Medical image noise reduction method and system

By extracting target image data from multi-frame medical image data and performing noise reduction processing, and by fitting noise reduction parameters using the target vector and set matrix, the problem of low signal-to-noise ratio in mPLD-ASL sequences is solved, achieving high signal-to-noise ratio and clarity in medical images.

CN118941456APending Publication Date: 2024-11-12SHANGHAI UNITED IMAGING HEALTHCARE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310514021.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-08
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In existing magnetic resonance imaging techniques, especially the mPLD-ASL sequence, the signal-to-noise ratio is low. This is particularly true in high-field magnetic resonance applications, which further reduces the signal-to-noise ratio of ASL images, affecting image quality and application effectiveness.

Method used

The process involves identifying at least one target medical image from multiple frames of medical image data, denoising each frame, fitting denoising parameters using the relationship between the target vector and the set matrix, and updating the image data.

Benefits of technology

It significantly improves the signal-to-noise ratio of medical images while maintaining image clarity, without requiring a large amount of additional training data, and avoids the problems of image blurring and increased scanning time in traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941456A_ABST
    Figure CN118941456A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides noise reduction of a medical image, and the method comprises the steps: obtaining multiple frames of medical image data, and determining at least one frame of target medical image data from the multiple frames of medical image data; performing noise reduction on each frame in the at least one frame of target medical image data, wherein noise reduction on each frame of target medical image comprises the following steps: forming a target vector by using a plurality of data points of the target medical image data; based on the positions of the plurality of data points, determining data points, the positions of which are the same as the positions of the plurality of data points, in the rest frames of medical image data, forming a set matrix by the data points of the rest frames of medical image data, the rest frames of medical image data being the rest frames of medical image data except the target medical image data in the multiple frames of medical image data; based on the relationship between the target vector and the set matrix, fitting a noise reduction parameter corresponding to the target vector; and performing noise reduction on the target medical image data based on the noise reduction parameter, and determining updated medical image data of the target medical image data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of magnetic resonance imaging, and in particular to a method and system for reducing noise in medical images. Background Art

[0002] The multiple post labeling delay (mPLD) arterial spin labeling (ASL) sequence is a multi-phase ASL scanning sequence. The MRI perfusion technology based on the mPLD-ASL sequence uses water molecules in the blood as endogenous tracers to measure the blood perfusion volume and arterial transit time (ATT) of the target tissue. These MRI quantitative parameters reflect the blood perfusion information of the target tissue and have been widely used in the research and clinical treatment of vascular diseases, tumors and other diseases. At present, an important development bottleneck of the mPLD-ASL technology is its low signal-to-noise ratio. Especially in high-field MRI applications, it is often necessary to acquire higher MRI image resolution, which leads to a further reduction in the signal-to-noise ratio of ASL images.

[0003] Therefore, it is necessary to provide a medical image denoising method and system to improve the signal-to-noise ratio of the image. Summary of the invention

[0004] One of the embodiments of the present specification provides a method for denoising a medical image. The method includes: acquiring multiple frames of medical image data, determining at least one frame of target medical image data from the multiple frames of medical image data; performing denoising on each frame of the at least one frame of target medical image data, wherein denoising each frame of the target medical image includes the following steps: composing a target vector with multiple data points of the target medical image data; based on the positions of the multiple data points, determining the data points in the remaining frames of medical image data that have the same positions as the multiple data points, composing a set matrix with the data points of the remaining frames of medical image data, wherein the remaining frames of medical image data are the medical image data of the remaining frames of the multiple frames of medical image data except the target medical image data; fitting the denoising parameters corresponding to the target vector based on the relationship between the target vector and the set matrix; denoising the target medical image data based on the denoising parameters, and determining the updated medical image data of the target medical image data.

[0005] In some embodiments, determining at least one frame of target medical image data from multiple frames of medical image data includes: determining the multiple frames of medical image data as target medical image data to perform noise reduction on the multiple frames of medical image data.

[0006] In some embodiments, forming a target vector from multiple data points of the target medical image data includes: dividing or partitioning the multiple data points of the target medical image data into blocks to obtain multiple data blocks, each data block including multiple data points; for each of the multiple data blocks, forming a target vector from the multiple data points contained in the data block.

[0007] In some embodiments, forming a target vector out of multiple data points of the target medical image data includes: arranging the multiple data points in sequence vertically to form a column vector, wherein the column vector is the target vector; forming a set matrix out of the data points of the remaining frames of medical image data includes: for each of the multiple data points, based on the position of the data point, determining the data points in the remaining frames of medical image data that have the same position as the data point, and arranging them in sequence horizontally to form a row vector; arranging the multiple row vectors corresponding to the multiple data points in sequence vertically to form a set matrix, wherein the arrangement order of the multiple row vectors corresponds to the arrangement order of the multiple data points in the target vector.

[0008] In some embodiments, the medical image data consists of real data and noise, the real data of multiple frames of medical image data are the same or meet the first feature, and the noise of the multiple frames of medical image data meet the second feature.

[0009] One of the embodiments of the present specification provides a medical image denoising system, the system comprising a determination module, used to acquire multiple frames of medical image data, and determine at least one frame of target medical image data from the multiple frames of medical image data; a denoising module, used to perform denoising on each frame of the at least one frame of target medical image data, the denoising module further comprising: a first composition unit, used to form a target vector from multiple data points of the target medical image data; a second composition unit, used to determine, based on the positions of the multiple data points, data points in the remaining frames of medical image data having the same positions as the multiple data points, and form a set matrix from the data points of the remaining frames of medical image data, the remaining frames of medical image data being medical image data of the remaining frames of the multiple frames of medical image data except for the target medical image data; a fitting unit, used to fit a denoising parameter corresponding to the target vector based on the relationship between the target vector and the set matrix; a determination unit, used to perform denoising on the target medical image data based on the denoising parameter, and determine updated medical image data of the target medical image data.

[0010] In some embodiments, the determination module is further used to: determine the multiple frames of medical image data as target medical image data to perform noise reduction on the multiple frames of medical image data.

[0011] In some embodiments, the noise reduction module is further used to: segment or partition multiple data points of the target medical image data to obtain multiple data blocks, each data block includes multiple data points; for each of the multiple data blocks, the multiple data points contained in the data block are combined into a target vector.

[0012] One of the embodiments of the present specification provides a medical image denoising apparatus, comprising a processing device, wherein the processing device is used to execute a medical image denoising method.

[0013] One of the embodiments of the present specification provides a computer-readable storage medium, wherein the storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes a medical image denoising method.

[0014] Currently, methods to improve the signal-to-noise ratio of magnetic resonance imaging include repeated scanning, traditional image-based denoising algorithms, and deep learning-based denoising methods.

[0015] The repeated scanning method is relatively simple to implement. By averaging the generated images after repeated scanning of the ASL sequence, the signal-to-noise ratio of the image can be effectively improved. However, this method will also cause the scanning time to increase exponentially, especially for the mPLD-ASL sequence, which has a long scanning time (for example, a single scan of 5 phases takes about 6 minutes). The increase in scanning time caused by repeated scanning seriously restricts the feasibility of this method.

[0016] Traditional image-based denoising methods usually use different filter functions to post-process the generated image to reduce image noise. However, while reducing image noise, image filtering-based denoising methods will cause a blurring effect on the image, that is, reduce the sharpness of the image, thus affecting the imaging effect and application.

[0017] In recent years, image denoising technology based on deep learning has been widely used. The ASL denoising method based on deep learning uses a large amount of training data to learn the noise model in the ASL image, thereby achieving the purpose of removing noise and showing good denoising effect. However, deep learning methods usually require a large amount of ASL images as training data to train the deep learning model to obtain reliable results. In addition, the acquisition of high-quality ASL images used as gold standard data in model training is very difficult (for example, it can be acquired by repeated scanning dozens of times), and usually requires a long scanning time, which is prone to problems such as motion artifacts caused by the patient's movement during the scan.

[0018] In order to solve the above problems, the present invention determines at least one frame of target medical image data from multiple frames of medical image data, and performs noise reduction on each frame of the at least one frame of target medical image data. This does not require the input of a large amount of additional training data, and can significantly improve the signal-to-noise ratio of the medical image while ensuring the clarity of the medical image. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0020] Figure 1 is a schematic diagram of an application scenario of a medical image denoising system according to some embodiments of this specification;

[0021] Figure 2 is a module diagram of a medical image denoising system according to some embodiments of this specification;

[0022] Figure 3 is an exemplary flow chart of denoising of medical images according to some embodiments of this specification;

[0023] Figure 4 is a schematic diagram of multiple data points of target medical image data according to some embodiments of this specification;

[0024] Figure 5 is a schematic diagram of partitioning multiple data points of target medical image data into blocks according to some embodiments of this specification;

[0025] Figure 6 is a comparison diagram of medical images before and after noise reduction according to some embodiments of this specification;

[0026] Figure 7 It is a schematic diagram of an exemplary process of denoising a medical image using a machine learning model according to some embodiments of this specification. DETAILED DESCRIPTION

[0027] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0028] It should be understood that the "system", "device", "unit" and / or "module" used herein are a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

[0029] As shown in this specification and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "comprise" and "include" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0030] Flowcharts are used in this specification to illustrate the operations performed by the system according to the embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed precisely in order. Instead, the steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or one or more operations may be removed from these processes.

[0031] Figure 1 1 is a schematic diagram of an application scenario of a medical image denoising system according to some embodiments of the present specification. The medical image denoising system can be used to denoise a medical image. In some embodiments, the application scenario 100 of the medical image denoising system may include a storage device 110, a processing device 120, and an imaging scanner 130.

[0032] Storage device 110 may store data, instructions, and / or any other information. In some embodiments, storage device 110 may store data and / or instructions related to noise reduction of medical images. For example, storage device 110 may store computer instructions for performing noise reduction of medical images. For another example, storage device 110 may store medical images, such as Figure 1 The target medical image 110-1 and the updated medical image 110-2 are shown. The target medical image 110-1 corresponds to the target medical image data, and the updated medical image 110-2 corresponds to the updated medical image data. The target medical image 110-1 and the updated medical image 110-2 are both brain images, but the target medical image 110-1 is a brain image before denoising, while the updated medical image 110-2 is a brain image after denoising.

[0033] Medical images refer to magnetic resonance images acquired using a multi-frame acquisition method. Medical images may include mPLD-ASL images, blood oxygen level dependent brain functional imaging (BOLD-fMRI) images, dynamic contrast enhanced MRI (DCE-MRI) images, diffusion weighted imaging (DWI), diffusion tensor imaging (DTI) images, etc. Multi-frame medical images may include medical images repeatedly acquired at the same delay time (i.e., the same phase) and medical images acquired at different delay times (i.e., different phases). A single scan image of a single phase is called a frame of medical image. In some embodiments, medical images may include 2D images, 3D images, etc. In some embodiments, the storage device 110 may be connected to a network so as to communicate with the imaging scanner 130 to obtain the medical image acquired by the imaging scanner 130.

[0034] In some embodiments, the storage device 110 may be a storage device inside or outside the medical image noise reduction system. In some embodiments, the storage device may be a part of the processing device 120. In some embodiments, the storage device 110 may include a mass storage device, a removable storage device, a volatile read-write memory, a read-only memory (ROM), etc., or any combination thereof.

[0035] The processing device 120 can be used to process data related to the medical image noise reduction system. For example, the processing device 120 can perform noise reduction on the target medical image 110-1 by executing the medical image noise reduction method disclosed in this specification to obtain an updated medical image 110-2. Exemplarily, the processing device 120 can obtain multiple frames of medical image data, determine at least one frame of target medical image data of the target medical image 110-1 from the multiple frames of medical image data; and perform noise reduction on each frame of the at least one frame of the target medical image data.

[0036] In some embodiments, the processing device 120 can be a single server or a server group. The server group can be centralized or distributed. In some embodiments, the processing device 120 can be local or remote. In some embodiments, the processing device 120 can be implemented on a cloud platform. As an example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof. In some embodiments, the processing device 120 can be integrated or installed on the imaging scanner 130.

[0037] The imaging scanner 130 can be used to scan a patient. In some embodiments, the imaging scanner 130 can obtain a three-dimensional medical image of a scanned part of the scanned object. In some embodiments, the imaging scanner 130 can include a medical imaging device, a combination of a medical imaging device and an interventional medical device, etc. An exemplary medical imaging device may include a magnetic resonance imaging (MRI) scanner. An exemplary interventional medical device may include a radiotherapy (RT) device, an ultrasound therapy device, a thermal therapy device, a surgical intervention device, etc., or a combination thereof.

[0038] In some embodiments, the application scenario 100 of the medical image denoising system may further include one or more other devices, such as a network and / or a user terminal (not shown in the figure).

[0039] The user can interact with the medical image noise reduction system via the user terminal. Exemplary user terminals may include mobile devices, tablet computers, laptop computers, etc., or any combination thereof. For example, when it is determined to update the medical image 110-2, the processing device 120 may display it to the user via the user terminal. In some embodiments, the processing device 120 may be part of the user terminal.

[0040] The network may include any suitable wired or wireless network that can facilitate information and / or data exchange. For example, the processing device 120 and the imaging scanner 130 can transmit information and / or data via the network. In some embodiments, the storage device 110 can be connected to the network to communicate with one or more other components (e.g., the processing device 120 and the imaging scanner 130, etc.) in the application scenario 100 of the denoising system for medical images. For example, one or more components of the application scenario 100 of the denoising system for medical images can access data or instructions stored in the storage device 110 through the network.

[0041] Figure 2 It is a module diagram of a medical image denoising system according to some embodiments of the present specification.

[0042] In some embodiments, the module diagram 200 of the medical image denoising system may include a determination module 210 and a denoising module 220. The denoising module 220 may further include a first component unit 221, a second component unit 222, a fitting unit 223 and a determination unit 224.

[0043] The determination module 210 may be configured to acquire multiple frames of medical image data and determine at least one frame of target medical image data from the multiple frames of medical image data.

[0044] In some embodiments, the determination module 210 may be further configured to determine the multiple frames of medical image data as target medical image data to perform noise reduction on the multiple frames of medical image data.

[0045] In some embodiments, the medical image data may consist of real data and noise, the real data of multiple frames of medical image data are the same or meet the first feature, and the noise of the multiple frames of medical image data meet the second feature.

[0046] The denoising module 220 may be configured to perform denoising on each frame of at least one frame of target medical image data.

[0047] In some embodiments, the first composition unit 221 may be configured to compose a plurality of data points of the target medical image data into a target vector.

[0048] In some embodiments, the first composition unit 221 can be further used to segment or partition multiple data points of the target medical image data to obtain multiple data blocks, each of which includes multiple data points; for each of the multiple data blocks, the multiple data points included in the data block are composed into a target vector. In some embodiments, the first composition unit 221 is further used to sequentially arrange the multiple data points in a vertical direction to form a column vector, and the column vector is the target vector.

[0049] In some embodiments, the second composition unit 222 can be used to determine the data points in the remaining frames of medical image data that have the same positions as the multiple data points based on the positions of the multiple data points, and to form a set matrix of the data points in the remaining frames of medical image data, where the remaining frames of medical image data are the medical image data of the remaining frames in the multiple frames of medical image data except for the target medical image data.

[0050] In some embodiments, the second composition unit 222 can be further used to determine, for each of the multiple data points, based on the position of the data point, the data points in the remaining frames of medical image data that have the same position as the data point, and arrange them in sequence horizontally to form row vectors; and arrange the multiple row vectors corresponding to the multiple data points in sequence vertically to form a set matrix, and the arrangement order of the multiple row vectors corresponds to the arrangement order of the multiple data points in the target vector.

[0051] In some embodiments, the fitting unit 223 may be configured to fit the noise reduction parameters corresponding to the target vector based on the relationship between the target vector and the set matrix.

[0052] In some embodiments, the determination unit 224 may be configured to perform noise reduction on the target medical image data based on the noise reduction parameters, and determine updated medical image data of the target medical image data.

[0053] For more details about the determination module 210 and the noise reduction module 220, see Figure 3-Figure 7 and its related description.

[0054] It should be understood that Figure 2 The system and its modules shown can be implemented in various ways. It should be noted that the above description of the medical image denoising system and its modules is only for the convenience of description and cannot limit this specification to the scope of the embodiments. It is understandable that for those skilled in the art, after understanding the principle of the system, it is possible to arbitrarily combine the modules or form a subsystem to connect with other modules without deviating from this principle. In some embodiments, Figure 2 The determination module 210 and the noise reduction module 220 disclosed in the specification can be different modules in a system, or a module can realize the functions of two or more modules. For example, each module can share a storage module, or each module can have its own storage module. Such variations are all within the protection scope of this specification.

[0055] Figure 3 is an exemplary flow chart of denoising of medical images according to some embodiments of this specification. Figure 3 As shown, process 300 includes the following steps. In some embodiments, Figure 3 One or more operations of the process 300 shown may be performed in Figure 1 The denoising system for medical images is implemented in the application scenario 100 shown. For example, Figure 3 The process 300 shown may be stored in the storage device 110 in the form of instructions, and called and / or executed by the processing device 120 .

[0056] Step 310 , acquiring multiple frames of medical image data, and determining at least one frame of target medical image data from the multiple frames of medical image data. In some embodiments, step 310 may be performed by the determination module 210 .

[0057] Medical image data refers to image data collected by an imaging scanner (for example, imaging scanner 130). Medical image data may be composed of real data and noise. Exemplarily, the mPLD-ASL sequence is a multi-phase ASL scanning sequence. The sequence collects medical image data at different delay times, which are called medical image data of different phases. Usually, medical image data of the same delay time are repeatedly collected multiple times. Multiple frames of medical image data may include medical image data repeatedly collected at the same delay time (i.e., the same phase) and medical image data collected at different delay times (i.e., different phases). A single scan data of a single phase is called a frame of medical image data, and each frame of medical image data is composed of real data and noise data. Medical images are reconstructed based on medical image data through an image reconstruction algorithm. For more details about medical images, see Figure 1 and its related description.

[0058] In some embodiments, the medical image data may include three-dimensional medical image data, four-dimensional medical image data, etc. The three-dimensional medical image data is two-dimensional composed of image data of a two-dimensional medical image in space, and a third dimension composed of multiple frames (corresponding to the number of multiple frames of medical image data); the four-dimensional medical image data is three-dimensional composed of image data of a three-dimensional medical image in space, and a fourth dimension composed of multiple frames.

[0059] Real data refers to useful data collected by an imaging scanner that can reflect the patient's condition. The real data of multiple frames of medical image data are the same or meet the first feature. Specifically, the real data in the medical image data collected repeatedly with the same delay time are the same; the real data in the medical image data collected with different delay times meet the first feature.

[0060] The first feature refers to the inherent property of real data, that is, the real data in the medical image data collected at different delay times are correlated.

[0061] Noise refers to useless data received during the image acquisition process of the imaging scanner. The process of noise reduction is to eliminate or reduce the noise while retaining the real data. The noise of multi-frame medical image data meets the second characteristic.

[0062] The second feature refers to the inherent properties of noise, that is, the noise in different frames of medical image data collected is independent and uncorrelated. The medical image data of different frames include medical image data with the same delay time and medical image data with different delay times that are repeatedly collected.

[0063] In some embodiments, the processing device (eg, the processing device 120 ) may repeatedly acquire medical image data at the same delay time based on an imaging scanner, and acquire medical image data at different delay times to obtain multiple frames of medical image data.

[0064] The target medical image data refers to the medical image data to be subjected to noise reduction. In some embodiments, the target medical image data may be one frame or multiple frames. In some embodiments, the processing device may determine the acquired multiple frames of medical image data as the target medical image data to perform noise reduction on the multiple frames of medical image data.

[0065] In some embodiments, the processing device may use the medical image data that needs to be denoised as the target medical image data specified by the user. For example, the processing device may reconstruct the collected multiple frames of medical image data into an initial medical image, number it, and then display it to the user through the user terminal. After judging the medical image displayed by the user terminal, the user inputs the number of the medical image that needs to be denoised (for example, a medical image with large noise recognized by the naked eye).

[0066] In some embodiments, the processing device may determine the acquired multiple frames of medical image data as target medical image data to perform noise reduction on the acquired multiple frames of medical image data. By performing noise reduction on the multiple frames of medical image data, the signal-to-noise ratio of all acquired medical images may be improved.

[0067] Step 320, denoising each frame of at least one frame of target medical image data. In some embodiments, step 320 may be performed by the denoising module 220. The denoising of each frame of target medical image data includes the following steps:

[0068] Step 321 , compose a target vector from a plurality of data points of the target medical image data. In some embodiments, step 321 is performed by the first composition unit 221 .

[0069] A data point refers to a data value corresponding to a distinguishable element that constitutes the smallest unit in a target medical image. In some embodiments, when the target medical image data is three-dimensional medical image data, a data point is a data value corresponding to a pixel of the target medical image data. In some embodiments, when the target medical image data is four-dimensional medical image data, a data point is a data value corresponding to a voxel of the target medical image data. For example, taking three-dimensional medical image data as an example, Figure 4 As shown in the figure, the multi-frame medical image data includes N frames (for example, 5 frames, 10 frames, 20 frames, etc.), S1, S j and S Nrepresents the 1st frame of medical image data, the jth frame of medical image data and the Nth frame of medical image data, respectively, where j is any value between 1 and N. Each frame of medical image data includes M (for example, an integer number of 50, 80, 100, etc.) data points, S 1, , S j,k and S N,l They respectively represent the kth data point of the 1st frame of medical image data, the kth data point of the jth frame of medical image data, and the kth data point of the Nth frame of medical image data, where k is any value between 1 and M.

[0070] The target vector is a vector consisting of the target medical image data to be denoised. Figure 4 As shown, the j-th frame medical image data is taken as the target medical image data for noise reduction below, and the noise reduction of the j-th frame medical image data (hereinafter referred to as the j-th frame target medical image data) is taken as an example for explanation. The target vector can be a vector composed of the j-th frame target medical image data.

[0071] In some embodiments, the processing device may sequentially arrange the multiple data points of the target medical image data in a longitudinal direction to form a column vector, and the column vector is the target vector. Figure 4 As shown, the processing device can process all data points S in the jth frame target medical image data. j,1 ……S j,k ……S j,M Arrange them vertically to form a target column vector, denoted as b:

[0072] In some embodiments, the processing device may segment or partition multiple data points of the target medical image data to obtain multiple data blocks, each of which includes multiple data points; for each of the multiple data blocks, the multiple data points included in the data block are combined into a target vector. Figure 5 As shown, the processing device may process the j-th frame target medical image data S of 6×6 j The multiple data points are partitioned along the thick solid line to obtain 9 data blocks, each of which includes 2×2 data points. For each of the 9 data blocks, the 4 data points contained in the data block are combined into a target vector.

[0073] Segmentation refers to the division of target medical image data based on image information. For example, segmenting multiple data points of target medical image data may be dividing the target medical image data into regions of interest and regions of no interest.

[0074] Partitioning blocks refer to the division of target medical image data not based on image information. For example, the processing device can directly cut multiple data points of the target medical image data into a preset number of blocks (for example, 4 blocks, 9 blocks, 16 blocks). For another example, the processing device can divide the target medical image data into several blocks in sequence based on a preset data block size (for example, 2×2). In this method, there may be several data blocks remaining at the boundary that cannot be divided into the preset data block size. The remaining data blocks can be directly divided into one data block.

[0075] By dividing or partitioning multiple data points of the target medical image data to obtain multiple data blocks, and forming a target vector with multiple data points contained in the data blocks, the data points in the target vector can be made more similar, which is conducive to subsequent fitting to obtain more accurate noise reduction parameters and thus obtain better noise reduction effects.

[0076] Step 322 , based on the positions of the multiple data points, determine the data points in the remaining frames of medical image data that have the same positions as the multiple data points, and assemble the data points of the remaining frames of medical image data into a set matrix. In some embodiments, step 322 is performed by the second assembly unit 222 .

[0077] The location of a data point refers to the location of the data point in the target medical image data. Figure 4 As shown, data point S j,k The position of the j-th frame target medical image data S j The fourth one in the third row.

[0078] The remaining frames of medical image data refer to the data other than the target medical image data in the multiple frames of medical image data. Figure 4 As shown, the processing device may process the target medical image data S except for the jth frame j Medical image data S1...S of other frames other than j-1 , S j+1 ……S N As the remaining frame medical image data (when j=1, there is no S j-1 ; When j = N, there is no S j+1 ).

[0079] The data points in the remaining frames of medical image data having the same position as the multiple data points refer to all the data points S in the remaining frames of medical image data having the same position as the j-th frame of target medical image data. j,1 ……S j,k ……S j,M For example, Figure 4 As shown, the kth data point S in the remaining frames of medical image data is related to the jth frame of target medical image data. j,k The data point with the same position is Si,k , the value of i is from 1 to N and i is not equal to j, the value of k is from 1 to M, that is, the kth data point S in the remaining frames of medical image data that is the same as the jth frame of target medical image data j,k The data points with the same position (black square) are S 1, ……S j-1,k , S j+1,k ……S N,k (Twill small squares).

[0080] The collection matrix refers to the matrix composed of the data points of the remaining frames of medical image data. Figure 4 As shown, the set matrix can be composed of the remaining frame medical image data S 1, ……S j-1,k , S j+1,k ……S N,k The matrix composed of.

[0081] In some embodiments, the processing device can determine, for each of the multiple data points, based on the position of the data point, the data points in the remaining frames of medical image data that have the same position as the data point, and arrange them in sequence horizontally to form row vectors; then arrange the multiple row vectors corresponding to the multiple data points in sequence vertically to form a set matrix, and the arrangement order of the multiple row vectors corresponds to the arrangement order of the multiple data points in the target vector.

[0082] For example, Figure 4 As shown, the processing device may combine the kth data point S in the remaining frames of medical image data with the jth frame of target medical image data. j,k The data point with the same position is S 1, ……S j-1,k , S j+1,k ……S N,k Arrange them horizontally to form a row vector, denoted as A k : A k =[S 1, ,…,S j-1,k ,S j+1,k ,…,S N,k ]

[0083] Then, the processing device can convert all row vectors A k (the value of k ranges from 1 to M) are arranged vertically to form a data set matrix, denoted as A:

[0084] Step 323 , based on the relationship between the target vector and the set matrix, fitting the noise reduction parameters corresponding to the target vector. In some embodiments, step 323 is performed by the fitting unit 223 .

[0085] The denoising parameter refers to a parameter used to denoise the target medical image. For example, the denoising parameter can be represented by w.

[0086] In some embodiments, the processing device may fit the noise reduction parameters corresponding to the target vector based on the relationship between the target vector and the set matrix. Figure 4 As shown, the processing device can linearly fit the noise reduction parameter w corresponding to the target vector based on the relationship between the target vector w and the set matrix b. That is, the noise reduction parameter w is solved by formula (1).

[0087] Among them, A represents the collection matrix, w represents the noise reduction parameter, and b represents the target vector.

[0088] Since the real data of the multi-frame medical image data are the same or satisfy the first characteristic, the real data of the set matrix and the real data of the target vector are the same or have correlation, that is, the relationship between the real data of the set matrix and the real data of the target vector can satisfy A1·w=b1, where A1 represents the real data of the set matrix and b1 represents the real data of the target vector. However, the noise of the multi-frame medical image data is randomly distributed and independent of each other, and there is no specific correlation, that is, the relationship between the noise of the set matrix and the noise of the target vector cannot satisfy A2·w=b2, where A2 represents the noise of the set matrix and b2 represents the noise of the target vector. Therefore, the sameness or correlation between the real data of the multi-frame medical image data plays a decisive role in the result of formula (1) (that is, the noise reduction parameter w), while the noise of the multi-frame medical image data does not affect the result of formula (1) (that is, the noise reduction parameter w).

[0089] Step 324 , denoising the target medical image data based on the denoising parameters, and determining updated medical image data of the target medical image data. In some embodiments, step 324 is performed by the determining unit 224 .

[0090] The updated medical image data refers to the target medical image data after noise reduction.

[0091] In some embodiments, the processing device may perform noise reduction on the target medical image data based on the noise reduction parameters to determine updated medical image data of the target medical image data. Figure 4 As shown, after obtaining the noise reduction parameter w, the processing device can determine to update the medical image data based on formula (2): b′=A·w (2)

[0092] Wherein, b′ represents the updated medical image data, A represents the set matrix, and w represents the noise reduction parameter.

[0093] Traditional image-based denoising methods use different filter functions to post-process the generated images. The processing is performed on the assumption that adjacent pixels of the target image data are similar (for example, the average value of a certain data point in the target image data and the surrounding data points is used as the filtered data point). In fact, there is not necessarily a specific relationship between adjacent pixels of the target image data. Therefore, traditional image-based denoising methods will cause blurring of medical images while reducing medical image noise. However, there is indeed a specific relationship between the real data of the multiple frames of medical image data of the present application (the same or satisfying the first feature). Therefore, the denoising method of the present application will not cause blurring of medical images while reducing medical image noise.

[0094] Figure 6 It is a comparison diagram of medical images before and after noise reduction according to some embodiments of this specification. Among them, the first row is the target medical image, the second row is the updated target medical image, the third row is the difference between the target medical image and the updated target medical image (that is, the image reconstructed from the noise), the first column is the first phase, the second column is the second phase, the third column is the third phase, the fourth column is the fourth phase, and the fifth column is the fifth phase. Figure 6 It can be seen that the updated target medical images in the second row have a higher signal-to-noise ratio than the target medical images in the first row. Among them, the signal strength of the fifth phase (i.e., the fifth column) is low, and the difference in signal-to-noise ratio between the target medical image and the updated target medical image is the largest, that is, it can be clearly seen that the edge of the target medical image is not clear, while the edge of the updated target medical image is clear. This is because the noise is random, and the signal strength of the noise has no obvious difference between multiple frames of medical image data. Therefore, the lower the signal strength of the real data, the greater the proportion of noise, and the greater the impact on the medical image.

[0095] By determining at least one frame of target medical image data from multiple frames of medical image data and performing noise reduction on each frame of the at least one frame of target medical image data, there is no need to input a large amount of additional training data, and the signal-to-noise ratio of the medical image can be significantly improved while ensuring the clarity of the medical image.

[0064] It should be noted that the above description of denoising of process medical images is only for example and explanation, and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the denoising of process medical images under the guidance of this specification. However, these modifications and changes are still within the scope of this specification. In some embodiments, denoising of medical images can also include taking multiple frames of medical image data as samples, taking updated medical image data as labels, training a machine learning model, and the machine learning model is used to denoise subsequent medical image data. For more details about machine learning models, see Figure 7and its related description.

[0096] Figure 7 7 is a schematic diagram of an exemplary process 700 for denoising a medical image using a machine learning model according to some embodiments of the present specification. In some embodiments, the process of generating updated medical image data 730 using a machine learning model 720 may be performed by the processing device 120. In some embodiments, the process of training the initial model 750 to obtain the machine learning model 720 may be performed by the processing device 120 or by other processing devices.

[0097] like Figure 7 As shown, the processing device 120 can input the medical image data 710 into the machine learning model 720, and the machine learning model 720 can output corresponding updated medical image data 730.

[0098] In some embodiments, the machine learning model 720 may be a deep learning neural network model. Exemplary deep learning neural network models may include convolutional neural networks (CNN), deep neural networks (DNN), recurrent neural networks (RNN), etc. or a combination thereof.

[0099] In some embodiments, Figure 7 As shown, the model parameters of the machine learning model 720 can be obtained by training the initial model 750 using multiple training samples. For example, multiple training samples can be obtained. Each training sample can include sample medical image data 740-1 and training labels 740-2. The sample medical image data 740-1 is medical image data repeatedly collected at the same delay time and / or medical image data collected at different delay times. The training label 740-2 is updated medical image data corresponding to the sample medical image data 740-1. In some embodiments, the processing device can be based on the sample medical image data 740-1, through Figure 3 The updated medical image data obtained by the method is used as the training label 740-2.

[0100] The training of the initial model 750 may include one or more iterations. As an example only, in the current iteration, for each training sample, the processing device 120 may determine the updated medical image data of the sample medical image data 740-1 using the intermediate model. If the current iteration is the first iteration, the intermediate model may be the initial model 750. If the current iteration is another iteration, the intermediate model may be the model updated in the previous iteration. The processing device 120 may determine the value of the loss function based on the accuracy of the updated medical image data of the sample medical image data 740-1, and update the intermediate model based on the value of the loss function. For example, the value of the loss function may be determined based on the difference between the updated medical image data of the sample medical image data 740-1 and the training label 740-2, and the intermediate model may be updated based on the value of the loss function.

[0101] In some embodiments, the parameters of the initial model 750 can be iteratively updated based on multiple training samples so that the loss function of the intermediate model meets the preset conditions. For example, the loss function converges, or the loss function value is less than the preset value. When the loss function meets the preset conditions, the model training is completed, and the trained initial model 750 can be used as the machine learning model 720.

[0102] By taking multiple frames of medical image data as samples, Figure 3 The updated medical image data obtained by the method is used as a label to train the machine learning model. The machine learning model can be used to quickly and accurately determine the updated medical image data of subsequent medical image data, while ensuring a high signal-to-noise ratio of the medical image and improving efficiency.

[0103] In some embodiments, a medical image denoising apparatus comprises a processing device and a memory; the memory is used to store instructions, and when the instructions are executed by the processing device, the apparatus implements the medical image denoising method.

[0104] In some embodiments, a computer-readable storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the method for reducing noise of medical images.

[0105] The beneficial effects that may be brought about by the embodiments of the present specification include but are not limited to: (1) by determining at least one frame of target medical image data from multiple frames of medical image data and performing noise reduction on each frame of the at least one frame of target medical image data, there is no need to input a large amount of additional training data, and the signal-to-noise ratio of the medical image can be significantly improved while ensuring the clarity of the medical image; (2) by dividing or partitioning multiple data points of the target medical image data into blocks to obtain multiple data blocks, and forming a target vector with multiple data points contained in the data blocks, the data points in the target vector can be made more similar, which is conducive to subsequent fitting to obtain more accurate noise reduction parameters, thereby obtaining better noise reduction effects; (3) using this machine learning model to quickly and accurately determine the updated medical image data of subsequent medical image data, while ensuring a high signal-to-noise ratio of the medical image, the efficiency is improved.

[0106] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements and corrections to this specification. Such modifications, improvements and corrections are suggested in this specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of this specification.

[0107] At the same time, this specification uses specific words to describe the embodiments of this specification. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of this specification. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures or characteristics in one or more embodiments of this specification can be appropriately combined.

[0108] In addition, unless explicitly stated in the claims, the order of the processing elements and sequences described in this specification, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this specification. Although the above disclosure discusses some invention embodiments that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0109] Similarly, it should be noted that in order to simplify the description disclosed in this specification and thus help understand one or more embodiments of the invention, in the above description of the embodiments of this specification, multiple features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the subject matter of this specification are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the single embodiment disclosed above.

[0110] In some embodiments, numbers describing the number of components and attributes are used. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise specified, "about", "approximately" or "substantially" indicate that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may change according to the required features of individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and adopt the general method of retaining digits. Although the numerical domains and parameters used to confirm the breadth of their range in some embodiments of this specification are approximate values, in specific embodiments, the setting of such numerical values ​​is as accurate as possible within the feasible range.

[0111] Each patent, patent application, patent application publication, and other materials, such as articles, books, specifications, publications, documents, etc., cited in this specification are hereby incorporated by reference in their entirety. Except for application history documents that are inconsistent with or conflicting with the contents of this specification, documents that limit the broadest scope of the claims of this specification (currently or later attached to this specification) are also excluded. It should be noted that if the descriptions, definitions, and / or use of terms in the materials attached to this specification are inconsistent or conflicting with the contents described in this specification, the descriptions, definitions, and / or use of terms in this specification shall prevail.

[0112] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A method for denoising medical images, It is characterized in that The method comprises the following steps: Acquire multiple frames of medical image data, and determine at least one frame of target medical image data from the multiple frames of medical image data; Noise reduction is performed on each frame of the at least one frame of target medical image data, wherein the noise reduction on each frame of the target medical image data comprises the following steps: Combining a plurality of data points of the target medical image data into a target vector; Based on the positions of the multiple data points, determine the data points in the remaining frames of medical image data that have the same positions as the multiple data points, and form a set matrix with the data points in the remaining frames of medical image data, wherein the remaining frames of medical image data are the medical image data of the remaining frames in the multiple frames of medical image data except the target medical image data; Based on the relationship between the target vector and the set matrix, fitting the noise reduction parameter corresponding to the target vector; The target medical image data is denoised based on the denoising parameters to determine updated medical image data of the target medical image data.

2. The method according to claim 1, It is characterized in that Determining at least one frame of target medical image data from the multiple frames of medical image data comprises: The multiple frames of medical image data are determined as the target medical image data to perform noise reduction on the multiple frames of medical image data.

3. The method according to claim 1, It is characterized in that The step of forming a target vector from a plurality of data points of the target medical image data comprises: Segmenting or partitioning the multiple data points of the target medical image data to obtain multiple data blocks, each of which includes multiple data points; For each of the multiple data blocks, multiple data points included in the data block are combined into the target vector.

4. The method according to claim 1, It is characterized in that The step of forming a target vector from a plurality of data points of the target medical image data comprises: Arrange the multiple data points in sequence vertically to form a column vector, where the column vector is the target vector; The step of forming a set matrix of the data points of the remaining frames of medical image data comprises: For each of the plurality of data points, based on the position of the data point, determine the data points in the remaining frames of medical image data that have the same position as the data point, and arrange them in sequence in a horizontal manner to form a row vector; The multiple row vectors corresponding to the multiple data points are sequentially arranged vertically to form the set matrix, and the arrangement order of the multiple row vectors corresponds to the arrangement order of the multiple data points in the target vector.

5. The method according to claim 1, It is characterized in that The medical image data consists of real data and noise, the real data of the multiple frames of medical image data are the same or meet the first characteristic, and the noise of the multiple frames of medical image data meet the second characteristic.

6. A medical image denoising system, It is characterized in that include: A determination module, configured to acquire multiple frames of medical image data and determine at least one frame of target medical image data from the multiple frames of medical image data; A denoising module, configured to perform denoising on each frame of the at least one frame of target medical image data, the denoising module further comprising: A first composition unit, configured to compose a target vector from a plurality of data points of the target medical image data; a second composition unit, configured to determine, based on the positions of the multiple data points, data points in remaining frames of medical image data having the same positions as the multiple data points, and to form a set matrix with the data points in the remaining frames of medical image data, wherein the remaining frames of medical image data are medical image data of remaining frames in the multiple frames of medical image data except the target medical image data; A fitting unit, configured to fit a noise reduction parameter corresponding to the target vector based on a relationship between the target vector and the set matrix; A determining unit is configured to perform noise reduction on the target medical image data based on the noise reduction parameters to determine updated medical image data of the target medical image data.

7. The system of claim 6, It is characterized in that The determination module is further used for: The multiple frames of medical image data are determined as the target medical image data to perform noise reduction on the multiple frames of medical image data.

8. The system according to claim 6, It is characterized in that The noise reduction module is further used for: Segmenting or partitioning the multiple data points of the target medical image data to obtain multiple data blocks, each of which includes multiple data points; For each of the multiple data blocks, multiple data points included in the data block are combined into the target vector.

9. A medical image denoising device, comprising a processing device, It is characterized in that The processing device is used to execute the medical image denoising method according to any one of claims 1 to 5.

10. A computer-readable storage medium, wherein the storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the medical image denoising method according to any one of claims 1 to 5.