Medical image processing method, model training method, equipment, medium and product

Abstract

The invention discloses a medical image processing method, a model training method, equipment, a medium and a product. The method comprises the following steps: dividing a to-be-processed image sequence of a target scanning part into at least two to-be-processed image combinations; selecting a reference image combination meeting a set similarity condition with each to-be-processed image combination from a pre-stored reference image sequence of the target scanning part; performing up-sampling on the to-be-processed image combination to obtain an up-sampling result, wherein the resolution of the up-sampling result is the same as that of the reference image sequence; a pre-trained image generation model is used to process each up-sampling result and the corresponding reference image combination, a target image combination corresponding to each to-be-processed image combination is determined, and the resolution of the target image combination is the same as the resolution of the reference image sequence; and determining a target image sequence corresponding to the to-be-processed image sequence according to all the target image combinations. According to the embodiment of the invention, high accuracy of the target image sequence can be ensured.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing technology, and in particular to a medical image processing method, a model training method, equipment, a medium and a product. Background Art

[0002] Super-resolution technology can generate high-resolution images from low-resolution CT images, significantly improving the details and clarity of the images. This is particularly important for detecting and diagnosing tiny lesions (such as early cancer, tiny vascular abnormalities, etc.). In addition, in surgical operations, high-resolution CT images can provide more accurate anatomical information to help surgeons plan operations and improve the success rate and safety of operations. High-resolution CT images provide better support for post-processing and analysis (such as three-dimensional reconstruction, image segmentation, functional imaging, etc.), thereby improving the accuracy and efficiency of related work.

[0003] By using super-resolution technology, high-quality CT images can be obtained while reducing scanning parameters (such as reducing X-ray dose), thereby reducing the radiation dose received by patients and reducing potential radiation risks.

[0004] The existing technology usually inputs low-resolution images into super-resolution models to obtain corresponding high-resolution images. Since it is difficult to directly generate high-resolution images based on low-resolution images, the accuracy of high-resolution images output by existing super-resolution models is low. Summary of the invention

[0005] The present invention provides a medical image processing method, a model training method, a device, a medium and a product to solve the problem that the prior art cannot determine an accurate high-resolution image based on a low-resolution image.

[0006] According to one aspect of the present invention, there is provided a medical image processing method, comprising:

[0007] Dividing the sequence of images to be processed of the target scanning part into at least two combinations of images to be processed, wherein the combination of images to be processed includes a first number of continuous images;

[0008] Selecting, from a pre-stored reference image sequence of the target scanning part, a reference image combination that meets a set similarity condition with each of the image combinations to be processed, wherein the number of consecutive images in the reference image combination is greater than the first number;

[0009] Upsampling the combination of images to be processed to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence;

[0010] Using a pre-trained image generation model to process each upsampling result and a corresponding reference image combination, determining a target image combination corresponding to each of the image combinations to be processed, wherein the resolution of the target image combination is the same as the resolution of the reference image sequence;

[0011] According to all the target image combinations, a target image sequence corresponding to the image sequence to be processed is determined.

[0012] According to another aspect of the present invention, there is provided a model training method for training a reference image determination model, comprising:

[0013] Acquire a first sample set, wherein the first sample set includes a set number of training image sequences for a target scanning part;

[0014] Randomly selecting an initial image combination from each of the training image sequences, and downsampling each image in the initial image combination to obtain a training image combination, wherein the training image combination includes a first number of continuous images;

[0015] Downsampling the training image sequence to obtain a downsampling result, wherein the resolution of the downsampling result is the same as the resolution of the training image combination;

[0016] The downsampling result and each of the training image combinations are input into a reference image determination model to obtain a reference image in the reference image sequence that meets a set similarity condition with a set image in each of the training image combinations, and parameter optimization of the reference image determination model is completed according to the similarity between each of the reference images and the corresponding set image to obtain a pre-trained reference image determination model.

[0017] According to another aspect of the present invention, there is provided a model training method for training an image generation model, comprising:

[0018] Acquire a second sample set, wherein the second sample set includes a reference image sequence for a target scanning part and a set number of training image sequences;

[0019] Randomly selecting a training image combination from each of the training image sequences, and downsampling the training image combination to obtain a target training image combination, wherein the training image combination includes a first number of continuous images;

[0020] Selecting, from the reference image sequence, a reference image combination that meets a set similarity condition with each of the target training image combinations, wherein the number of consecutive images in the reference image combination is greater than the first number;

[0021] Upsampling the target training image combination to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence;

[0022] Each of the upsampling results and the corresponding reference image combination is input into the image generation model to obtain a target image combination corresponding to each upsampling result, and according to the similarity between each of the target image combination and the corresponding training image combination, the parameter optimization of the image generation model is completed to obtain a pre-trained image generation model.

[0023] According to another aspect of the present invention, there is provided a medical image processing apparatus, comprising:

[0024] An image grouping module, used to divide the sequence of images to be processed of the target scanning part into at least two combinations of images to be processed, wherein the combination of images to be processed includes a first number of continuous images;

[0025] A reference image module, configured to select a reference image combination that meets a set similarity condition with each of the image combinations to be processed from a pre-stored reference image sequence of the target scanning part, wherein the number of consecutive images in the reference image combination is greater than the first number;

[0026] A sampling module, used for upsampling the combination of images to be processed to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence;

[0027] A prediction module, used to process each upsampling result and the corresponding reference image combination using a pre-trained image generation model, and determine a target image combination corresponding to each of the image combinations to be processed, wherein the resolution of the target image combination is the same as the resolution of the reference image sequence;

[0028] The result module is used to determine a target image sequence corresponding to the image sequence to be processed according to all the target image combinations.

[0029] According to another aspect of the present invention, there is provided a model training device for training a reference image determination model, comprising:

[0030] A first acquisition module, configured to acquire a first sample set, wherein the first sample set includes a set number of training image sequences for a target scanning part;

[0031] A first selection module is used to randomly select an initial image combination from each of the training image sequences, and downsample each image in the initial image combination to obtain a training image combination, wherein the training image combination includes a first number of continuous images;

[0032] A downsampling module, used for downsampling the training image sequence to obtain a downsampling result, wherein the resolution of the downsampling result is the same as the resolution of the image combination to be processed;

[0033] The first training module is used to input the downsampling result and each of the training image combinations into a reference image determination model to obtain a reference image in the reference image sequence that meets a set similarity condition with a set image in each of the training image combinations, and to optimize the parameters of the reference image determination model according to the similarity between each of the reference images and the corresponding set image to obtain a pre-trained reference image determination model.

[0034] According to another aspect of the present invention, there is provided a model training device for training an image generation model, comprising:

[0035] A second acquisition module, used to acquire a second sample set, wherein the second sample set includes a reference image sequence for a target scanning part and a set number of training image sequences;

[0036] A second selection module is used to randomly select a training image combination from each of the training image sequences, and downsample the training image combination to obtain a target training image combination, wherein the training image combination includes a first number of continuous images;

[0037] A third selection module is used to select, from the reference image sequence, a reference image combination that meets a set similarity condition with each of the target training image combinations, wherein the number of consecutive images in the reference image combination is greater than the first number;

[0038] An upsampling module, used for upsampling the target training image combination to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence;

[0039] The second model training module is used to input each of the upsampling results and the corresponding reference image combination into the image generation model to obtain a target image combination corresponding to each upsampling result, and according to the similarity between each of the target image combination and the corresponding training image combination, complete the parameter optimization of the image generation model to obtain a pre-trained image generation model.

[0040] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0041] at least one processor; and

[0042] a memory communicatively connected to the at least one processor; wherein,

[0043] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the medical image processing method described in any embodiment of the present invention, or the model training method for training a reference image determination model described in any embodiment, or the model training method for training an image generation model described in any embodiment.

[0044] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the medical image processing method described in any embodiment of the present invention, or the model training method for training a reference image determination model described in any embodiment, or the model training method for training an image generation model described in any embodiment.

[0045] According to another aspect of the present invention, a computer program product is provided, which includes a computer program, and when the computer program is executed by a processor, it implements the medical image processing method according to any embodiment, or the model training method for training a reference image determination model according to any embodiment, or the model training method for training an image generation model according to any embodiment.

[0046] The technical solution of the medical image processing method provided by the embodiment of the present invention is to match the most similar reference image combination in the reference image sequence for each combination of images to be processed in the image sequence to be processed; up-sample the image combination to be processed to obtain an up-sampling result; since the up-sampling result includes the global information of each image in the image combination to be processed, and each image in the reference image sequence includes high-resolution detail information, the pre-trained image generation model can combine the low-resolution prior global information of each image in the up-sampling result with the high-resolution detail information of the reference image combination, and infer and generate a target image combination corresponding to each combination of images to be processed, which reduces the difficulty of model reasoning, helps to improve the accuracy of the target image combination, and thus improves the accuracy of the target image sequence determined based on each target image combination.

[0047] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 is a flowchart of a medical image processing method provided according to an embodiment of the present invention;

[0050] Figure 2 is a structural schematic diagram of an image generation model provided according to an embodiment of the present invention;

[0051] Figure 3 is another flow chart of a medical image processing method provided according to an embodiment of the present invention;

[0052] Figure 4 is a flow chart of a model training method for training a reference image determination model provided according to an embodiment of the present invention;

[0053] Figure 5 is a schematic diagram of the structure of a reference image determination model provided according to an embodiment of the present invention;

[0054] Figure 6 is a flowchart of a model training method for training an image generation model provided according to an embodiment of the present invention;

[0055] Figure 7 is a schematic structural diagram of a medical image processing device provided according to an embodiment of the present invention;

[0056] Figure 8 is a structural schematic diagram of a model training device for training a reference image determination model provided in an embodiment of the present invention;

[0057] Fig. 9 is a structural schematic diagram of a model training device for training an image generation model provided according to an embodiment of the present invention;

[0058] Fig.10 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] 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.

[0060] It should be noted that the terms "first", "second", 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 of the present invention described herein can be implemented in an order other than those 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 that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0061] Figure 1 This is a flowchart of a medical image processing method provided by an embodiment of the present invention. This embodiment can be applied to the case where a high-resolution image corresponding to a low-resolution image is obtained by inferring a high-resolution reference image sequence of a corresponding body part. The method can be executed by a medical image processing device, which can be implemented in the form of hardware and / or software, and can be configured in a processor of an electronic device. Figure 1 As shown, the method includes:

[0062] S110, dividing the sequence of images to be processed of the target scanning part into at least two combinations of images to be processed, where the combination of images to be processed includes a first number of continuous images.

[0063] The target scanning part is any scanning part of the human body, such as the head, chest, upper abdomen, lower abdomen, etc.

[0064] The image sequence to be processed refers to an image sequence reconstructed based on projection data. Each image in the image sequence corresponds to a unique image identifier, wherein the image identifier corresponds to a coordinate position.

[0065] After the image sequence to be processed is determined, the number of images included in the preset image combination to be processed is obtained; the image sequence to be processed is divided into at least two image combinations to be processed according to the number of images. Exemplarily, the image sequence to be processed includes 50 images, and one image combination to be processed includes 5 consecutive images. Therefore, the image sequence to be processed is divided into 10 image combinations to be processed, each of which includes 5 consecutive images. In one embodiment, the number of images is configured to be greater than or equal to 2 and less than or equal to 15.

[0066] S120. Selecting, from a pre-stored reference image sequence of the target scanning part, a reference image combination that meets a set similarity condition with each image combination to be processed, wherein the number of consecutive images in the reference image combination is greater than the first number.

[0067] The resolution of the reference image sequence is the desired image resolution.

[0068] In one embodiment, a reference image library is pre-set, and the reference image library includes reference image sequences of various common scanning parts. For example, the reference image library includes a reference image sequence for the head, a reference image sequence for the chest, a reference image sequence for the upper abdomen, a reference image sequence for the lower abdomen, etc.

[0069] This step aims to select the reference image combination that is most similar to each image combination to be processed from the reference image sequence, and the number of images included in the reference image combination is greater than the number of continuous images included in the corresponding image combination to be processed.

[0070] In one embodiment, a reference image combination that is most similar to each image combination to be processed is selected from a reference image sequence based on an existing similarity determination method. Optionally, a reference image that is most similar to a set image in each image combination to be processed is selected from a reference image sequence based on an existing similarity determination method, and then the position of the reference image in the reference image combination is determined based on the position of the set image in the image combination to be processed; then, other similar images are selected from the image combination to be processed according to the position of the reference image in the reference image, the image identifier of the reference image, and the number of set images corresponding to the reference image combination, and finally all the selected images are used as the reference image combination corresponding to the image combination to be processed. Wherein, if the set image is the starting image of the image combination to be processed; then the reference image is the starting image of the corresponding reference image sequence; if the set image is the ending image of the image combination to be processed, then the reference image is the ending image of the corresponding reference image sequence; if the set image is the intermediate image of the image combination to be processed, then the reference image is the intermediate image of the corresponding reference image sequence.

[0071] S130 , up-sampling the image combination to be processed to obtain an up-sampling result, wherein the resolution of the up-sampling result is the same as the resolution of the reference image sequence.

[0072] The resolution of the image sequence to be processed is smaller than the resolution of the reference image sequence. The image combination to be processed is upsampled to obtain an updated image combination to be processed, so that the resolution of the updated image combination to be processed is the same as the resolution of the reference image sequence, and the resolution of the updated image combination to be processed is also the target resolution.

[0073] S140. Use a pre-trained image generation model to process each upsampling result and the corresponding reference image combination, determine a target image combination corresponding to each image combination to be processed, and the resolution of the target image combination is the same as the resolution of the reference image sequence.

[0074] The pre-trained image generation model is trained to combine the prior global information of the low-resolution image with the detailed reference information of the high-resolution image to infer the high-resolution target image corresponding to the low-resolution image; that is, to combine the prior global information of the image combination to be processed with the detailed reference information of the reference image combination to infer the high-resolution target image combination corresponding to the image combination to be processed. It can be understood that the resolution of the target image combination is the same as the resolution of the reference image sequence, that is, the resolution of the target image combination is also the image resolution expected by the user.

[0075] In one embodiment, the image generation model includes an L-layer residual convolution network for compression and an L-1-layer residual convolution network for image fusion; the output of the previous residual convolution network for compression is the input of the next residual convolution network for compression; the upsampling result of the output result of the L-layer residual convolution network for compression and the output result of the L-1-layer residual convolution network for compression are both used as the input of the L-1-layer residual convolution network for image fusion; the upsampling result of the output result of the previous residual convolution network for image fusion and the output result of the residual convolution network for compression at the target level are both used as the input of the next residual convolution network for image fusion, and the output result of the residual convolution network for compression at the target level has the same resolution as the upsampling result of the output result of the previous residual convolution network for image fusion. In this embodiment, the residual convolution network for image fusion is configured to fuse the current upsampling result with the corresponding reference image combination. By introducing residual convolutional neural network to solve the gradient vanishing problem in deep neural network training, it helps to improve the accuracy of model output results.

[0076] In one embodiment, Figure 2As shown, the image generation model also includes a residual convolution network for selecting an image, whose input end is connected to the output end of the residual convolution network of the last layer for image fusion, and is used to select a target image combination from the output result of the residual convolution network of the last layer for image fusion. In this embodiment, the reference convolution network for image fusion is configured to fuse the upsampling result with the reference image sequence, and the reference image without a matching upsampling result remains unchanged; that is, the output result of the residual convolution network of the last layer for image fusion includes the fusion result of the reference image combination in the reference image sequence and the corresponding upsampling result, and the remaining reference images in the reference image sequence; the residual convolution network for selecting an image is configured to select the fusion result of the reference image combination and the corresponding upsampling result from the output result of the residual convolution network of the last layer for image fusion as the target image combination. Wherein, the identifier of c corresponds to the number of channels, for example, cL represents L channels. In this embodiment, the reference image sequence can be processed as a whole, without matching the corresponding reference image combination for each upsampling result, and the image processing process is simpler.

[0077] S150. Determine a target image sequence corresponding to the image sequence to be processed according to all target image combinations.

[0078] After the target image combinations corresponding to the image combinations to be processed are determined, the images in all the target image combinations are sorted based on the image identifiers to obtain a target image sequence corresponding to the image sequence to be processed.

[0079] The technical solution of the medical image processing method provided by the embodiment of the present invention is to match the most similar reference image combination in the reference image sequence for each combination of images to be processed in the image sequence to be processed; up-sample the image combination to be processed to obtain an up-sampling result; since the up-sampling result includes the global information of each image in the image combination to be processed, and each image in the reference image sequence includes high-resolution detail information, the pre-trained image generation model can combine the low-resolution prior global information of each image in the up-sampling result with the high-resolution detail information of the reference image combination, and infer and generate a target image combination corresponding to each combination of images to be processed, which reduces the difficulty of model reasoning, helps to improve the accuracy of the target image combination, and thus improves the accuracy of the target image sequence determined based on each target image combination.

[0080] Figure 3 This is a flow chart of a medical image processing method provided by an embodiment of the present invention. This embodiment is used to refine the reference image combination determination step in the above-mentioned embodiment. Figure 3 As shown, the method includes:

[0081] S210: Divide the sequence of to-be-processed images of the target scanning part into at least two to-be-processed image combinations, where the to-be-processed image combinations include a first number of continuous images.

[0082] S2201. Down-sample a pre-stored reference image sequence of the target scanning part to obtain a target reference image sequence, wherein the resolution of the target reference image sequence is the same as the resolution of the image sequence to be processed.

[0083] In order to improve the accuracy of image similarity determination, the downsampling multiple of the reference image sequence is determined according to the resolution multiple relationship between the reference image sequence and the image combination to be processed, such as two times; then the reference image sequence is downsampled according to the downsampling multiple to obtain the target reference image sequence. In this way, the resolution of the target reference image sequence is the same as that of the image sequence to be processed.

[0084] Exemplarily, the resolution of the reference image sequence is h×w×d. The resolution of the image combination to be processed is M×w / 2×d / 2, where M is the number of images included in the image combination to be processed. Therefore, the downsampling result of the reference image sequence, that is, the resolution of the target reference image sequence is h×w / 2×d / 2.

[0085] S2202: Select, from the target reference image sequence, a reference image that meets a set similarity condition with a set image in each combination of images to be processed.

[0086] A predetermined setting image is selected from the image combination to be processed, and the setting image may be an initial image, an intermediate image or an end image of the image combination to be processed.

[0087] This step aims to select a reference image that is most similar to the set image from the target reference image sequence, so the set similarity condition can be optionally set to a similarity threshold range.

[0088] In one embodiment, a pre-trained reference image determination model is used to select a reference image that meets a set similarity condition with a set image in each combination of images to be processed from a target reference image sequence. This embodiment can quickly and accurately select a reference image that is most similar to a set image in each combination of images to be processed from a target reference image sequence.

[0089] The reference image determination model may be a classification model, such as a 2D image classification model such as a convolutional network or a residual network, and different images to be processed in the combination of images to be processed are configured to be input into the pre-trained reference image determination model from different channels of the pre-trained reference image determination model. It is understandable that the use of a 2D image classification model improves the speed of determining the reference image.

[0090] In one embodiment, Figure 4 As shown in FIG, the reference image determination model includes multiple layers of residual convolutional networks for compression connected in sequence and a fully connected layer connected to the last upsampling residual convolutional network. The output result is a reference image combination, and the resolution of the reference image combination is equal to the resolution of the reference image sequence.

[0091] S2203: Determine the position of the reference image in the reference image combination according to the position of the set image in the image combination to be processed.

[0092] If the image is set as the middle image of the image combination to be processed, then the reference image is also the middle image of the reference image combination; if the image is set as the starting image of the image combination to be processed, then the reference image is the Nth image of the reference image combination, and the number of images included in the reference image combination is 2N+M, where M is the number of images included in the image combination to be processed; if the image is set as the ending image of the image combination to be processed, then the reference image is the Nth to last image of the reference image combination, and the number of images included in the reference image combination is still 2N+M.

[0093] S2204. Select a group of images from the reference image sequence according to the position of the reference image in the reference image combination, the image identifier of the reference image and the number of target images included in the reference image combination, and use the group of images as a reference image combination that meets a set similarity condition with the image combination to be processed.

[0094] Exemplarily, the reference image is in the middle of the reference image combination, the image of the reference image is identified as S, and the reference image combination is set to include T images. Then, the image identified as S in the reference image sequence and the combination of the first (T-1) / 2 images and the last (T-1) / 2 images of the image identified as S are used as the reference image combination.

[0095] S230 , up-sampling the image combination to be processed to obtain an up-sampling result, wherein the resolution of the up-sampling result is the same as the resolution of the reference image sequence.

[0096] S240, using a pre-trained image generation model to process each upsampling result and the corresponding reference image combination, determine a target image combination corresponding to each image combination to be processed, and the resolution of the target image combination is the same as the resolution of the reference image sequence.

[0097] S250: Determine a target image sequence corresponding to the image sequence to be processed according to all target image combinations.

[0098] The embodiment of the present invention selects a reference image that is most similar to a set image in the image combination to be processed from a target reference image sequence through a pre-trained reference image determination model, and then selects a reference image combination from the reference image sequence based on the image identifier corresponding to the reference image, the number of images in the reference image combination, and the position of the set image in the image combination to be processed, thereby achieving the technical effect of quickly selecting a reference image combination corresponding to the image combination to be processed from the reference image sequence.

[0099] Figure 5 Flow chart of a model training method for training a reference image determination model provided by an embodiment of the present invention. Figure 5 As shown, the method includes:

[0100] S310: Acquire a first sample set, where the first sample set includes a set number of training image sequences for a target scanning part.

[0101] The training image sequence is an image sequence to be processed, and its resolution is the first resolution.

[0102] This embodiment does not impose any specific limitation on the set number, and it can be determined according to specific circumstances during actual model training.

[0103] S320: Randomly select an initial image combination from each of the training image sequences, and downsample each image in the initial image combination to obtain a training image combination, where the training image combination includes a first number of continuous images.

[0104] A first number of continuous images are selected from the training image sequence, and the first number of continuous images are used as the initial image combination. The initial image combination is selected by random selection, so that the image identifiers in the initial reminder combination are random and irregular. This form of initial image combination is used for model training to avoid the influence of the image combination order on the training result.

[0105] Specifically, in the training image sequence x (i) Randomly select an image label K from the image, and use K-1, K and K+1 as training image combinations

[0106] S330, down-sample the training image sequence to obtain a down-sampling result, wherein the resolution of the down-sampling result is the same as the resolution of the training image combination.

[0107] In order to simplify the training process, this embodiment uses the training image sequence as the reference image sequence and downsamples it to obtain the downsampled result x refSince the initial image combination is part of the training image sequence, when the resolution of the down-sampling result is the same as the resolution of the training image combination, the down-sampling result can be regarded as including the training image combination.

[0108] S340. Input the downsampling result and each training image combination into the reference image determination model to obtain a reference image in the reference image sequence that meets the set similarity conditions with the set image in each training image combination, and optimize the parameters of the reference image determination model according to the similarity between each reference image and the corresponding set image to obtain a pre-trained reference image determination model.

[0109] The downsampled result x ref Combined with a training image As a set of input data; input the input data into the reference image determination model to obtain a predicted reference image that is most similar to the set image in the training image combination Based on the predicted reference image Compared with the actual reference image y (i) The difference between them is used to adjust the model parameters of the predicted reference image until the predicted reference image Compared with the actual reference image y (i) The target similarity condition is met between them, and the pre-trained reference image determination model is obtained. Among them, the actual reference image y (i) For, set the image identifier of the image in the downsampling result x ref The corresponding image in .

[0110] The embodiment of the present invention simplifies the training process of the reference image determination model and improves the accuracy of the model training result by using the same image sequence as the training image sequence and the reference image sequence.

[0111] Figure 6 Flowchart of a model training method for training an image generation model provided by an embodiment of the present invention. Figure 6 As shown, the method includes:

[0112] S410 , obtaining a second sample set, where the second sample set includes a reference image sequence for a target scanning part and a set number of training image sequences.

[0113] If the image generation model is expected to be trained into a model for only one target scanning part, the second training sample set may include a reference image sequence for the target scanning part and a set number of training image sequences; if the image generation model is expected to be trained into a model for multiple target scanning parts, the second training sample set may include reference image sequences for different target scanning parts and a set number of training image sequences corresponding to each target scanning part.

[0114] S420: Randomly select a training image combination from each training image sequence, and downsample the training image combination to obtain a target training image combination, where the training image combination includes a first number of continuous images.

[0115] A first number of continuous images are selected from the training image sequence, and the first number of continuous images are used as the initial image combination. The initial image combination is selected by random selection, so that the image identifiers in the initial reminder combination are random and irregular. This form of initial image combination is used for model training to avoid the influence of the image combination order on the training result.

[0116] Specifically, in the training image sequence x (i) Randomly select an image identifier K, and use K-1, K and K+1 as the initial image combination The initial image combination is downsampled to obtain a target training image combination.

[0117] S430: Select, from the reference image sequence, a reference image combination that meets a set similarity condition with each target training image combination, wherein the number of consecutive images in the reference image combination is greater than a first number.

[0118] The reference image combination selection method described in the above embodiment is adopted to select a reference image combination that meets the set similarity condition with each target training image combination from the reference image sequence, and this embodiment will not be described in detail here.

[0119] In one embodiment, if the second training sample set includes reference image sequences for different target scanning parts, the scanning part corresponding to the current training image sequence is determined, and based on the scanning part, a reference image sequence for the same scanning part is selected from the second training sample set; and then, from the reference image sequence, a reference image combination that meets the set similarity condition with each target training image combination is selected.

[0120] S440 , up-sampling the target training image combination to obtain an up-sampling result, wherein the resolution of the up-sampling result is the same as the resolution of the reference image sequence.

[0121] After the target training image combination is determined, it is upsampled to obtain an upsampled result, and the resolution of the upsampled result is the same as the resolution of the reference image sequence. In this way, each image in the updated target training image combination retains the previous global information and adds detail information based on the global information. The corresponding reference image combination includes the original high-resolution detail information of the same body part.

[0122] S450. Input each upsampling result and the corresponding reference image combination into the image generation model to obtain a target image combination corresponding to each upsampling result, and optimize the parameters of the image generation model according to the similarity between each target image combination and the corresponding training image combination to obtain a pre-trained image generation model.

[0123] It is understandable that the most accurate super-resolution image reconstruction result corresponding to the target training image combination is the training image combination. Therefore, each updated target training image combination and the corresponding reference image combination are input into the image generation model to obtain a target image combination corresponding to each upsampling result, and then the model parameters of the image generation model are adjusted according to the similarity between each target image combination and the corresponding training image combination until the target image combination determined by the image generation model meets the similarity requirement with the corresponding training image combination, thereby obtaining a pre-trained image generation model.

[0124] In one embodiment, while setting the second training sample set, a validation set and a prediction set are also set. Based on the above-mentioned model training method and the second training sample set, multiple pre-trained image generation models are determined, and then the optimal model is selected from the multiple pre-trained image generation models based on the validation set, and the optimal model is used as the final pre-trained image generation model; the prediction set is used to observe the model prediction effect.

[0125] The technical solution provided by the embodiment of the present invention is to obtain a target training image combination by downsampling the training image combination, and use the training image combination as the super-resolution reconstructed image true value corresponding to the target training image combination; select the image combination most similar to the target training image combination in the reference image sequence to obtain the reference image combination; upsample the target training image combination to obtain an upsampled result, and the upsampled result has the same resolution as the reference image combination; then input the reference image combination corresponding to each upsampled result into the image generation model, and the image generation model will obtain a predicted target image combination; then determine the similarity between the target image combination and the training image combination, and adjust the model parameters of the image generation model based on the similarity graph. The low-resolution training image combination and its corresponding true value are cleverly solved, which ensures the accuracy of the image generation model training method and the accuracy of the prediction results of the pre-trained image generation model.

[0126] Figure 7 Schematic diagram of the structure of the image processing device provided by the embodiment of the present invention. Figure 7 As shown, the device comprises:

[0127] An image grouping module 51 is used to divide the sequence of to-be-processed images of the target scanning part into at least two to-be-processed image combinations, wherein the to-be-processed image combinations include a first number of continuous images;

[0128] A reference image module 52, configured to select a reference image combination that meets a set similarity condition with each of the image combinations to be processed from a pre-stored reference image sequence of the target scanning part, wherein the number of consecutive images in the reference image combination is greater than the first number;

[0129] A sampling module 53, configured to upsample the image combination to be processed to obtain an upsampled result, wherein the resolution of the upsampled result is the same as the resolution of the reference image sequence;

[0130] A prediction module 54 is used to process each upsampling result and the corresponding reference image combination using a pre-trained image generation model, and determine a target image combination corresponding to each of the image combinations to be processed, wherein the resolution of the target image combination is the same as the resolution of the reference image sequence;

[0131] The result module 55 is used to determine a target image sequence corresponding to the image sequence to be processed according to all the target image combinations.

[0132] In one embodiment, the reference image module 52 includes:

[0133] A down-sampling unit, configured to down-sample a pre-stored reference image sequence of the target scanning part to obtain a target reference image sequence, wherein the resolution of the target reference image sequence is the same as the resolution of the image sequence to be processed;

[0134] A selection unit, configured to select, from the target reference image sequence, a reference image that meets a set similarity condition with a set image in each of the combinations of images to be processed;

[0135] a position determination unit, configured to determine the position of the reference image in the reference image combination according to the position of the set image in the image combination to be processed;

[0136] A reference image determination unit is used to select a group of images from the reference image sequence according to the position of the reference image in the reference image combination, the image identification of the reference image and the number of target images included in the reference image combination, and use the group of images as a reference image combination that meets a set similarity condition with the image combination to be processed.

[0137] In one embodiment, the selection unit is specifically used for:

[0138] Using a pre-trained reference image determination model, a reference image that meets a set similarity condition with a set image in each of the to-be-processed image combinations is selected from the target reference image sequence.

[0139] In one embodiment, the set quasi-image is a starting image, an intermediate image or an ending image of an image sequence corresponding to the combination of images to be processed.

[0140] In one embodiment, the image generation model includes L layers of residual convolutional networks for compression and L-1 layers of residual convolutional networks for image fusion;

[0141] The output of the previous residual convolutional network for compression is the input of the next residual convolutional network for compression;

[0142] The up-sampling result of the output result of the residual convolution network used for compression in the L layer and the output result of the residual convolution network used for compression in the L-1 layer are both used as the input of the residual convolution network used for image fusion in the L-1 layer;

[0143] The upsampling result of the output result of the previous residual convolution network used for image fusion and the output result of the residual convolution network used for compression at the target level are both used as inputs of the subsequent residual convolution network used for image fusion, and the output result of the residual convolution network used for compression at the target level has the same resolution as the upsampling result of the output result of the previous residual convolution network used for image fusion.

[0144] The technical solution of the medical image processing device provided by the embodiment of the present invention is to match the most similar reference image combination in the reference image sequence for each combination of images to be processed in the image sequence to be processed; upsample the image combination to be processed to obtain an upsampling result; since the upsampling result includes the global information of each image in the image combination to be processed, and each image in the reference image sequence includes high-resolution detail information, the pre-trained image generation model can combine the low-resolution prior global information of each image in the upsampling result with the high-resolution detail information of the reference image combination, and infer and generate a target image combination corresponding to each combination of images to be processed, which reduces the difficulty of model reasoning, helps to improve the accuracy of the target image combination, and thus improves the accuracy of the target image sequence determined based on each target image combination.

[0145] The medical image processing device provided in the embodiment of the present invention can execute the medical image processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0146] Figure 8 A schematic diagram of the structure of a model training device for training a reference image determination model provided by an embodiment of the present invention. Figure 8 As shown, the device comprises:

[0147] A first acquisition module 61 is used to acquire a first sample set, wherein the first sample set includes a set number of training image sequences for a target scanning part;

[0148] A first selection module 62, configured to randomly select an initial image combination from each of the training image sequences, and downsample each image in the initial image combination to obtain a training image combination, wherein the training image combination includes a first number of continuous images;

[0149] A downsampling module 63, configured to downsample the training image sequence to obtain a downsampling result, wherein the resolution of the downsampling result is the same as the resolution of the image combination to be processed;

[0150] The first training module 64 is used to input the downsampling result and each of the training image combinations into a reference image determination model to obtain a reference image in the reference image sequence that meets a set similarity condition with a set image in each of the training image combinations, and to optimize the parameters of the reference image determination model according to the similarity between each of the reference images and the corresponding set image to obtain a pre-trained reference image determination model.

[0151] The embodiment of the present invention simplifies the training process of the reference image determination model and improves the accuracy of the model training result by using the same image sequence as the training image sequence and the reference image sequence.

[0152] The model training device for training a reference image determination model provided in an embodiment of the present invention can execute the model training method for training a reference image determination model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0153] Fig. 9 A schematic diagram of the structure of a model training device for training an image generation model provided by an embodiment of the present invention. Fig. 9 As shown, the device comprises:

[0154] A second acquisition module 71, used to acquire a second sample set, wherein the second sample set includes a reference image sequence for a target scanning part and a set number of training image sequences;

[0155] A second selection module 72 is used to randomly select a training image combination from each of the training image sequences, and downsample the training image combination to obtain a target training image combination, wherein the training image combination includes a first number of continuous images;

[0156] A third selection module 73 is used to select, from the reference image sequence, a reference image combination that meets a set similarity condition with each of the target training image combinations, wherein the number of consecutive images in the reference image combination is greater than the first number;

[0157] An upsampling module 74, configured to upsample the target training image combination to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence;

[0158] The second training module 75 is used to input each of the upsampling results and the corresponding reference image combination into the image generation model to obtain a target image combination corresponding to each upsampling result, and according to the similarity between each of the target image combination and the corresponding training image combination, complete the parameter optimization of the image generation model to obtain a pre-trained image generation model.

[0159] The technical solution provided by the embodiment of the present invention is to obtain a target training image combination by downsampling the training image combination, and use the training image combination as the super-resolution reconstructed image true value corresponding to the target training image combination; select the image combination most similar to the target training image combination in the reference image sequence to obtain the reference image combination; upsample the target training image combination to obtain an upsampled result, and the upsampled result has the same resolution as the reference image combination; then input the reference image combination corresponding to each upsampled result into the image generation model, and the image generation model will obtain a predicted target image combination; then determine the similarity between the target image combination and the training image combination, and adjust the model parameters of the image generation model based on the similarity graph. The low-resolution training image combination and its corresponding true value are cleverly solved, which ensures the accuracy of the image generation model training method and the accuracy of the prediction results of the pre-trained image generation model.

[0160] The model training method device for training an image generation model provided in an embodiment of the present invention can execute the model training method for training an image generation model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0161] Fig.10 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0162] like Fig.10As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0163] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0164] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 performs the various methods and processes described above, such as a medical image processing method, or a model training method for training a reference image determination model, or a model training method for training an image generation model.

[0165] In some embodiments, the medical image processing method, or the model training method for training the reference image determination model, or the model training method for training the image generation model may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the medical image processing method described above, or the model training method for training the reference image determination model, or the model training method for training the image generation model may be executed. Alternatively, in other embodiments, the processor 11 may be configured to execute the medical image processing method, or the model training method for training the reference image determination model, or the model training method for training the image generation model, by any other appropriate means (e.g., by means of firmware).

[0166] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0167] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0168] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0169] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0170] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0171] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0172] An embodiment of the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements a medical image processing method as provided in any embodiment of the present application, or a model training method for training a reference image determination model, or a model training method for training an image generation model.

[0173] The computer program product may be implemented by writing computer program code for performing the operations of the present invention in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0174] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0175] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A medical image processing method, characterized in that: include: Dividing the sequence of images to be processed of the target scanning part into at least two combinations of images to be processed, wherein the combination of images to be processed includes a first number of continuous images; Selecting, from a pre-stored reference image sequence of the target scanning part, a reference image combination that meets a set similarity condition with each of the image combinations to be processed, wherein the number of consecutive images in the reference image combination is greater than the first number; Upsampling the combination of images to be processed to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence; Using a pre-trained image generation model to process each upsampling result and a corresponding reference image combination, determining a target image combination corresponding to each of the image combinations to be processed, wherein the resolution of the target image combination is the same as the resolution of the reference image sequence; According to all the target image combinations, a target image sequence corresponding to the image sequence to be processed is determined.

2. The method according to claim 1, characterized in that The step of selecting a reference image combination that meets a set similarity condition with each of the image combinations to be processed from a pre-stored reference image sequence of the target scanning part comprises: Downsampling a pre-stored reference image sequence of the target scanning part to obtain a target reference image sequence, wherein the resolution of the target reference image sequence is the same as the resolution of the image sequence to be processed; Selecting, from the target reference image sequence, a reference image that meets a set similarity condition with a set image in each of the combinations of images to be processed; Determining the position of the reference image in the reference image combination according to the position of the set image in the to-be-processed image combination; According to the position of the reference image in the reference image combination, the image identification of the reference image and the number of target images included in the reference image combination, a group of images is selected from the reference image sequence, and the group of images is used as a reference image combination that meets the set similarity condition with the image combination to be processed.

3. The method according to claim 2, characterized in that The step of selecting a reference image corresponding to a set image of each combination of the to-be-processed images from the target reference image sequence comprises: Using a pre-trained reference image determination model, a reference image that meets a set similarity condition with a set image in each of the to-be-processed image combinations is selected from the target reference image sequence.

4. The method according to claim 2, characterized in that: The set quasi-image is a starting image, an intermediate image or an ending image of an image sequence corresponding to the image combination to be processed.

5. The method according to claim 1, characterized in that The image generation model includes L layers of residual convolutional networks for compression and L-1 layers of residual convolutional networks for image fusion; The output of the previous residual convolutional network for compression is the input of the next residual convolutional network for compression; The up-sampling result of the output result of the residual convolution network used for compression in the L layer and the output result of the residual convolution network used for compression in the L-1 layer are both used as the input of the residual convolution network used for image fusion in the L-1 layer; The upsampling result of the output result of the previous residual convolution network used for image fusion and the output result of the residual convolution network used for compression at the target level are both used as inputs of the subsequent residual convolution network used for image fusion, and the output result of the residual convolution network used for compression at the target level has the same resolution as the upsampling result of the output result of the previous residual convolution network used for image fusion.

6. A model training method for training a reference image determination model, characterized in that: include: Acquire a first sample set, wherein the first sample set includes a set number of training image sequences for a target scanning part; Randomly selecting an initial image combination from each of the training image sequences, and downsampling each image in the initial image combination to obtain a training image combination, wherein the training image combination includes a first number of continuous images; Downsampling the training image sequence to obtain a downsampling result, wherein the resolution of the downsampling result is the same as the resolution of the training image combination; The downsampling result and each of the training image combinations are input into a reference image determination model to obtain a reference image in the reference image sequence that meets a set similarity condition with a set image in each of the training image combinations, and parameter optimization of the reference image determination model is completed according to the similarity between each of the reference images and the corresponding set image to obtain a pre-trained reference image determination model.

7. A model training method for training an image generation model, characterized in that: include Acquire a second sample set, wherein the second sample set includes a reference image sequence for a target scanning part and a set number of training image sequences; Randomly selecting a training image combination from each of the training image sequences, and downsampling the training image combination to obtain a target training image combination, wherein the training image combination includes a first number of continuous images; Selecting, from the reference image sequence, a reference image combination that meets a set similarity condition with each of the target training image combinations, wherein the number of consecutive images in the reference image combination is greater than the first number; Upsampling the target training image combination to obtain an upsampling result, wherein the resolution of the upsampling result is the same as the resolution of the reference image sequence; Each of the upsampling results and the corresponding reference image combination is input into the image generation model to obtain a target image combination corresponding to each upsampling result, and according to the similarity between each of the target image combination and the corresponding training image combination, the parameter optimization of the image generation model is completed to obtain a pre-trained image generation model.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the medical image processing method described in any one of claims 1 to 5, or the model training method for training a reference image determination model described in claim 6, or the model training method for training an image generation model described in claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which are used to enable a processor to implement the medical image processing method described in any one of claims 1 to 5, or the model training method for training a reference image determination model described in claim 6, or the model training method for training an image generation model described in claim 7 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program which, when executed by a processor, implements the medical image processing method according to any one of claims 1 to 5, or the model training method for training a reference image determination model according to claim 6, or the model training method for training an image generation model according to claim 7.

Images