Method for optimizing bone scanning image and related device

Abstract

The embodiment of the invention provides a method for optimizing a bone scanning image and a related device, which are used for improving the examination efficiency of the bone scanning image. The method comprises the steps that a to-be-optimized bone scanning image is obtained, and the to-be-optimized bone scanning image comprises an accelerated bone scanning image obtained when an object executes accelerated bone scanning or a low-dosage bone scanning image obtained when the object injects a low medicine dosage; a to-be-optimized bone scanning image is input into the accelerated image optimization model or the low-dose image optimization model to obtain a bone scanning image generated by the accelerated image optimization model or the low-dose image optimization model, the generated bone scanning image comprises a standard speed bone scanning image obtained at a standard bone scanning speed, and the standard speed bone scanning image is a standard speed bone scanning image obtained at a standard bone scanning speed. Or a standard dose bone scanning image obtained by scanning according to a standard drug dose; and converting non-integer pixel points in the generated bone scanning image into integer pixel points to obtain an optimized bone scanning image.

Description

Technical Field

[0001] The present invention relates to the technical field of whole-body bone scan image processing, and in particular to a method for optimizing bone scan images and related devices. Background Art

[0002] Cancer research and curing cancer patients are issues that governments around the world are currently paying great attention to. Bone metastasis is one of the most serious hazards of malignant tumors and the main cause of clinical death in cancer patients. Bone metastasis is a common disease in advanced tumors. Many types of malignant tumors may metastasize to the bone. Once bone metastasis occurs, most patients cannot be effectively treated.

[0003] With the development and progress of medicine, bone scanning has become one of the main examination items for bone tumor examination. So far, regardless of the development of various imaging technologies, including the emergence of PET / CT functional imaging, whole-body bone imaging still plays an irreplaceable role in the screening of bone metastasis of malignant tumors. For patients with malignant tumors, regular follow-up of whole-body bone scanning can detect bone metastasis early, 3-6 months earlier than X-ray bone films.

[0004] In order to meet the image quality standards required for diagnosis, whole-body bone scans require a long scan time, which greatly reduces the efficiency of hospital equipment use. Although there are many patents for the optimization of medical images such as fast MRI and PET, compared with the optimization technologies of MRI and PET, bone scans, as a planar scanning technology, cannot utilize the spatial correlation between layers, and there is no K-space raw data that retains the characteristics of the entire image. Therefore, the goal of improving the image quality of bone scans can only be achieved by extending the scan time.

[0005] Therefore, speeding up the efficiency of whole-body bone scanning imaging examinations is a technical problem that urgently needs to be solved. Summary of the invention

[0006] The embodiment of the present invention provides a method and a related device for optimizing bone scan images, which are used to improve the inspection efficiency of bone scan images.

[0007] A first aspect of an embodiment of the present application provides a method for optimizing a bone scan image, comprising:

[0008] Acquiring a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan;

[0009] Inputting the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose;

[0010] The non-integer pixels in the generated bone scan image are converted into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical features.

[0011] Preferably, converting non-integer pixels in the generated bone scan image into integer pixels comprises:

[0012] Acquire all pixel points in the generated bone scan image;

[0013] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0014] If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X, if Y is less than or equal to the decimal part of X, X is adjusted to a minimum positive integer greater than X, if Y is greater than the decimal part of X, X is adjusted to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixel points.

[0015] Preferably, converting non-integer pixels in the generated bone scan image into integer pixels comprises:

[0016] Acquire all pixel points in the generated bone scan image;

[0017] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0018] If the pixel value of the target pixel point is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle, wherein the target pixel point is any pixel point among all the pixel points.

[0019] Preferably, the accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model includes a convolutional neural network or a generative adversarial network;

[0020] The convolutional neural network includes a backbone optimized network structure.

[0021] Preferably, before converting the non-integer pixels in the generated bone scan image into integer pixels, the method further comprises:

[0022] The generated bone scan image is dot-multiplied with the acceleration factor matrix to obtain a quantitatively corrected bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the distribution of the pixel matrix of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value.

[0023] Preferably, the convolutional neural network also includes a quantitative correction network structure and a first convolution module, and the quantitative correction network structure includes a second convolution module and a third convolution module.

[0024] Preferably, the step of inputting the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model comprises:

[0025] Inputting the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtaining a first feature three-dimensional matrix of the bone scan image to be optimized;

[0026] The first characteristic three-dimensional matrix is ​​superimposed with the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values;

[0027] Inputting the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix;

[0028] Inputting the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized;

[0029] Multiply the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix;

[0030] The fourth characteristic three-dimensional matrix is ​​input into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0031] Preferably, the loss function of the accelerated image optimization model or the low-dose image optimization model includes:

[0032] At least one of an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

[0033] Preferably, before inputting the bone scan image to be optimized into the accelerated image optimization model or the low-dose image optimization model, the method further comprises:

[0034] Acquire first training sample data, and use the first training sample data to train the accelerated image optimization model until the loss of the accelerated image optimization model meets the first target loss, wherein the first training sample data includes: accelerated bone scan images and standard bone scan images of the same morphological lesions of each subject in a plurality of subjects when the same drug dose is injected and the bone scan speeds are respectively the accelerated bone scan speed and the standard bone scan speed, and the standard bone scan speed is less than 40 cm / min;

[0035] Acquire second training sample data, and use the second training sample data to train the low-dose image optimization model until the loss of the low-dose image optimization model meets the second target loss, wherein the second training sample data includes: low-dose bone scan images and standard drug dose bone scan images with the same morphological lesions acquired at the same scanning speed when each of the multiple subjects is injected with a low drug dose and a standard drug dose.

[0036] Preferably, acquiring an accelerated bone scan image and a standard speed bone scan image of each subject comprises:

[0037] acquiring a standard speed bone scan image for each subject at a standard bone scan speed;

[0038] performing Poisson downsampling on the standard speed bone scan image to obtain a corresponding accelerated bone scan image;

[0039] Low-dose bone scan images and standard drug-dose bone scan images were obtained for each subject, including:

[0040] Obtaining standard drug dose bone scan images for each subject when a standard drug dose is infused;

[0041] Poisson downsampling was performed on the standard drug dose bone scan images to obtain the corresponding low dose bone scan images.

[0042] Preferably, performing Poisson downsampling on the standard speed bone scan image or the standard drug dose bone scan image comprises:

[0043] Get the downsampling multiple u, the 0 <u<1;

[0044] Obtaining a pixel value n of each pixel position (x, y) in a standard speed bone scan image or a standard drug dose bone scan image;

[0045] Acquire a first image template corresponding to a standard speed bone scan image or a standard drug dose bone scan image, wherein the image size of the first image template is the same as that of the standard speed bone scan image or the standard drug dose bone scan image, but pixel values ​​at corresponding positions in the first image template are all 0;

[0046] For the pixel value of the target pixel position in the first image template, a number σ is randomly obtained with equal probability, and if σ≤u, the pixel value of the target pixel position is increased by 1, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template;

[0047] For the pixel value 0 of each pixel position in the first image template, repeat the first step n times until the pixel value of each pixel position is traversed, and the first step includes: for the pixel value of the target pixel position in the first image template, randomly and equally likely obtain a number σ, if the σ≤u, then add 1 to the pixel value of the target pixel position, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template.

[0048] Before performing Poisson downsampling on the standard speed bone scan image or the standard drug dose bone scan image, the method further includes:

[0049] A filtering algorithm is used to perform denoising filtering on a standard speed bone scan image or a standard drug dose bone scan image, wherein the filtering algorithm includes at least one of a self-supervised learning algorithm, a Gaussian filtering algorithm, a median filtering algorithm, a bilateral filtering algorithm, an image denoising algorithm based on a Markov expert field, and a three-dimensional block matching algorithm.

[0050] Preferably, using the self-supervised learning algorithm to perform denoising filtering on a standard speed bone scan image or a standard drug dose bone scan image comprises:

[0051] After setting the pixel value of a pixel point in the standard speed bone scan image or the standard drug dose bone scan image to zero each time, a plurality of processed standard speed bone scan images or a plurality of standard drug dose bone scan images are obtained;

[0052] The processed multiple standard speed bone scan images or multiple standard drug dose bone scan images are input into a pre-trained self-supervised neural network model to obtain filtered standard speed bone scan images or standard drug dose bone scan images.

[0053] Preferably, before using the first training sample data to train the accelerated image optimization model, the method further includes:

[0054] A registration algorithm is used to register the position coordinates of the same part images in the standard speed bone scan image and the accelerated bone scan image;

[0055] Before training the low-dose image optimization model using the second training sample data, the method further includes:

[0056] The registration algorithm is used to register the position coordinates of the same part images in the standard drug dose bone scan image and the low drug dose scan image.

[0057] A second aspect of an embodiment of the present application provides a device for optimizing a bone scan image, characterized in that it includes:

[0058] an acquisition unit, configured to acquire a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan;

[0059] a generating unit, configured to input the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose;

[0060] The optimization unit is used to convert non-integer pixels in the generated bone scan image into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical characteristics.

[0061] Preferably, the optimization unit is specifically used for:

[0062] Acquire all pixel points in the generated bone scan image;

[0063] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0064] If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X, if Y is less than or equal to the decimal part of X, X is adjusted to a minimum positive integer greater than X, if Y is greater than the decimal part of X, X is adjusted to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixel points.

[0065] Preferably, the optimization unit is specifically used for:

[0066] Acquire all pixel points in the generated bone scan image;

[0067] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0068] If the pixel value of the target pixel point is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle, wherein the target pixel point is any pixel point among all the pixel points.

[0069] Preferably, the accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model includes a convolutional neural network or a generative adversarial network;

[0070] The convolutional neural network includes a backbone optimized network structure.

[0071] Preferably, the device further comprises:

[0072] A correction unit is used to perform a dot multiplication of the generated bone scan image and an acceleration factor matrix before converting non-integer pixels in the generated bone scan image into integer pixels to obtain a quantitatively corrected bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the distribution of the pixel matrix of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value.

[0073] Preferably, the convolutional neural network also includes a quantitative correction network structure and a first convolution module, and the quantitative correction network structure includes a second convolution module and a third convolution module.

[0074] Preferably, the generating unit is specifically used for:

[0075] Inputting the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtaining a first feature three-dimensional matrix of the bone scan image to be optimized;

[0076] The first characteristic three-dimensional matrix is ​​superimposed with the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values;

[0077] Inputting the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix;

[0078] Inputting the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized;

[0079] Multiply the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix;

[0080] The fourth characteristic three-dimensional matrix is ​​input into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0081] Preferably, the loss function of the accelerated image optimization model or the low-dose image optimization model includes:

[0082] At least one of an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

[0083] Preferably, the device further comprises:

[0084] A first training unit is used to obtain first training sample data, and train the accelerated image optimization model using the first training sample data until the loss of the accelerated image optimization model meets the first target loss, wherein the first training sample data includes: an accelerated bone scan image and a standard bone scan image of the same morphological lesion when each of the multiple subjects is injected with the same drug dose and the bone scan speeds are the accelerated bone scan speed and the standard bone scan speed, respectively, and the standard bone scan speed is less than 40 cm / min;

[0085] The second training unit is used to obtain second training sample data and use the second training sample data to train the low-dose image optimization model until the loss of the low-dose image optimization model meets the second target loss. The second training sample data includes: low-dose bone scan images and standard drug dose bone scan images with the same morphological lesions obtained at the same scanning speed when each object in multiple objects is injected with a low drug dose and a standard drug dose.

[0086] The first training unit is specifically used for:

[0087] acquiring a standard speed bone scan image for each subject at a standard bone scan speed;

[0088] performing Poisson downsampling on the standard speed bone scan image to obtain a corresponding accelerated bone scan image;

[0089] The second training unit is specifically used for:

[0090] Obtaining standard drug dose bone scan images for each subject when a standard drug dose is infused;

[0091] Poisson downsampling was performed on the standard drug dose bone scan images to obtain the corresponding low dose bone scan images.

[0092] Preferably, the first training unit or the second training unit is specifically used for:

[0093] Get the downsampling multiple u, the 0 <u<1;

[0094] Obtaining a pixel value n of each pixel position (x, y) in a standard speed bone scan image or a standard drug dose bone scan image;

[0095] Acquire a first image template corresponding to a standard speed bone scan image or a standard drug dose bone scan image, wherein the image size of the first image template is the same as that of the standard speed bone scan image or the standard drug dose bone scan image, but pixel values ​​at corresponding positions in the first image template are all 0;

[0096] For the pixel value of the target pixel position in the first image template, a number σ is randomly obtained with equal probability, and if σ≤u, the pixel value of the target pixel position is increased by 1, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template;

[0097] For the pixel value 0 of each pixel position in the first image template, repeat the first step n times until the pixel value of each pixel position is traversed, and the first step includes: for the pixel value of the target pixel position in the first image template, randomly and equally likely obtain a number σ, if the σ≤u, then add 1 to the pixel value of the target pixel position, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template.

[0098] Preferably, the device further comprises:

[0099] A denoising unit is used to: before performing Poisson downsampling on the standard speed bone scan image or the standard drug dose bone scan image, use a filtering algorithm to perform denoising filtering on the standard speed bone scan image or the standard drug dose bone scan image, wherein the filtering algorithm includes at least one of a self-supervised learning algorithm, a Gaussian filtering algorithm, a median filtering algorithm, a bilateral filtering algorithm, an image denoising algorithm based on a Markov expert field, and a three-dimensional block matching algorithm.

[0100] Preferably, the denoising unit is specifically used for:

[0101] After setting the pixel value of a pixel point in the standard speed bone scan image or the standard drug dose bone scan image to zero each time, a plurality of processed standard speed bone scan images or a plurality of standard drug dose bone scan images are obtained;

[0102] The processed multiple standard speed bone scan images or multiple standard drug dose bone scan images are input into a pre-trained self-supervised neural network model to obtain filtered standard speed bone scan images or standard drug dose bone scan images.

[0103] Preferably, the device further comprises:

[0104] A registration unit is used to register the position coordinates of the images of the same part in the standard speed bone scan image and the accelerated bone scan image using a registration algorithm before training the accelerated image optimization model using the first training sample data; or, before training the low-dose image optimization model using the second training sample data, register the position coordinates of the images of the same part in the standard drug dose bone scan image and the low drug dose scan image using a registration algorithm.

[0105] It can be seen from the above technical solutions that the embodiments of the present invention have the following advantages:

[0106] In an embodiment of the present application, a bone scan image to be optimized is obtained, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the object performs an accelerated bone scan, or a low-dose bone scan image when the object is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan; the bone scan image to be optimized is input into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose; non-integer pixels in the generated bone scan image are converted into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical characteristics.

[0107] Therefore, the embodiments of the present application can improve the efficiency of bone scanning imaging examinations, or reduce the damage to the body caused by drug dosage during bone scanning imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 A schematic diagram of an embodiment of a method for optimizing a bone scan image in an embodiment of the present application;

[0109] Figure 2 for Figure 1 A refinement of step 103 in the embodiment;

[0110] Figure 3 for Figure 1 Another refinement of step 103 in the embodiment;

[0111] Figure 4 This is a schematic diagram of the structure of an accelerated image optimization model or a low-dose image optimization model in an embodiment of the present application;

[0112] Figure 5 for Figure 1 A refinement of step 102 in the embodiment;

[0113] Figure 6 Schematic diagram of an embodiment of accelerating the training process of an image optimization model or a low-dose image optimization model in an embodiment of the present application;

[0114] Figure 7 A schematic diagram of an embodiment of a device for optimizing bone scan images in an embodiment of the present application. DETAILED DESCRIPTION

[0115] The embodiment of the present invention provides a method and a related device for optimizing bone scan images, which are used to improve the inspection efficiency of bone scan images.

[0116] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0117] The terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0118] For ease of understanding, the following first describes the method for optimizing the bone scan image in the embodiment of the present application. Figure 1 , an embodiment of the method for optimizing bone scan images in the embodiments of the present application includes:

[0119] 101. Acquire a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan;

[0120] For ease of understanding, the imaging principle of bone scanning is described first:

[0121] A bone scan requires that the subject be injected with a bone imaging agent, which is a radioactive drug, before the examination. The scan will be performed after the bones have fully absorbed the drug. If the radioactive absorption of a certain part of the bone is abnormal, it can prove that the bone metabolism is abnormal. When imaging the bone scan image, the imaging result not only depends on the scanning speed. If the scanning speed is slower, the bone absorption is more uniform and the imaging result is more accurate. At the same time, the imaging result of the bone scan image also depends on the amount of bone imaging agent injected. If the more bone imaging agent is injected, the more signals are detected and the more accurate the imaging result is.

[0122] Therefore, in order to improve the imaging quality of bone scan images, the traditional approach is to either extend the bone scan time or increase the amount of drug injected.

[0123] The embodiment of the present application proposes a method for optimizing bone scan images, which is used to optimize accelerated bone scan images or low-dose bone scan images, so as to obtain a standard speed bone scan image corresponding to a standard bone scan speed, or to obtain a standard dose bone scan image corresponding to a low-dose bone scan image, thereby improving the efficiency of bone imaging examinations or reducing the impact of drug doses on the body during bone scanning.

[0124] Before optimizing the bone scan image to be optimized, it is necessary to first obtain the bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the object performs an accelerated bone scan, or a low-dose bone scan image when the object is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan.

[0125] The standard bone scanning speed in this embodiment is less than 40 cm / min, and the standard drug dosage is the drug dosage required to be injected when a normal examination is performed on the subject.

[0126] The bone scan image to be optimized in this embodiment can be generated by scanning on the spot by a bone scanning device, or can be an image that has been scanned in advance and stored locally or by a third party. There is no specific limitation on the source of the image to be optimized.

[0127] 102. Input the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard speed bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose;

[0128] Because the signal amount detected by the bone scan image to be optimized (accelerated bone scan image or low-dose bone scan image) in the embodiment of the present application is relatively small, in order to increase the signal amount of the bone scan image, the embodiment of the present application inputs the bone scan image to be optimized into the accelerated image optimization model or the low-dose image optimization model for optimization to obtain a bone scan image generated by the model, wherein the generated bone scan image includes a standard speed bone scan image corresponding to the accelerated bone scan image, or a standard dose bone scan image corresponding to the low-dose bone scan image.

[0129] 103. Convert non-integer pixels in the generated bone scan image into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical features.

[0130] Furthermore, unlike CT images, because each pixel point in the bone scan image represents the number of detected signal quantities, and the number of signal quantities is generally an integer, the embodiment of the present application converts non-integer pixel points in the bone scan image generated by the accelerated image optimization model or the low-dose image optimization model into integer pixel points, so that the optimized bone scan image and the real bone scan image have similar physical characteristics.

[0131] Specifically, when converting non-integer pixels into integer pixels, different conversion algorithms make the converted bone scan image have different characteristics. The different characteristics of different conversion algorithms will be described in the following embodiments and will not be repeated here.

[0132] In an embodiment of the present application, the bone scan image to be optimized is input into an accelerated image optimization model or a low-dose image optimization model, so that the bone scan image output by the accelerated image optimization model or the low-dose image optimization model is a corresponding standard speed bone scan image, or a standard dose bone scan image, and the embodiment of the present application converts non-integer pixels in the bone scan image generated by different models into integer pixels, so that the optimized bone scan image is closer to the real bone scan image, thereby improving the quality of the optimized bone scan image on the one hand, and improving the efficiency of the scanning image examination on the other hand, or reducing the damage to the body caused by the drug dose during bone scanning imaging.

[0133] based on Figure 1 In the embodiment described above, step 103 is described in detail below. Figure 2 , Figure 2 The detailed steps of step 103 are as follows:

[0134] 201. Acquire all pixel points in the generated bone scan image;

[0135] In the embodiment of the present application, all pixel points in the generated bone scan image are first obtained. If the generated bone scan image includes X pixel points, the X pixel points are obtained.

[0136] 202. For a target pixel point, if the pixel value of the target pixel point is less than 0, adjust the pixel value of the target pixel point to 0;

[0137] For a target pixel point in the X pixel points, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0, wherein the target pixel point is any pixel point of all the pixel points.

[0138] 203. If the pixel value of the target pixel is a non-integer X greater than 0, generate a random number Y greater than 0 and less than 1 for X, and if Y is less than or equal to the decimal part of X, adjust X to a minimum positive integer greater than X; if Y is greater than the decimal part of X, adjust X to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixels.

[0139] If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X. If Y is less than or equal to the decimal part of X, X is adjusted to the smallest positive integer greater than X. If Y is greater than the decimal part of X, X is adjusted to the largest positive integer less than X.

[0140] The following example illustrates that, assuming the pixel value of the target pixel is 245.2, a random number of 0.4 is generated for 245.2. Since 0.4 is greater than 0.2 (the decimal part of 245.2), 245.2 is adjusted to 245. If a random number of 0.1 is generated, since 0.1 is less than 0.2 (the decimal part of 245.2), 245.2 is adjusted to 246.

[0141] Because the embodiment of the present application uses a random probability method to adjust the pixel value of the target pixel point in the process of adjusting the non-integer pixel point of the target pixel point to an integer pixel point, the random probability method is used. Figure 2 The adjustment method in the embodiment can make the optimized bone scan image have physical characteristics similar to those of the real bone scan image.

[0142] based on Figure 1 In the embodiment described above, step 103 is described in detail below. Figure 3 , Figure 3 Another refinement of step 103:

[0143] 301. Acquire all pixel points in the generated bone scan image;

[0144] In the embodiment of the present application, all pixel points in the generated bone scan image are first obtained. If the generated bone scan image includes Y pixel points, the Y pixel points are obtained.

[0145] 302. For a target pixel point, if the pixel value of the target pixel point is less than 0, adjust the pixel value of the target pixel point to 0;

[0146] For a target pixel point among the Y pixel points, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0, wherein the target pixel point is any pixel point of all the pixel points.

[0147] 303. If the pixel value of the target pixel is a non-integer greater than 0, adjust the non-integer greater than 0 to an integer according to the rounding principle, wherein the target pixel is any pixel among all the pixel points.

[0148] If the pixel value of the target pixel is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle.

[0149] The following example illustrates that, assuming that the pixel value of the target pixel is 245.2, then according to the rounding principle, 245.2 is adjusted to 245; if the pixel value of the target pixel is 245.6, then according to the rounding principle, 245.2 is adjusted to 246.

[0150] The present application adopts the principle of rounding to adjust the pixel value of the target pixel point, so that the edge features of the optimized bone scan image are smoother and more continuous.

[0151] against Figure 1 In the described embodiment, the accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model also includes a convolutional neural network or a generative adversarial network; and the convolutional neural network includes a backbone optimization network structure. The following describes the process of generating a bone scan image when the accelerated image optimization model or the low-dose image optimization model only includes a backbone optimization network structure, wherein the backbone optimization network structure includes a UNet network structure, an RRDBNet network structure, or a Transformer network structure. Because when performing a bone scan, the scan is performed at an accelerated speed, after the bone scan image is generated, it is necessary to perform quantitative correction on the generated bone scan image, that is, before performing step 103, it is necessary to perform a dot multiplication of the generated bone scan image and the acceleration factor matrix to perform quantitative correction on the generated bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the pixel matrix distribution of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value. For example, if the pixel matrix of the generated bone scan image is 256*256, the acceleration factor matrix is ​​also 256*256, but the value of the acceleration factor matrix is ​​a preset acceleration factor value, such as 3 or 2, etc., and no specific limitation is made here.

[0152] against Figure 1 In the embodiment described, if the convolutional neural network of the accelerated image optimization model or the low-dose image optimization model includes a backbone optimization network structure, a quantitative correction network structure (wherein the quantitative correction network structure includes a second convolution module and a third convolution module) and a first convolution module, the process of generating a bone scan image by the accelerated image optimization model or the low-dose image optimization model is described for ease of understanding. Figure 4The model structure of the accelerated image optimization model or the low-dose image optimization model is given. Figure 5 The process of generating bone scan images with accelerated image optimization model or low-dose image optimization model is given:

[0153] 501. Input the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtain a first feature three-dimensional matrix of the bone scan image to be optimized;

[0154] Because when a subject is subjected to a bone scan by a bone scanning device, the bone scanning device is set with an acceleration factor, so that the bone scanning device can only obtain a sparse bone scanning image, the embodiment of the present application adds a quantitative correction network structure to the image optimization model structure to perform quantitative correction on the obtained bone scan image to be optimized, so as to restore the sparse bone scan image to a bone scan image with a denser number of pixels.

[0155] Therefore, the embodiment of the present application performs a calculation on the bone scan image to be optimized and the second convolution module to extract the first image feature of the bone scan image to be optimized, and obtains the first feature three-dimensional matrix of the bone scan image to be optimized.

[0156] 502. Superimpose the first characteristic three-dimensional matrix and the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values;

[0157] Furthermore, the embodiment of the present application superimposes the first characteristic three-dimensional matrix and the acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the distribution of the acceleration factor characteristic matrix is ​​the same as the distribution of the first characteristic three-dimensional matrix on two dimensions other than the target dimension, and the pixel values ​​in the acceleration factor characteristic matrix are preset acceleration factor values.

[0158] For example, when the first characteristic three-dimensional matrix is ​​256*256*512, assuming that the acceleration factor characteristic matrix is ​​superimposed on the first characteristic three-dimensional matrix in the Z-axis direction, the size of the acceleration factor characteristic matrix in the X-axis and Y-axis directions is 256*256. If the acceleration factor is 2 when the bone scanning device generates an accelerated bone scanning image, the corresponding pixel value in the acceleration factor characteristic map is set to 2.

[0159] 503. Input the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix;

[0160] After obtaining the second characteristic three-dimensional matrix (that is, after obtaining multiple bone scan feature images to be optimized), in order to highlight the background noise in the bone scan feature image to be optimized, the normal tissue and the lesion tissue in the foreground feature differently, such as weakening the Beijing noise in the bone scan feature image to be optimized, performing a first enhancement on the normal tissue in the bone scan feature image to be optimized, and performing a second enhancement on the lesion tissue in the bone scan feature image to be optimized, the embodiment of the present application inputs the second characteristic three-dimensional matrix into the third convolution module for convolution operation to obtain a quantitative correction weight matrix.

[0161] 504. Input the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized;

[0162] Furthermore, the embodiment of the present application also inputs the bone scan image to be optimized into the backbone optimization network structure to extract multiple third feature maps of the bone scan image to be optimized, that is, to obtain the third feature three-dimensional matrix of the bone scan image to be optimized.

[0163] 505. Perform a dot multiplication of the quantitative correction weight matrix and the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix;

[0164] Furthermore, the present application multiplies the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix, that is, to obtain multiple fourth characteristic maps.

[0165] 506. Input the fourth characteristic three-dimensional matrix into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0166] In order to synthesize multiple fourth feature maps into one image, the embodiment of the present application further inputs the fourth feature map into the first convolution module, that is, inputs the fourth feature three-dimensional matrix into the first convolution module, so as to synthesize the multiple fourth feature maps into one image, and the image is the bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0167] Because a quantitative correction network structure is added to the accelerated image optimization model or the low-dose image optimization model in the embodiments of the present application, the neural network model structure in the embodiments of the present application can directly output the corrected bone scan image when used, without having to input the bone scan image generated by the accelerated image optimization model or the low-dose image optimization model into the quantitative correction module for quantitative correction, thereby improving the convenience of obtaining the bone scan image.

[0168] Furthermore, a quantitative correction network structure is added to the accelerated image optimization model or the low-dose image optimization model in the embodiments of the present application, and the quantitative correction network structure can weaken the background noise of the bone scan image and enhance the foreground features of the bone scan image to varying degrees, thereby further improving the clarity and image quality of the optimized bone scan image.

[0169] It is easy to understand that before executing step 101, the accelerated image optimization model or the low-dose image optimization model needs to be trained. The training process of the accelerated image optimization model and the low-dose image optimization model is described below. Figure 6 , an embodiment of accelerating the training process of an image optimization model and a low-dose image optimization model, comprising:

[0170] 601. Acquire first training sample data, and use the first training sample data to train the accelerated image optimization model until the loss of the accelerated image optimization model meets the first target loss, wherein the first training sample data includes: accelerated bone scan images and standard bone scan images of each subject in a plurality of subjects with the same morphological lesions when the same drug dose is injected and the bone scan speeds are respectively the accelerated bone scan speed and the standard bone scan speed, and the standard bone scan speed is less than 40 cm / min;

[0171] It is easy to understand that before training the accelerated image optimization model, it is necessary to first obtain first training sample data, and use the first training sample data to train the accelerated image optimization model until the loss of the accelerated image optimization model meets the first target loss, wherein the first training sample data includes: accelerated bone scan images and standard bone scan images of the same morphological lesions when each of the multiple subjects is injected with the same drug dose and the bone scan speeds are the accelerated bone scan speed and the standard bone scan speed, respectively, and the standard bone scan speed is less than 40 cm / min;

[0172] Furthermore, the loss function of the accelerated image optimization model in the embodiment of the present application includes: at least one of: an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

[0173] When acquiring the accelerated bone scan image and the standard speed bone scan image of each object, the standard speed bone scan image of each object can be acquired at the standard bone scan speed, and Poisson downsampling is performed on the standard speed bone scan image to obtain the corresponding accelerated bone scan image.

[0174] Furthermore, before using the first training sample to train the accelerated image optimization model, it is necessary to use a registration algorithm to align the position coordinates of the images of the same parts in the standard speed bone scan image and the accelerated bone scan image, so that during the training process, the same parts of the standard speed bone scan image and the accelerated bone scan image correspond to each other.

[0175] 602. Obtain second training sample data, and use the second training sample data to train the low-dose image optimization model until the loss of the low-dose image optimization model meets the second target loss, wherein the second training sample data includes: low-dose bone scan images and standard drug dose bone scan images with the same morphological lesions acquired at the same scanning speed when each of the multiple subjects is injected with a low drug dose and a standard drug dose.

[0176] Before training the low-dose image optimization model, it is necessary to first obtain second training sample data and use the second training sample data to train the accelerated image optimization model until the loss of the accelerated image optimization model meets the second target loss, wherein the second training sample data includes: low-dose bone scan images and standard drug dose bone scan images with the same morphological lesions obtained at the same scanning speed for each object in multiple objects when a low drug dose and a standard drug dose are injected.

[0177] Furthermore, the loss function of the low-dose image optimization model in the embodiment of the present application also includes: at least one of: an L1 loss function, an L2 loss function, a perceptual loss function and a structural similarity loss function.

[0178] When acquiring the low-dose bone scan image and the standard drug dose bone scan image of each object, the standard drug dose bone scan image of each object when the standard drug dose is injected can be acquired; and Poisson downsampling is performed on the standard drug dose bone scan image to obtain the corresponding low-dose bone scan image.

[0179] Before using the second training sample data to train the low-dose image optimization model, a registration algorithm is used to align the position coordinates of the images of the same parts in the standard drug dose bone scan image and the low drug dose scan image, so that during the training process, the same parts of the standard drug dose bone scan image and the low drug dose scan image correspond to each other.

[0180] The following describes the process of performing Poisson downsampling on a standard speed bone scan image or a standard drug dose bone scan image:

[0181] A. Get the preset downsampling multiple u, the 0 <u<1;

[0182] For the convenience of explanation, the embodiment of the present application assumes that u is 0.5.

[0183] B. Obtaining a pixel value n at each pixel position (x, y) in a standard speed bone scan image or a standard drug dose bone scan image;

[0184] Assume that the pixel value at the pixel position (1,1) in the standard speed bone scan image or the standard drug dose bone scan image is 256.

[0185] C. Acquire a first image template corresponding to a standard speed bone scan image or a standard drug dose bone scan image, wherein the image size of the first image template is the same as that of the standard speed bone scan image or the standard drug dose bone scan image, but corresponding pixel values ​​in the first image template are all 0;

[0186] Assuming that the size of the standard speed bone scan image or the standard drug dose bone scan image is 256*512, and the pixel value at the pixel position (1,1) is 256, when acquiring the first image template, the size of the first image template is also 256*512, but each pixel value in the first image template is 0.

[0187] D. For the pixel value of the target pixel position in the first image template, randomly and equally likely obtain a number σ, and if σ≤u, add 1 to the pixel value of the target pixel position, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template;

[0188] After obtaining the first image template, a number σ is randomly and equally likely obtained for the pixel value of the target pixel position in the first image template (assuming that the pixel position is (1,1)), where 0<σ<1. If σ≤u, the pixel value of the target pixel position is increased by 1. Assuming that for the pixel value at the pixel position (1,1) in the first image template, σ is also 0.5, the pixel value at the pixel position (1,1) in the first image template is set to 1.

[0189] E. For the pixel value 0 at each pixel position in the first image template, repeat step D n times until the pixel value at each pixel position is traversed.

[0190] After step D is completed, step D is repeated n times (256) for the pixel value at the pixel position (1,1) in the first image template until the pixel value at each pixel position is traversed.

[0191] The number of repetitions n in this application is the pixel value 256 at the pixel position (1,1) in the first bone scan image.

[0192] Because the embodiment of the present application uses Poisson downsampling to process standard speed bone scan images or standard drug dose bone scan images, the efficiency of obtaining accelerated bone scan images or low dose bone scan images is improved compared to manual screening.

[0193] Furthermore, because Poisson sampling assumes that the numerical distribution in the standard speed bone scan image or the standard drug dose bone scan image is consistent with the density distribution of the actual drug in the human body, however, in reality, due to the fact that the acquisition time cannot be infinitely long, the metabolic state of the drug in the human body changes during the long acquisition process, and the subject moves during the scanning process, it is assumed that there is a difference between the standard speed bone scan image or the standard drug dose bone scan image and the actual drug distribution. Therefore, we propose the following optimized Poisson downsampling process.

[0194] A1. Obtain standard speed bone scan images or standard drug dose bone scan images;

[0195] B1. Perform denoising filtering on the standard speed bone scan image or the standard drug dose bone scan image according to a first preset algorithm to obtain a filtered standard speed bone scan image or a standard drug dose bone scan image. The first preset algorithm includes: at least one of a self-supervised learning algorithm, a Gaussian filtering algorithm, a median filtering algorithm, a bilateral filtering algorithm, an image denoising algorithm based on a Markov expert field, and a three-dimensional block matching algorithm.

[0196] For ease of understanding, the following describes the execution process of the first preset algorithm by taking the self-supervised learning algorithm as an example:

[0197] (1) respectively setting the pixel value of each pixel point in the standard speed bone scan image or the standard drug dose bone scan image to zero, thereby obtaining a plurality of processed standard speed bone scan images or standard drug dose bone scan images;

[0198] For easier understanding, the following example is given:

[0199] Assuming that the size of the standard speed bone scan image is 256*512, it means that the standard speed bone scan image has 256*512=65536 pixels. The present application sets the pixel value of each pixel in the standard speed bone scan image to zero, such as setting the pixel at position (1,1) to zero for the first time, and setting the pixel at position (1,2) to zero for the second time, until the pixel at position (256*512) is set to zero, thereby obtaining 65536 processed standard speed bone scan images.

[0200] (2) inputting the processed multiple standard speed bone scan images or standard drug dose bone scan images into the self-supervised neural network model to obtain filtered standard speed bone scan images or standard drug dose bone scan images;

[0201] The processed 65536 standard speed bone scan images or standard drug dose bone scan images are input into the pre-trained self-supervised neural network model, and then the filtered standard speed bone scan images or standard drug dose bone scan images output by the self-supervised neural network model can be obtained.

[0202] It is easy to understand that before the processed multiple standard speed bone scan images are input into the pre-trained self-supervised neural network model, the self-supervised neural network model needs to be trained. The following takes the standard speed bone scan image as an example to describe the training process of the self-supervised neural network model:

[0203] a. respectively setting the pixel point at each pixel position in the standard speed bone scan image of each of the multiple subjects to zero, so as to obtain multiple zeroed images of each patient after the target pixel position is set to zero;

[0204] Before training the self-supervised neural network model, it is necessary to first obtain training samples, wherein the training samples include zeroed images after the pixel points at each pixel position in the standard speed bone scan image of each object among the multiple objects are respectively set to zero.

[0205] Assume that there are m patients, m is greater than or equal to 2, and the size of the standard speed bone scan image of each patient is 256*512, then each patient has 256*512=65536 zero images corresponding to them, and m patients have 65536m zero images corresponding to them.

[0206] b. taking a plurality of zeroed images of each of a plurality of objects as input and the standard speed bone scan image of each patient as output, training the self-supervised neural network model until the output of the self-supervised neural network model satisfies a first target loss;

[0207] Assuming that the target object Xiaohua among multiple objects is taken as an example, 65536 zeroed images of Xiaohua are taken as input, and the standard speed bone scan images of Xiaohua are taken as output to train the self-supervised neural network model, wherein the specific training process is to take each zeroed image of Xiaohua as input to obtain the output image X of the self-supervised neural network model, and then calculate the loss between the image X and the standard speed bone scan image of Xiaohua, and use the loss and the back propagation algorithm to train the parameters of the self-supervised neural network model until the output of the self-supervised neural network model is the standard speed bone scan image of Xiaohua, that is, until the output of the self-supervised neural network model meets the first target loss.

[0208] C1. performing Poisson downsampling on the filtered standard speed bone scan image or the standard drug dose bone scan image to obtain an accelerated bone scan image or a low dose bone scan image;

[0209] After the above steps obtain the filtered standard speed bone scan image or standard drug dose bone scan image of each object, Poisson downsampling is further performed on the filtered standard speed bone scan image or standard drug dose bone scan image to obtain an accelerated bone scan image or a low-dose bone scan image, wherein the filtered standard speed bone scan image or standard drug dose bone scan image is closer to the distribution of the actual drug.

[0210] The process of performing Poisson downsampling on the filtered standard speed bone scan image or standard drug dose bone scan image is similar to the process of performing Poisson downsampling in the above embodiment, and will not be described in detail here.

[0211] In an embodiment of the present application, Poisson downsampling is performed on the filtered standard speed bone scan image or the standard drug dosage bone scan image, so that the distribution of the signal in the filtered standard speed bone scan image is closer to the distribution of the real drug signal, thereby improving the similarity between the signal distribution in the standard speed bone scan image and the real drug distribution in the patient's body.

[0212] The above describes in detail the method for optimizing the bone scan image in the embodiment of the present application. Next, the device for optimizing the bone scan image in the embodiment of the present application is described. Figure 7 , an embodiment of the device for optimizing bone scan images in the embodiment of the present application includes:

[0213] An acquisition unit 701 is used to acquire a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan;

[0214] A generating unit 702 is used to input the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose;

[0215] The optimization unit 703 is used to convert non-integer pixels in the generated bone scan image into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical features.

[0216] Preferably, the optimization unit 703 is specifically used for:

[0217] Acquire all pixel points in the generated bone scan image;

[0218] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0219] If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X, if Y is less than or equal to the decimal part of X, X is adjusted to a minimum positive integer greater than X, if Y is greater than the decimal part of X, X is adjusted to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixel points.

[0220] Preferably, the optimization unit 703 is specifically used for:

[0221] Acquire all pixel points in the generated bone scan image;

[0222] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0223] If the pixel value of the target pixel point is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle, wherein the target pixel point is any pixel point among all the pixel points.

[0224] Preferably, the accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model includes a convolutional neural network or a generative adversarial network;

[0225] The convolutional neural network includes a backbone optimized network structure.

[0226] Preferably, the device further comprises:

[0227] The correction unit 704 is used to perform a dot multiplication of the generated bone scan image and an acceleration factor matrix before converting the non-integer pixels in the generated bone scan image into integer pixels to obtain a quantitatively corrected bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the distribution of the pixel matrix of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value.

[0228] Preferably, the convolutional neural network also includes a quantitative correction network structure and a first convolution module, and the quantitative correction network structure includes a second convolution module and a third convolution module.

[0229] Preferably, the generating unit 702 is specifically configured to:

[0230] Inputting the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtaining a first feature three-dimensional matrix of the bone scan image to be optimized;

[0231] The first characteristic three-dimensional matrix is ​​superimposed with the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values;

[0232] Inputting the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix;

[0233] Inputting the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized;

[0234] Multiply the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix;

[0235] The fourth characteristic three-dimensional matrix is ​​input into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0236] Preferably, the loss function of the accelerated image optimization model or the low-dose image optimization model includes:

[0237] At least one of an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

[0238] Preferably, the device further comprises:

[0239] A first training unit 705 is used to obtain first training sample data, and train the accelerated image optimization model using the first training sample data until the loss of the accelerated image optimization model meets the first target loss, wherein the first training sample data includes: an accelerated bone scan image and a standard bone scan image of the same morphological lesion when each of the multiple subjects is injected with the same drug dose and the bone scan speeds are the accelerated bone scan speed and the standard bone scan speed, respectively, and the standard bone scan speed is less than 40 cm / min;

[0240] The second training unit 706 is used to obtain second training sample data and use the second training sample data to train the low-dose image optimization model until the loss of the low-dose image optimization model meets the second target loss. The second training sample data includes: low-dose bone scan images and standard drug dose bone scan images with the same morphological lesions obtained at the same scanning speed for each object among multiple objects when the low drug dose and the standard drug dose are injected.

[0241] The first training unit 705 is specifically configured to:

[0242] acquiring a standard speed bone scan image for each subject at a standard bone scan speed;

[0243] performing Poisson downsampling on the standard speed bone scan image to obtain a corresponding accelerated bone scan image;

[0244] The second training unit 706 is specifically configured to:

[0245] Obtaining standard drug dose bone scan images for each subject when a standard drug dose is infused;

[0246] Poisson downsampling was performed on the standard drug dose bone scan images to obtain the corresponding low dose bone scan images.

[0247] Preferably, the first training unit 705 or the second training unit 706 is specifically used for:

[0248] Get the downsampling multiple u, the 0 <u<1;

[0249] Obtaining a pixel value n of each pixel position (x, y) in a standard speed bone scan image or a standard drug dose bone scan image;

[0250] Acquire a first image template corresponding to a standard speed bone scan image or a standard drug dose bone scan image, wherein the image size of the first image template is the same as that of the standard speed bone scan image or the standard drug dose bone scan image, but pixel values ​​at corresponding positions in the first image template are all 0;

[0251] For the pixel value of the target pixel position in the first image template, a number σ is randomly obtained with equal probability, and if σ≤u, the pixel value of the target pixel position is increased by 1, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template;

[0252] For the pixel value 0 of each pixel position in the first image template, repeat the first step n times until the pixel value of each pixel position is traversed, and the first step includes: for the pixel value of the target pixel position in the first image template, randomly and equally likely obtain a number σ, if the σ≤u, then add 1 to the pixel value of the target pixel position, wherein 0<σ<1, and the target pixel position is any pixel position in the first image template.

[0253] Preferably, the device further comprises:

[0254] The denoising unit 707 is used to: before performing Poisson downsampling on the standard speed bone scan image or the standard drug dose bone scan image, use a filtering algorithm to perform denoising filtering on the standard speed bone scan image or the standard drug dose bone scan image, wherein the filtering algorithm includes at least one of a self-supervised learning algorithm, a Gaussian filtering algorithm, a median filtering algorithm, a bilateral filtering algorithm, an image denoising algorithm based on a Markov expert field, and a three-dimensional block matching algorithm.

[0255] Preferably, the denoising unit 707 is specifically used for:

[0256] After setting the pixel value of a pixel point in the standard speed bone scan image or the standard drug dose bone scan image to zero each time, a plurality of processed standard speed bone scan images or a plurality of standard drug dose bone scan images are obtained;

[0257] The processed multiple standard speed bone scan images or multiple standard drug dose bone scan images are input into a pre-trained self-supervised neural network model to obtain filtered standard speed bone scan images or standard drug dose bone scan images.

[0258] Preferably, the device further comprises:

[0259] The registration unit 708 is used to register the position coordinates of the images of the same part in the standard speed bone scan image and the accelerated bone scan image using a registration algorithm before using the first training sample data to train the accelerated image optimization model; or, before using the second training sample data to train the low-dose image optimization model, register the position coordinates of the images of the same part in the standard drug dose bone scan image and the low drug dose scan image using a registration algorithm.

[0260] In the embodiment of the present application, the bone scan image to be optimized is input into the accelerated image optimization model or the low-dose image optimization model through the generation unit 702, so that the bone scan image output by the accelerated image optimization model or the low-dose image optimization model is the corresponding standard speed bone scan image, or the standard dose bone scan image, and the embodiment of the present application converts the non-integer pixels in the bone scan images generated by different models into integer pixels through the optimization unit 703, so that the optimized bone scan image is closer to the real bone scan image, thereby improving the quality of the optimized bone scan image on the one hand, and improving the efficiency of the scanning image examination on the other hand, or reducing the damage to the body caused by the drug dose during bone scanning imaging.

[0261] The above describes the device for optimizing bone scan images in the embodiment of the present invention from the perspective of modular functional entities. The following describes the computer device in the embodiment of the present invention from the perspective of hardware processing:

[0262] The computer device is used to implement the function of the device for optimizing bone scan images. In an embodiment of the present invention, an embodiment of the computer device includes:

[0263] Processor and memory;

[0264] The memory is used to store a computer program, and when the processor is used to execute the computer program stored in the memory, the following steps can be implemented:

[0265] Acquiring a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan;

[0266] Inputting the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose;

[0267] The non-integer pixels in the generated bone scan image are converted into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical features.

[0268] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0269] Acquire all pixel points in the generated bone scan image;

[0270] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0271] If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X, if Y is less than or equal to the decimal part of X, X is adjusted to a minimum positive integer greater than X, if Y is greater than the decimal part of X, X is adjusted to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixel points.

[0272] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0273] Acquire all pixel points in the generated bone scan image;

[0274] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0275] If the pixel value of the target pixel point is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle, wherein the target pixel point is any pixel point among all the pixel points.

[0276] In some embodiments of the present invention, the accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model includes a convolutional neural network or a generative adversarial network; the convolutional neural network includes a backbone optimization network structure.

[0277] In some embodiments of the present invention, before converting the non-integer pixels in the generated bone scan image into integer pixels, the processor may also be used to implement the following steps:

[0278] The generated bone scan image is dot-multiplied with the acceleration factor matrix to obtain a quantitatively corrected bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the distribution of the pixel matrix of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value.

[0279] In some embodiments of the present invention, the convolutional neural network also includes a quantitative correction network structure and a first convolution module, and the quantitative correction network structure includes a second convolution module and a third convolution module.

[0280] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0281] Inputting the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtaining a first feature three-dimensional matrix of the bone scan image to be optimized;

[0282] The first characteristic three-dimensional matrix is ​​superimposed with the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values;

[0283] Inputting the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix;

[0284] Inputting the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized;

[0285] Multiply the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix;

[0286] The fourth characteristic three-dimensional matrix is ​​input into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0287] In some embodiments of the present invention, the loss function of the accelerated image optimization model or the low-dose image optimization model includes:

[0288] At least one of an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

[0289] It can be understood that when the processor in the computer device described above executes the computer program, it can also implement the functions of the various units in the corresponding device embodiments described above, which will not be repeated here. Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the neural network training device / optimized bone scanning image device. For example, the computer program can be divided into the various units in the above-mentioned neural network training device, and each unit can implement the specific functions described in the above-mentioned corresponding neural network training device.

[0290] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that a processor and a memory are merely examples of computer devices and do not constitute a limitation of the computer device. The computer device may include more or fewer components, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, and the like.

[0291] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, and uses various interfaces and lines to connect various parts of the entire computer device.

[0292] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0293] The present invention also provides a computer-readable storage medium, which is used to implement the function of the device for optimizing bone scan images, and stores a computer program thereon. When the computer program is executed by a processor, the processor can be used to perform the following steps:

[0294] Acquiring a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan;

[0295] Inputting the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose;

[0296] The non-integer pixels in the generated bone scan image are converted into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical features.

[0297] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0298] Acquire all pixel points in the generated bone scan image;

[0299] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0300] If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X, if Y is less than or equal to the decimal part of X, X is adjusted to a minimum positive integer greater than X, if Y is greater than the decimal part of X, X is adjusted to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixel points.

[0301] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0302] Acquire all pixel points in the generated bone scan image;

[0303] For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0;

[0304] If the pixel value of the target pixel point is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle, wherein the target pixel point is any pixel point among all the pixel points.

[0305] In some embodiments of the present invention, the accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model includes a convolutional neural network or a generative adversarial network; the convolutional neural network includes a backbone optimization network structure.

[0306] In some embodiments of the present invention, before converting the non-integer pixels in the generated bone scan image into integer pixels, the processor may also be used to implement the following steps:

[0307] The generated bone scan image is dot-multiplied with the acceleration factor matrix to obtain a quantitatively corrected bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the distribution of the pixel matrix of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value.

[0308] In some embodiments of the present invention, the convolutional neural network also includes a quantitative correction network structure and a first convolution module, and the quantitative correction network structure includes a second convolution module and a third convolution module.

[0309] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0310] Inputting the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtaining a first feature three-dimensional matrix of the bone scan image to be optimized;

[0311] The first characteristic three-dimensional matrix is ​​superimposed with the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values;

[0312] Inputting the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix;

[0313] Inputting the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized;

[0314] Multiply the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix;

[0315] The fourth characteristic three-dimensional matrix is ​​input into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

[0316] In some embodiments of the present invention, the loss function of the accelerated image optimization model or the low-dose image optimization model includes:

[0317] At least one of an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

[0318] It is understandable that if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a corresponding computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned corresponding embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0319] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0320] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0321] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0322] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing bone scan images, characterized in that: include: Acquiring a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan; Inputting the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard speed bone scan image obtained at a standard bone scan speed, or a standard dose bone scan image obtained when scanning at a standard drug dose; The non-integer pixels in the generated bone scan image are converted into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical features.

2. The method according to claim 1, characterized in that The step of converting non-integer pixels in the generated bone scan image into integer pixels comprises: Acquire all pixel points in the generated bone scan image; For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0; If the pixel value of the target pixel is a non-integer X greater than 0, a random number Y greater than 0 and less than 1 is generated for X, if Y is less than or equal to the decimal part of X, X is adjusted to a minimum positive integer greater than X, if Y is greater than the decimal part of X, X is adjusted to a maximum positive integer less than X, wherein the target pixel is any pixel among all the pixel points.

3. The method according to claim 1, characterized in that The step of converting non-integer pixels in the generated bone scan image into integer pixels comprises: Acquire all pixel points in the generated bone scan image; For a target pixel point, if the pixel value of the target pixel point is less than 0, the pixel value of the target pixel point is adjusted to 0; If the pixel value of the target pixel point is a non-integer greater than 0, the non-integer greater than 0 is adjusted to an integer according to the rounding principle, wherein the target pixel point is any pixel point among all the pixel points.

4. The method according to claim 1, characterized in that: The accelerated image optimization model includes a convolutional neural network or a generative adversarial network, and the low-dose image optimization model includes a convolutional neural network or a generative adversarial network; The convolutional neural network includes a backbone optimized network structure.

5. The method according to claim 4, characterized in that Before converting the non-integer pixels in the generated bone scan image into integer pixels, the method further includes: The generated bone scan image is dot-multiplied with the acceleration factor matrix to obtain a quantitatively corrected bone scan image, wherein the distribution of the acceleration factor matrix is ​​the same as the distribution of the pixel matrix of the generated bone scan image, and the value of the acceleration factor matrix is ​​a preset acceleration factor value.

6. The method according to claim 4, characterized in that The convolutional neural network also includes a quantitative correction network structure and a first convolution module, and the quantitative correction network structure includes a second convolution module and a third convolution module.

7. The method according to claim 6, characterized in that The step of inputting the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model comprises: Inputting the bone scan image to be optimized into the second convolution module to extract image features of the bone scan image to be optimized, and obtaining a first feature three-dimensional matrix of the bone scan image to be optimized; The first characteristic three-dimensional matrix is ​​superimposed with the pre-input acceleration factor characteristic matrix on the target dimension to obtain a second characteristic three-dimensional matrix, wherein the acceleration factor characteristic matrix and the first characteristic three-dimensional matrix have the same distribution on two dimensions other than the target dimension, and the values ​​in the acceleration factor characteristic matrix are preset acceleration factor values; Inputting the second characteristic three-dimensional matrix into the third convolution module to weaken the background noise in the image corresponding to the second characteristic three-dimensional matrix, perform a first enhancement on the normal tissue in the image corresponding to the second characteristic three-dimensional matrix, and perform a second enhancement on the lesion tissue in the image corresponding to the second characteristic three-dimensional matrix, so as to obtain a quantitative correction weight matrix; Inputting the bone scan image to be optimized into the backbone optimization network structure to extract a third characteristic three-dimensional matrix of the bone scan image to be optimized; Multiply the quantitative correction weight matrix by the third characteristic three-dimensional matrix to obtain a fourth characteristic three-dimensional matrix; The fourth characteristic three-dimensional matrix is ​​input into the first convolution module to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model.

8. The method according to any one of claims 1 to 7, characterized in that The loss function of the accelerated image optimization model or the low-dose image optimization model includes: At least one of an L1 loss function, an L2 loss function, a perceptual loss function, and a structural similarity loss function.

9. A device for optimizing bone scan images, characterized in that: include: an acquisition unit, configured to acquire a bone scan image to be optimized, wherein the bone scan image to be optimized includes an accelerated bone scan image obtained when the subject performs an accelerated bone scan, or a low-dose bone scan image when the subject is injected with a low drug dose, wherein the scanning speed of the accelerated bone scan is greater than the scanning speed of the standard bone scan, and the low drug dose is less than the standard drug dose during the standard bone scan; a generating unit, configured to input the bone scan image to be optimized into an accelerated image optimization model or a low-dose image optimization model to obtain a bone scan image generated by the accelerated image optimization model or the low-dose image optimization model, wherein the generated bone scan image includes a standard bone scan image obtained at a standard bone scan speed, or a standard dose scan image obtained when scanning at a standard drug dose; The optimization unit is used to convert non-integer pixels in the generated bone scan image into integer pixels to obtain an optimized bone scan image, wherein the optimized bone scan image and the real bone scan image have similar physical characteristics.

10. A computer device comprising a processor, characterized in that: When the processor executes the computer program stored in the memory, the processor is used to implement the method for optimizing the bone scan image according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a computer device, it is used to implement the method for optimizing bone scan images according to any one of claims 1 to 8.

Images