imaged by a computed tomography (CT) scanner as an example, the sinogram is generated by converting raw X-ray projection data acquired from various angles into a 2D graphical representation. X-ray projection refers to the process by which an X-ray tube emits an X-ray beam at a given gantry angle, which then passes through the ROI and is intercepted by an imaging detector. The detector converts the intercepted X-ray beam signal into an electronic signal. The trajectory of the X-ray projection corresponding to the ROI follows a sine wave, where the X-ray angle and detector position are considered as axes of the sinogram. Furthermore, taking a region of interest imaged using a ring positron emission tomography (PET) scanner as an example, the ROI is injected with a PET reagent that positions the ROI and decays with the emission of positrons. The positrons are then annihilated by electrons, leading to the detection of a coincidence event, where their mass is converted into energy in the form of two photons emitted back-to-back. Annihilation occurs along the line connecting the two detectors, known as the response line (LOR). To collect raw detector data, the LOR is characterized by the angle of its orientation and the shortest distance between the LOR and the center of the PET stage. If a large number of LORs are plotted from the same region of interest, the resulting graph is represented by a sine wave. This graph visualizes the raw detector data and is called a sine plot. The proposed imaging device allows filtering functions to be directly built onto the sine plot, facilitating subsequent data processing steps with less noise.

[0027] In a possible embodiment of the imaging apparatus, the data acquisition module is configured to capture and receive multiple time-ordered datasets comprising reconstructed image data represented by multiple reconstructed images. The proposed imaging apparatus also allows for filtering of the reconstructed medical images. The process of creating an image using raw detector data is called image reconstruction. The reconstructed image data can come from data encoded in a sine wave. In CT imaging, once a medical image has been reconstructed from raw detector data, each pixel or voxel is assigned a specific value to express CT density in a normalized manner. Reconstructed image data can also come from data encoded in other forms of transition images. For example, magnetic resonance imaging (MRI) is an imaging process that collects raw detector data and converts it into transition data represented in k-space. K-space is a domain representing the spatial frequency components and phase information of an imaging of a region of interest. Each point in k-space corresponds to a specific combination of spatial frequency and phase information. MRI reconstructs a medical image from an image derived from k-space data. The proposed imaging apparatus allows for filtering of the reconstructed medical images, further increasing the flexibility of the imaging apparatus application. Furthermore, the proposed imaging apparatus can allow for simultaneous filtering of both the sine wave and the reconstructed medical images.

[0028] In a possible embodiment of the imaging device, the data acquisition module is configured such that the spatial domain information includes 3D spatial domain information. 3D spatial domain information can be acquired from an image representing the region of interest in the dimensions of length, width, and depth. For example, 3D spatial domain information can be acquired from CT scans, characterized by tomographic imaging, where cross-sectional images from different layers of the region of interest are imaged and aggregated into a 3D image. 3D spatial domain information can be acquired from MRI scans, which use magnetic field gradients and electromagnetic pulses to spatially encode the signal intensity in the region of interest. 3D spatial domain information can be acquired from PET scans, which detect radiation from particle annihilation to form a 3D medical image. Therefore, in this case, time-resolved imaging can also be referred to as 4D imaging.

[0029] In a possible embodiment of the imaging apparatus, the data acquisition module is configured such that the spatial domain information includes 2D spatial domain information. 2D spatial domain information can be acquired from imaging that represents the region of interest in the dimensions of length and width. 2D spatial domain information can be acquired from conventional X-ray radiography, which uses X-rays to produce 2D images of the region of interest. 2D spatial domain information can be acquired from optical microscopy, which uses visible light to produce 2D images of cells and tissues at the microscopic level. 2D spatial domain imaging can be acquired from nuclear medicine planar scintillation imaging, which detects gamma rays emitted by a radioactive tracer to create 2D images of physiological processes. 2D spatial domain information can be acquired from conventional 2D angiography, which uses X-rays and contrast agents injected into blood vessels to visualize their structure and flow. In these cases, the additional dimension provided by time-resolved imaging is time.

[0030] In a possible embodiment of the imaging device, the data acquisition module is configured to capture and receive multiple time-ordered datasets during at least one of the following predetermined time periods: respiratory cycle; cardiac cycle; diffusion activity period; fluoroscopic imaging period; and perfusion period. In conventional 2D or 3D medical imaging, motion in the region of interest has a significant impact on image quality. In some cases, even the patient's respiratory activity can affect image quality. Respiratory activity is essentially periodic. Time-resolved medical imaging utilizes this particular characteristic. During a respiratory cycle or cardiac cycle, at least one predetermined time point can be defined for capturing 2D or 3D images. For example, if only one predetermined time point is defined within a respiratory cycle or cardiac cycle, then an image of the same predetermined time point can be captured at least for the next subsequent respiratory cycle or cardiac cycle, respectively. If a pixel or voxel value differs significantly from surrounding pixel or voxel values ​​from the same predetermined time point but in different respiratory cycles, it can be assumed to be noise. If more than one predetermined time point is defined within a respiratory or cardiac cycle, the proposed imaging apparatus can also use this information to denoise time-resolved medical images taken at different predetermined time points within the same respiratory cycle, and alternatively, from images taken separately from different respiratory or cardiac cycles. The more scans performed within the region of interest, the easier the noise filtering becomes. The proposed imaging apparatus can also be used for other substantially periodic activities. For medical imaging procedures focusing on medical activities over a specific time period, such as diffusion activity imaging, fluorescence fluoroscopy, and perfusion imaging, the proposed imaging apparatus can also be applied to such medical imaging procedures. When these pixel or voxel values ​​differ significantly from surrounding pixel or voxel values ​​from scans that include similar anatomical structures but are recoded at different predetermined time points within a time period or time cycle, the pixel or voxel value can also be assumed to be noise.

[0031] In one possible embodiment of this imaging apparatus, the apparatus includes at least one of the following: a CT scanner; an MRI scanner; a PET scanner; and an X-ray machine. CT scanners, etc., are characterized by tomographic imaging, in which cross-sectional images from different layers of a region of interest are imaged and aggregated into a 3D image. MRI scanners, etc., use magnetic field gradients and electromagnetic pulses to spatially encode the signal intensity in the region of interest. PET scanners, etc., detect radiation from particle annihilation to form 3D medical images. X-ray machines, etc., emit X-ray beams to penetrate the region of interest to obtain projection data for reconstructing medical images. X-ray machines can be used in radiographic procedures, fluoroscopy systems, mammography, angiography, etc. The imaging apparatus proposed herein enables cross-platform applications and improves application accessibility and flexibility.

[0032] In a possible embodiment of the imaging device, at least one denoising module further includes an optimization module configured to set / adjust a configurable / adjustable bilateral filter by implementing an optimization algorithm, wherein the optimization algorithm allows the parameterization of the configurable / adjustable bilateral filter to be automatically set. The setting / adjustment process of the configurable / adjustable bilateral filter is computer-executable. The configurable / adjustable bilateral filter may include at least one configurable / adjustable bilateral filter layer. Multiple configurable / adjustable bilateral filter layers established by the computer-executable denoising module may exist, wherein each configurable / adjustable bilateral filter layer may have different filter parameters according to different noise characteristics. The input configurable / adjustable parameters of the configurable / adjustable bilateral filter may come from a low-SNR medical image obtained by simulating a low-SNR image from an existing image with high image quality. The optimization algorithm may involve a backpropagation algorithm. With the implementation of the backpropagation algorithm, the difference between the predicted image output and the actual image with high image quality is backpropagated through the configurable / adjustable bilateral filter to optimize the parameterization. The configurable / adjustable bilateral filter layer can be differentiable, meaning that the backpropagation algorithm can be used as a gradient estimation method to update the parameters by reducing the value of the cost / loss function during the setup / adjustment of the bilateral filter. Therefore, the parameterization of the configurable / adjustable bilateral filter is automated. During this process, filtering actions for the sine wave and the reconstructed medical image can be constructed simultaneously. The setup / adjustment of the bilateral filter can be self-supervised training. In self-supervised training, the training parameters may not require manual labeling, which is more time- and cost-effective, and the training parameters are less likely to get trapped in local optima.

[0033] In a possible embodiment of the imaging device, the optimization module is further configured such that the parameterization of the settable / adjustable bilateral filter includes two or three spatial parameters, an intensity range parameter, and a time-intensity distribution of these parameters. Depending on the characteristics of the bilateral filter and time-resolved medical imaging, the settable / adjustable parameters may use only two spatial parameters, one intensity range value, and the time-intensity distribution of these parameters. Alternatively, the settable / adjustable parameters may use only three spatial parameters, one intensity range value, and the time-intensity distribution of these parameters. In medical imaging, bilateral filters typically include domain filters and range filters. Domain filters operate based on the spatial characteristics of the image. This refers to modifying the pixel or voxel values ​​of the data based on the location of the pixel or voxel values. Range filters selectively process pixel or voxel values ​​based on their intensity to highlight or suppress brightness in a specific range. When the settable / adjustable parameters of the bilateral filter can use only two spatial parameters, one intensity range value, and the time-intensity distribution of these parameters, the settable / adjustable parameters may refer to the spatial kernel size width of the length and width dimensions of the image data used in the image generation process, the intensity range kernel size width of these parameters, and the time-intensity distribution. When the settable / adjustable parameters of at least one bilateral filter can be achieved using only three spatial parameters, one intensity range value, and the temporal-intensity distribution of these parameters, the settable / adjustable parameters can refer to the spatial kernel size width with respect to the length, width, and depth dimensions of the image data used in the image generation process, the intensity range kernel size width of these parameters, and the temporal-intensity distribution. The kernel can be a Gaussian kernel, which is used to apply Gaussian smoothing to the image by convolving the image with a Gaussian function. A Gaussian kernel is essentially a 2D or 3D matrix where values ​​closer to the center of the matrix (also called weights in the matrix) are higher, and the values ​​decrease symmetrically as they move away from the center. Compared to other denoising methods using thousands of parameters, such as convolutional neural networks, the proposed method uses only a small number of parameters for its setup / adjustment process, achieving similar filtering performance with low computational load. The method is also interpretable and does not produce illusions, as it is based on established filter operations that can be optimized and fine-tuned for each specific noise characteristic.

[0034] In a possible embodiment of the method, the order of the multiple time-ordered datasets and the weights used in the time-intensity distribution are determined based on at least one of the following: temporal proximity; anatomical similarity; and automatic similarity measures. The time-ordered datasets do not necessarily have to be ordered in sequential chronological order. Images were captured at different time points and can be ordered based on temporal proximity of the time-resolved imaging, anatomical similarity of the time-resolved imaging, or automatic similarity measures such as structural similarity index, SSIM, mean squared error, MSE, etc. The weights of the time-intensity distribution may differ in different specific use cases; therefore, multiple time-ordered datasets do not necessarily have to be ordered in sequential chronological order. If a pixel or voxel value differs significantly from the surrounding pixel or voxel values ​​of the same point in the region of interest from different time-series records, it can be assumed to be noise.

[0035] In a possible embodiment of this method, the amount of deposited radiation dose required to acquire medical images is reduced when the steps of capturing and receiving multiple time-sequential datasets involve ionizing radiation. Ionizing radiation used in medical imaging, such as X-rays and gamma rays, can cause stochastic effects in living tissue. The effective dose limit for adults in any given year is limited to 20 millisieverts (mSv). The effective dose assesses the likelihood of long-term effects that may occur after radiation exposure. The absorbed dose assesses the likelihood of biochemical changes in a specific tissue after radiation exposure, measured in milligrels (mGy). Since the calculation of the effective dose is proportional to the absorbed dose to all organs of the body, the higher the absorbed dose, the more severe the potential radiation damage to the body. Typically, in CT scans, a relatively high radiation dose of 30–60 mGy is required to allow for time-resolved image reconstruction with acceptable image quality. The proposed denoising method denoises low-dose images, resulting in denoised low-dose images with performance compatible with high-dose images. Using the proposed denoising method, noise reduction can be applied to reduce the dose required to acquire medical images, and in cases of medical imaging involving ionizing radiation, additional repeated imaging can be permitted to accommodate treatment plans.

[0036] In a possible embodiment of this method, the method operates in a registration manner. Registration brings significant improvements, especially when combined with registration, the performance of the filter can be improved. However, it goes without saying that this method can also operate without registration. Registration in temporally resolved medical imaging refers to the process of aligning two or more images into a common coordinate system. It preferably ensures that anatomical or functional features in the images correspond spatially, even if the images were captured at different times, in different modalities, or from different viewpoints. Since temporally resolved medical imaging processes capture images chronologically within at least one time period, the registration process is automatically performed during the imaging capture process. When the region of interest no longer includes the same tissue over time, information from these time points is inherently excluded by the intensity threshold of the bilateral filter. This is important because deformable registration is a nontrivial task with high computational load and its own error magnitude.

[0037] Where appropriate, the above configurations and developments can be combined. Implementations can be combined with each other as needed, provided it is reasonable. Other possible configurations, developments, and implementations of this disclosure include combinations of features of this disclosure that have been previously described or are described below with reference to embodiments that are not explicitly mentioned. In particular, in such cases, those skilled in the art will also add individual aspects as improvements or supplements to the basic form of this disclosure.

[0038] Figure 1 A block diagram of an example of an image device is shown.

[0039] Figure 1 The image apparatus in the figure is indicated by reference numeral 10. The image apparatus 10 includes a data acquisition module 11, at least one denoising module 14, and an output module 17. The data acquisition module 11 is configured to capture and receive multiple time-ordered datasets 12 used in the image generation process for entities over at least one predetermined time period, such that the multiple time-ordered datasets 12 include spatial domain information and temporal domain information. Receiving the multiple time-ordered datasets 12 may refer to temporarily storing the captured datasets and awaiting later processing. Receiving the multiple time-ordered datasets 12 may also refer to receiving datasets from an external data source. At least one denoising module 14 is coupled to the data receiving module 13 and includes a settable / adjustable bilateral filter 15. The at least one denoising module 14 is configured to use the received spatial domain information and temporal domain information to set / adjust the filter parameters of the settable / adjustable bilateral filter 15, thereby denoising the multiple time-ordered datasets 12 using the settable / adjustable bilateral filter 15 and acquiring multiple denoised time-ordered datasets 16. The output module 17 is coupled to at least one denoising module 14 and is configured to output at least one time-resolved medical image 18 for an entity using multiple denoised time-ordered datasets 16.

[0040] Figure 2 A block diagram of another example of an imaging device is shown. Figure 2 In the middle, the data receiving module 13 includes an input module 19 for capturing multiple time-sorted datasets 12 and a data receiving module for receiving the captured multiple time-sorted datasets 12. Figure 2 The rest of the parts are basically the same as Figure 1 Same as shown.

[0041] The entity to be imaged can refer to a patient, a diseased animal, a healthy person, or a region of interest in an animal, which provides comparative data for research or other purposes through medical imaging. The entity to be imaged can also refer to a non-living object, such as an imaging phantom designed to simulate the anatomy or tissue characteristics of a human or animal.

[0042] Multiple time-ordered datasets 12 may include raw detector data described by multiple sinusoids. In some medical imaging processes, sinusoids are used as 2D transition images between raw detector data and the final medical image. The proposed imaging device allows for direct filtering of the sinusoids. The proposed imaging device can directly construct filtering functions on the sinusoids, which facilitates subsequent data processing steps with less noise.

[0043] Multiple time-ordered datasets 12 can include reconstructed image data represented by multiple reconstructed images. The proposed imaging apparatus also allows filtering of the reconstructed medical images. The process of creating an image using the original detector data is called image reconstruction. The reconstructed image data can come from data encoded in a sinusoidal graph. The reconstructed image data can also come from data encoded in other forms of transitional images, such as k-space images or other images. The proposed imaging apparatus allows filtering of the reconstructed medical images, further increasing the flexibility of the imaging apparatus application. In addition, the proposed imaging apparatus can allow filtering of both the sinusoidal graph and the reconstructed medical images simultaneously.

[0044] Multiple time-ordered datasets12 include spatial and temporal information, with the spatial information comprising 3D spatial information. 3D spatial information can be obtained from images representing the region of interest in the dimensions of length, width, and depth. Time is considered an additional dimension in medical imaging. Therefore, in this context, time-resolved imaging can also be referred to as 4D imaging.

[0045] Multiple temporally ordered datasets12 include spatial and temporal information, with the spatial information comprising 2D spatial information. The 2D spatial information can be obtained from imaging that represents the region of interest in length and width dimensions. In this case, the additional dimension provided by temporally resolved imaging is time.

[0046] Multiple time-ordered datasets 12 are captured and received within at least one predetermined time period, which may be at least one respiratory cycle or cardiac cycle; at least one diffusion activity period; at least one fluorescence fluoroscopy imaging period; or at least one perfusion period. The time periods, which can be represented by a time period, can be substantially periodic, such as respiratory cycles or cardiac cycles, respectively. Time-resolved medical imaging utilizes this particular feature. The proposed imaging device 10 can also be used for other substantially periodic activities. If a pixel or voxel value differs significantly from surrounding pixel or voxel values ​​from the same predetermined time point but in different periods, it can be assumed to be noise. If more than one predetermined time point is defined within the time period, the proposed imaging device 10 can also use this information to denoise time-resolved medical images from images taken at different predetermined time points but within the same time period, and alternatively, from images taken in different time periods. The proposed imaging device 10 can also be applied to medical imaging processes focusing on medical activities within a specific time period, such as diffusion activity imaging, fluorescence fluoroscopy, and perfusion imaging. When these pixel or voxel values ​​differ significantly from the surrounding pixel or voxel values ​​from scans that include similar anatomical structures but are recoded at different predetermined time points within a time period or time cycle, it can also be assumed that the pixel or voxel value is noise.

[0047] At least one denoising module 14 may further include an optimization module 101. The optimization module 101 is configured to set / adjust a configurable / adjustable bilateral filter 15 by implementing an optimization algorithm that allows for automatic parameterization of the configurable / adjustable bilateral filter 15. The setting / adjustment process of the configurable / adjustable bilateral filter 15 is computer-executable. The configurable / adjustable bilateral filter 15 may include at least one configurable / adjustable bilateral filter layer. Multiple configurable / adjustable bilateral filter layers may exist, established by the computer-executable denoising module, wherein each configurable / adjustable bilateral filter layer may have different filter parameters according to different noise characteristics. The input configurable / adjustable parameters of the configurable / adjustable bilateral filter 15 may come from a low-SNR medical image obtained by simulating a low-SNR image from an existing image with high image quality. The optimization algorithm may involve a backpropagation algorithm. With the implementation of the backpropagation algorithm, the difference between the predicted image output and the actual image with high image quality is backpropagated through the configurable / adjustable filter to optimize the parameterization. The configurable / adjustable bilateral filter layer can be differentiable, meaning that backpropagation can be used as a gradient estimation method to update parameters by decreasing the value of the cost / loss function during the training of the bilateral filter. Therefore, the parameterization of the configurable / adjustable bilateral filter is automatic. During this process, filtering actions for the sine wave and the reconstructed medical image can be constructed simultaneously. The setup / adjustment of the configurable / adjustable bilateral filter can be self-supervised training. In self-supervised training, the training parameters may not require manual labeling, which is more time- and cost-effective, and the training parameters are less likely to get trapped in local optima.

[0048] The optimization module 101 is also configured to enable the parameterization of the settable / adjustable bilateral filter to include two or three spatial parameters, intensity range parameters, and time-intensity distributions of these parameters.

[0049] The imaging device 10 can be a CT scanner, MRI scanner, PET scanner, or X-ray machine. Therefore, the imaging device 10 proposed in this paper enables cross-platform applications and improves the accessibility and flexibility of the applications.

[0050] Output module 17 uses multiple denoised temporally ordered datasets 16 to output at least one time-resolved image 18 for an entity. The at least one time-resolved image 18 can be received and displayed by at least one screen integrated into imaging device 10. The at least one time-resolved image 18 can be received and displayed by at least one screen having a wired or wireless connection to imaging device 10. The at least one time-resolved image 18 can be received and displayed in a cloud-based repository linked to imaging device 10, allowing others to download the at least one time-resolved image 18 from the cloud-based repository. The at least one time-resolved image 18 can be stored in at least one piece of machine code, allowing others to obtain the at least one time-resolved image 18 by decoding the at least one piece of machine code.

[0051] Figure 3 A figure illustrating an example of multiple time-ordered datasets captured within a time period is shown. Figure 20 shows multiple time-ordered datasets captured within respiratory cycle 21. The vertical axis represents respiratory amplitude. The horizontal axis represents time. At least ten time points P1, P2, ..., P10 are predetermined within respiratory cycle 21. For each time point, a dataset is captured. After one respiratory cycle 21, at least ten datasets are captured and ordered temporally to obtain multiple time-ordered datasets 12. Figure 2 As shown, each of the multiple time-ordered datasets 12 can represent a 3D volume. A predetermined time point within a time period can have more or fewer than ten time points.

[0052] For example, if only one predetermined time point is defined within a respiratory cycle, then an image of the same predetermined time point can be captured at least for the next subsequent respiratory cycle. If a pixel or voxel value differs significantly from surrounding pixel or voxel values ​​from the same predetermined time point but in different respiratory cycles, it can be assumed to be noise. If more than one predetermined time point is defined within a respiratory cycle, the proposed imaging apparatus can also use this information to denoise time-resolved medical images from images taken at different predetermined time points within the same respiratory cycle, and alternatively, from images taken in different respiratory cycles. Therefore, the more scans performed in the region of interest, the easier the noise filtering becomes. The proposed imaging apparatus can also be used for other substantially periodic activities. For medical imaging procedures focusing on medical activities over a specific time period, such as diffusion activity imaging, fluorescence fluoroscopy, perfusion imaging, etc., the proposed imaging apparatus can also be applied to such medical imaging procedures. When these pixel or voxel values ​​differ significantly from surrounding pixel or voxel values ​​from scans that include similar anatomical structures but are recoded at different predetermined time points within a time period or cycle, the pixel or voxel value can also be assumed to be noise.

[0053] Figure 4 A schematic diagram illustrating an example of the filtering process for an exemplary image using a configurable / adjustable bilateral filter is shown.

[0054] exist Figure 4 In this example, taking a 2D reconstructed medical image 30 as an example, a filtering process for a reconstructed medical image 30 among multiple time-ordered reconstructed medical images is presented using a configurable / adjustable bilateral filter 35. The configurable / adjustable bilateral filter 35 includes a domain filter 31 implementing a Gaussian kernel and a range filter 32. In this example, the Gaussian kernel is essentially a 2D matrix where values ​​closer to the center of the matrix (also called weights in the matrix) are higher, and the values ​​decrease symmetrically as they move away from the center. The domain filter 31 operates based on the spatial characteristics of the image. It refers to modifying the intensity value of the pixel region 34 by performing a convolution process based on a normalized Gaussian space kernel 37 and the pixel region 34. Therefore, a denoised pixel region with the same imaging area as the pixel region 34 can be obtained. The pixel region 34 has been smoothed by the domain filter 31. This filtering process reduces random fluctuations around the pixel region 34. In this example, the shape of the Gaussian space kernel 37 has two dimensions, allowing the shape of the Gaussian space kernel 37 to be set / adjusted according to different noise characteristics relative to its length and width dimensions. Range filter 32 processes pixel values ​​based on their intensity to highlight or suppress brightness within a specific range. This refers to modifying the intensity value of pixel region 33 by performing a convolution process with normalized Gaussian range kernel 36 and pixel region 33. Therefore, a denoised pixel region with the same imaging area as pixel region 33 can be obtained. Pixel region 33 has been denoised by range filter 32; however, edges within pixel region 33 have been preserved. The shape of Gaussian range kernel 36 can also be set / adjusted according to different noise characteristics.

[0055] The configurable / adjustable bilateral filter 35 successfully smooths the image while preserving edges in the reconstructed medical image 30. Figure 3 The example can be extended by performing a filtering process in 3D reconstructed medical images. In this scenario, pixel regions are replaced by voxel regions. The shape of the Gaussian space kernel 37 has three dimensions, allowing the shape of the Gaussian space kernel 37 to be adjusted relative to its length, width, and depth dimensions.

[0056] Such a filtering process can be performed on each pixel or voxel of the reconstructed medical image to construct denoising, so that a denoised reconstructed medical image with high SNR can be acquired.

[0057] Figure 5 A diagram illustrating an example of a method for time-resolved medical imaging is shown.

[0058] The proposed method 40 includes three steps: Step S41, capturing and receiving multiple time-ordered datasets 12 used in the image generation process of an entity within at least one predetermined time period, such that the multiple time-ordered datasets 12 include spatial domain information and temporal domain information; Step S42, setting / adjusting a configurable / adjustable bilateral filter 15 using the spatial domain information and temporal domain information of the multiple time-ordered datasets 12, thereby denoising the multiple time-ordered datasets 12 using the configurable / adjustable bilateral filter 15, and obtaining multiple denoised time-ordered datasets 16; Step S43, using the multiple denoised time-ordered datasets 16 to output at least one time-resolved medical image 18 of the entity.

[0059] Apart from Figure 4 Beyond the description, the filtering process for multiple temporally ordered datasets 12 also involves the temporal-intensity distribution of spatial and range parameters. The order of these multiple temporally ordered datasets 12 and the weights used in their temporal-intensity distribution are determined based on at least one of the following: temporal proximity; anatomical similarity; and automatic similarity measures. The temporally ordered datasets do not necessarily have to be ordered in sequential chronological order. Images were captured at different time points and can be ordered based on temporal proximity of the temporally resolved imaging, anatomical similarity of the temporally resolved imaging, or automatic similarity measures such as structural similarity index, SSIM, mean squared error, MSE, etc. The weights of the temporal-intensity distribution may differ in different specific use cases; therefore, multiple temporally ordered datasets do not necessarily have to be ordered in sequential chronological order. If a pixel or voxel value differs significantly from the surrounding pixel or voxel values ​​of the same point in the region of interest from different temporal records, it can be assumed to be noise.

[0060] Compared to other methods that train bilateral filters using thousands of parameters, such as convolutional neural networks, the proposed method uses only a small number of parameters in its setup / tuning process, achieving similar filtering performance with low computational load. The method is also interpretable and does not produce illusions, as it is based on established filter operations that can be optimized and fine-tuned for each specific noise characteristic. By employing a time-intensity distribution, the amount of deposited radiation dose and prolonged acquisition time required to acquire medical images are reduced when the data used in the captured image generation process involves ionizing radiation and / or long acquisition times, without compromising image quality such as contrast, temporal resolution, and geometric accuracy. Furthermore, the method is operable without registration.

[0061] Although this disclosure has been described above by way of embodiments, it is not limited thereto and can be modified in a wide range of ways. In particular, this disclosure can be changed or modified in various ways without departing from its core.

[0062] In this patent application, nouns and pronouns referring to people generally do not specify a particular gender.

Claims

1. An imaging device (10), the imaging device (10) comprising: The data acquisition module (11) is configured to capture and receive multiple time-ordered datasets (12) used in the image generation process for an entity over at least a predetermined time period, wherein the multiple time-ordered datasets (12) include spatial domain information and temporal domain information; At least one denoising module (14) is coupled to the data acquisition module (13) and includes a configurable / adjustable bilateral filter (15), wherein the at least one denoising module (14) is configured to set / adjust the filter parameters of the configurable / adjustable bilateral filter (15) using the received spatial domain information and the temporal domain information, such that the configurable / adjustable bilateral filter (15) is used to denoise the plurality of time-sorted datasets (12) and acquire a plurality of denoised time-sorted datasets (16). The output module (17) is configured to use the plurality of denoised temporally ordered datasets (16) to output at least one temporally resolved medical image (18) for the entity.

2. The imaging apparatus according to claim 1, wherein the data acquisition module (11) is configured to capture and receive the plurality of time-ordered datasets (12) comprising raw detector data described by a plurality of sine waves.

3. The imaging apparatus according to any one of the preceding claims, wherein the data acquisition module (11) is configured to capture and receive the plurality of time-ordered datasets (12) comprising reconstructed image data represented by a plurality of reconstructed images.

4. The imaging apparatus according to any one of the preceding claims, wherein the data acquisition module (11) is configured such that the spatial domain information includes three-dimensional 3D spatial domain information.

5. The imaging apparatus according to any one of claims 1 to 3, wherein the data acquisition module (11) is configured such that the spatial domain information includes two-dimensional 2D spatial domain information.

6. The imaging apparatus according to any one of the preceding claims, wherein the data acquisition module (11) is configured to capture and receive the plurality of time-ordered datasets (12) during at least one of the following predetermined time periods: Respiratory cycle; Heart cycle; The period of diffusion activity; Perspective imaging time period; Infusion period.

7. The imaging apparatus according to any one of the preceding claims, wherein the imaging apparatus (10) comprises at least one of the following: Computational computed tomography (CT) scanners; Magnetic resonance imaging scanner; Positron emission tomography (PET) scanner; X-ray machine.

8. The imaging apparatus according to any one of the preceding claims, wherein the at least one denoising module (14) further comprises an optimization module configured to set / adjust the settable / adjustable bilateral filter (15) by implementing an optimization algorithm, wherein the optimization algorithm allows the parameterization of the settable / adjustable bilateral filter (15) to be automatically set.

9. The imaging apparatus according to any one of the preceding claims, wherein the optimization module is further configured such that the parameterization of the settable / adjustable bilateral filter (15) includes two or three spatial parameters, an intensity range parameter, and a time-intensity distribution of these parameters.

10. A method (40) for time-resolved medical imaging, particularly by employing an imaging apparatus according to any one of claims 1 to 9, the method comprising: Capture and receive (S41) multiple temporally ordered datasets (12) used in the image generation process for entities within at least one predetermined time period, wherein the multiple temporally ordered datasets (12) include spatial domain information and temporal domain information; Using the received spatial domain information and temporal domain information, the filter parameters of the settable / adjustable bilateral filter (15) are set / adjusted (S42) so that the settable / adjustable bilateral filter (15) is used to denoise the plurality of time-sorted datasets (12) and obtain a plurality of denoised time-sorted datasets (16). The plurality of denoised temporally ordered datasets (16) are used to output at least one temporally resolved medical image (18) for the entity.

11. The method according to any one of the preceding method-based claims, wherein the order of the plurality of time-ordered datasets (12) and the weighting used therein in the time-intensity distribution are determined according to at least one of the following: Time proximity; Anatomical similarity; Automatic similarity measurement.

12. The method according to any one of the preceding method-based claims, wherein when the step of capturing and receiving (S41) the plurality of time-ordered datasets (12) involves ionizing radiation, the amount of deposited radiation dose required to acquire medical images is reduced.

13. The method according to any one of the preceding method-based claims, wherein the method operates by registration.

14. A computer program product comprising a computer program that, when a program portion of the computer program is executed by a computer, causes the imaging apparatus (10) according to any one of claims 1 to 9 to perform the method (40) according to any one of claims 10 to 13.

15. A non-transitory computer-readable storage medium on which a computer stores a computer program according to claim 14.

Images