Medical image super-resolution reconstruction method and device based on unsupervised algorithm

Abstract

The invention discloses a medical image super-resolution reconstruction method and device based on an unsupervised algorithm. The method comprises the following steps: acquiring a low-resolution medical image to be processed; inputting the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model; wherein the resolution reconstruction model is obtained by training a pre-constructed generator network and a pre-constructed discriminator network by using a sample image, and the sample image is generated by using a synthetic degradation algorithm. The problem of medical image blurring caused by mechanical motion, focusing and the like is solved, an unsupervised algorithm is adopted, manual annotation is not needed, the labor cost is greatly reduced, the efficiency is improved, the recognition effect of downstream tasks such as target detection, image classification, image segmentation and the like can be improved through the medical image reconstructed through the method, and the recognition efficiency is improved. The technical problem that identification is difficult due to the fact that medical images are not clear enough is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and medical data processing, in particular to a medical image super-resolution reconstruction method and device based on an unsupervised algorithm. BACKGROUND

[0002] The clarity of medical images is crucial for disease diagnosis and treatment. Clear medical images can help doctors accurately identify and locate lesions such as tumors, fractures, infections, or other pathological changes, and also help doctors more accurately measure the size and extent of lesions, which is very important for determining the severity of the disease and choosing a treatment plan.

[0003] However, in actual medical scenarios, medical images are not clear due to the following factors: 1) limitations of imaging technology, different imaging technologies (such as X-ray, CT, MRI, etc.) have different resolutions and contrasts, which directly affect the clarity of the image; 2) motion blur, if the patient or imaging equipment moves during imaging, it may cause image blur; 3) noise interference, image noise is one of the common problems in medical image processing, noise interference can cause loss of image details and affect doctors' judgment of the disease; 4) insufficient contrast, making it difficult to identify the details of the medical image and affecting doctors' judgment of the disease; 5) image artifacts, such as stripe artifacts, blur artifacts, step artifacts, etc., which may be caused by imaging equipment, patient movement, or image reconstruction algorithms; 6) improper image post-processing, such as image enhancement, denoising, etc., which can also affect the clarity of the image; 7) focusing problem, when the camera is not in optimal focus when taking medical images, it can cause image blur.

[0004] In the medical image super-resolution task, the size and quality of the training data set have an important impact on the performance of the algorithm. Due to the high cost of medical image annotation, uneven quality, data privacy, etc., and the clarity annotation is easily affected by subjective factors, it is difficult to obtain a large amount of high-quality training data. Based on the above reasons, the present application performs medical image super-resolution reconstruction based on an unsupervised algorithm. SUMMARY

[0005] To this end, the present application provides a medical image super-resolution reconstruction method and device based on an unsupervised algorithm, which solves the technical problem of difficulty in identification due to unclear medical images.

[0006] In order to achieve the above purpose, the embodiments of the present application provide the following technical solutions:

[0007] The present application provides a medical image super-resolution reconstruction method based on an unsupervised algorithm, the method comprising:

[0008] obtaining a low-resolution medical image to be processed;

[0009] inputting the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model;

[0010] The resolution reconstruction model is obtained by training a pre-constructed generator network and a discriminator network using sample images, and the sample images are generated by using a synthetic degradation algorithm.

[0011] In some embodiments, the synthetic degradation algorithm specifically includes:

[0012] The obtained real medical image is sequentially filtered, sampled, noise-added and image-compressed to obtain a degraded image, and the degraded image is used as the sample image.

[0013] In some embodiments, the expression of the synthetic degradation algorithm is:

[0014] x = D(y) = [(y * k) ↓ r +n] JPEG

[0015] Wherein, D represents the degradation process, y represents the real image, * represents the filtering operation, k represents the size of the filter operator, ↓ r represents down-sampling r times, n represents adding noise, and JPEG represents image compression.

[0016] In some embodiments, the generator network includes a convolutional layer, a Basic Block and an up-sampling layer; in the generator network, the Basic Block adopts a network architecture combining RRDB, multi-level residual network and dense connection, and the Basic Block is multiplied by a coefficient β before the main branch is added, where 0 < β < 1.

[0017] In some embodiments, the discriminator network adopts a Transunet structure, specifically including:

[0018] An encoder, which down-samples through convolution operation and combines a transformer structure to extract features containing more information by means of the global self-attention mechanism of the transformer;

[0019] A decoder, which is up-sampled to match the output dimension and the input.

[0020] In some embodiments, the encoder of Transunet discriminates global features, and the loss function is:

[0021]

[0022] wherein x represents a real image, z represents a low-resolution image, and G represents a generator.

[0023] In some embodiments, the decoder of Transunet discriminates local features, and the loss function of the decoder is :

[0024]

[0025] wherein x represents a real image, z represents a low-resolution image, and G represents a generator.

[0026] The application further provides a medical image super-resolution reconstruction device based on an unsupervised algorithm, and the device comprises:

[0027] An image acquisition unit is configured to acquire a low-resolution medical image to be processed.

[0028] A result generation unit is configured to input the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model.

[0029] The resolution reconstruction model is obtained by training a generator network and a discriminator network pre-constructed by using sample images, and the sample images are generated by using a synthetic degradation algorithm.

[0030] The application further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method as described above when executing the program.

[0031] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, and the computer program is executable on a processor to implement the steps of the method as described above.

[0032] The medical image super-resolution reconstruction method and device based on an unsupervised algorithm provided by the application can obtain a high-resolution medical image output by a resolution reconstruction model by inputting a low-resolution medical image to be processed into the resolution reconstruction model pre-trained, wherein the resolution reconstruction model is obtained by training a generator network and a discriminator network pre-constructed using sample images, and the sample images are generated using a synthetic degradation algorithm. Thus, the method and device provided by the application generate sample images using a synthetic degradation algorithm of unsupervised learning, do not need manual labeling, can obtain a large number of high-quality medical training images, train a deep learning generation algorithm using the data, can generate high-resolution clear images in multiple medical application scenarios, and thus can better judge a target in a downstream task. The technical problem of difficult recognition caused by unclear medical images is solved. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can be obtained from the provided drawings without creative labor.

[0034] The structures, proportions, sizes, etc. shown in the specification are only used to cooperate with the content disclosed in the specification, to be understood and read by those skilled in the art, and do not define the limiting conditions under which the application can be implemented, so they do not have technical significance. Any modification of the structure, change of the proportion relationship or adjustment of the size, without affecting the effect and purpose that the application can produce, should still fall within the scope of the technical content disclosed by the application.

[0035] Figure 1 The overall flowchart of the medical image super-resolution reconstruction method based on an unsupervised algorithm provided by the application;

[0036] Figure 2 The comparison effect diagram before and after the medical image super-resolution reconstruction method based on an unsupervised algorithm provided by the application is used for peripheral blood image reconstruction;

[0037] Figure 3 The comparison effect diagram before and after the medical image super-resolution reconstruction method based on an unsupervised algorithm provided by the application is used for microorganism image reconstruction;

[0038] Figure 4 The comparison effect diagram before and after the medical image super-resolution reconstruction method based on an unsupervised algorithm provided by the application is used for urine image reconstruction;

[0039] Figure 5 A flowchart for synthesizing a degraded image;

[0040] Figure 6 A structural diagram of a generator SRResNet;

[0041] Figure 7 A structural diagram of RRDB;

[0042] Figure 8 A structural diagram of a Dense Block;

[0043] Figure 9 A structural diagram of a generator discriminator;

[0044] Figure 10 A structural block diagram of the medical image super-resolution reconstruction device based on the unsupervised algorithm provided by the present application;

[0045] Figure 11 A structural block diagram of the computer device provided by the present application. DETAILED DESCRIPTION

[0046] The embodiments of the present application will be described in detail by specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0047] In the medical image super-resolution task, the size and quality of the training data set have an important influence on the performance of the algorithm. Due to the high labeling cost, strong subjectivity of labeling, uneven quality, and high data privacy of medical images, it is difficult to obtain a large amount of high-quality training data. In order to solve this problem, the present patent adopts an unsupervised learning method, which can obtain a large amount of high-quality medical training images without manual labeling. Using these data to train a deep learning generation algorithm can generate high-resolution clear images in multiple medical application scenarios.

[0048] In one specific embodiment, referring to Figure 1 The medical image super-resolution reconstruction method based on the unsupervised algorithm provided by the present application comprises the following steps:

[0049] S110: obtaining a low-resolution medical image to be processed;

[0050] S120: inputting the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model;

[0051] wherein the resolution reconstruction model is obtained by training a pre-constructed generator network and a discriminator network using sample images generated by a synthetic degradation algorithm.

[0052] It should be understood that the resolution reconstruction model is obtained by training, and the training process is an iterative process, each iteration including the following steps:

[0053] S1: Generator update;

[0054] S11: input low-resolution image:

[0055] A batch of low-resolution images z is randomly extracted from a dataset constructed from sample images.

[0056] S12: generate high-resolution image:

[0057] The low-resolution image z is input into the generator G to generate a high-resolution image G(z).

[0058] S13: update generator parameters by backpropagation:

[0059] The gradient of the generator loss L G is calculated, and the parameters of the generator are updated by backpropagation.

[0060] S2: Discriminator update;

[0061] S21: input real image and generated image:

[0062] A batch of real high-resolution images x is randomly extracted from the dataset;

[0063] The high-resolution image G(z) generated by the generator is used.

[0064] S22: calculate discriminator loss:

[0065] Global feature loss: using the encoder part of the discriminator, the global feature loss of the real image x and the generated image G(z) is calculated.

[0066] Local feature loss: using the decoder part of the discriminator, the local feature loss of the real image x and the generated image G(z) is calculated.

[0067] S23: update discriminator parameters by backpropagation:

[0068] The gradient of the discriminator loss is calculated, and the parameters of the discriminator are updated by backpropagation.

[0069] S3: repeat the training loop until the model converges or reaches a preset number of training rounds;

[0070] S4. Verification and testing;

[0071] Validation set evaluation: Evaluate model performance on the validation set and tune hyperparameters.

[0072] The quality of the generated images is evaluated using metrics such as PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity).

[0073] The generation and real images are compared visually to evaluate texture sharpness, shape prominence, and background noise reduction.

[0074] Test set evaluation: Test the model's generalization ability on the test set to ensure that the model can also perform well on unseen data.

[0075] In specific application scenarios, the algorithm proposed in this invention can be applied to various medical fields. It takes a low-resolution image as input and outputs a high-resolution image. Examples are given below for three medical fields: peripheral blood, microorganisms, and urine. Figure 2 , Figure 3 , Figure 4 After super-resolution reconstruction, the textures of white blood cells, microorganisms, and formed elements in urine are clearer and their shapes are more prominent. At the same time, background noise is reduced, which effectively helps doctors to identify target types more accurately and measure target size more precisely.

[0076] In some embodiments, the synthetic degradation algorithm specifically includes:

[0077] The acquired real medical images are sequentially filtered, sampled, have noise added, and compressed to obtain degraded images, which are then used as the sample images.

[0078] Specifically, during the data generation process, a low-resolution image is synthesized according to equation (1), simulating various degradation processes. Gaussian filtering and other filtering methods are used. Resizing can be performed using nearest neighbor interpolation, bilinear interpolation, etc., for upsampling or downsampling. When adding noise, Gaussian noise, Poisson noise, etc. can be added.

[0079] Furthermore, the degradation algorithm can be refined multiple times to simulate complex degradation processes closer to real-world scenarios, with different hyperparameters used in each iteration. A sinc filter is added to simulate ringing artifacts and overshoot artifacts found in actual medical settings. Data augmentation, such as rotation, flipping, and cropping, is applied to the generated low-resolution images to increase data diversity.

[0080] Specifically, the process of synthesizing degraded images is as follows: Figure 5 As shown, in the face of unknown and complex medical image degradation, low-resolution images can be synthesized according to formula (1).

[0081] x=D(y=[(y*k)↓ r +n] JPEG (1)

[0082] Where D represents the degradation process, y represents the true image, * represents the filtering operation, and k represents the size of the filtering operator. r This indicates downsampling by a factor of r, where n represents the addition of noise, and JPEG indicates JPEG compression.

[0083] See degradation process Figure 5 This includes 1) filtering, such as using Gaussian filtering, 2) resizing, such as upsampling or downsampling, using methods such as nearest neighbor interpolation and bilinear interpolation, 3) adding noise, such as adding Gaussian noise or Poisson noise, and 4) JPEG compression.

[0084] The above process can solve some practical problems, but when encountering more complex and diverse degradation phenomena in real-world scenarios, such as camera imaging, image editing, network transmission, sensor noise, and unknown noise, the low-resolution images synthesized by the above process still have a significant gap from the degradation images in real-world scenarios. Degradation images in real-world scenarios are usually a combination of multiple different degradation processes. Such complex degradation processes cannot be simulated by a single degradation algorithm. Therefore, multiple degradation algorithms are needed to simulate medical degradation images that are closer to real-world scenarios. Each degradation algorithm uses different hyperparameters. In addition, to simulate ringing artifacts, overshoot artifacts, etc. in real-world medical scenarios, an additional sinc filter is added, which effectively improves the algorithm performance.

[0085] In some embodiments, the generator network includes convolutional layers, a basic block, and an upsampling layer; in the generator network, the basic block adopts RRDB (Residual-in-Residual Dense Block), which is a network architecture that combines multi-level residual networks and dense connections. In order to ensure the stability of the algorithm, it is multiplied by a coefficient β before being added to the main branch, where 0 < β < 1.

[0086] Specifically, the structure of generator G is as follows:

[0087] Convolutional layer: Extracts features from the input image;

[0088] Basic Block (RRDB): Extracts richer features through multi-level residual networks and dense connections;

[0089] Upsampling layer: Upsamples the low-resolution feature map to a high-resolution feature map.

[0090] Specifically, medical image generators utilize, for example Figure 6The SRResNet structure shown consists of convolution, Basic Block, and upsampling. Most of the computation is concentrated in the low-resolution feature space. The Basic Block can be selected from Residual Block, Dense Block, RRDB, etc. Figure 8 The Dense Block shown is as follows: Figure 7 The RRDB shown is an example.

[0091] Compared to other generator algorithms, the following two main structural improvements were made:

[0092] 1) Since the BN layer uses the mean and variance of the features within the batch to normalize the features during training, and uses the mean and variance estimated from the entire training set during inference, the BN layer may cause artifacts when the distribution of the training data and the test data differs greatly, thus limiting the ability to generate images. Therefore, all BN layers are removed.

[0093] 2) The Basic Block uses Residual-in-Residual Dense Block (RRDB), see... Figure 7 It combines a multi-level Residual Network with dense connections. To ensure the algorithm's stability, a β coefficient is applied before adding the main branch. The β coefficient ranges from 0 to 1. The structure of the Dense Block is shown in [see section 1]. Figure 8 Each layer of this structure is connected to the previous layer, and the feature maps of the previous layers are directly passed to the subsequent layers. This design helps with the transmission of information and the reuse of features, while reducing the number of parameters and alleviating the gradient vanishing problem to some extent.

[0094] In some embodiments, the discriminator network adopts a Transunet architecture, specifically including:

[0095] The encoder downsamples through convolution operations and combines them with a transformer structure, leveraging the transformer's global self-attention mechanism to extract features containing more information.

[0096] The decoder upsamples the data to match the output dimension with the input dimension.

[0097] Specifically, conventional discriminators only consider global features and the details of the generated image. This patent adds the discrimination of local features, replacing the VGG-type discriminator with Transunet. Transunet can discriminate not only global features but also local features, providing the generator with pixel-level detailed information. The discriminator structure is updated through a gradient backpropagation algorithm. (See attached diagram).Figure 9 .

[0098] The Transunet discriminator consists of an encoder and a decoder. The encoder downsamples through convolution operations and incorporates a transformer structure. With the help of the transformer's global self-attention mechanism, it can extract features containing more information. The decoder upsamples the input and matches the output dimension to the input. In addition, Skip Connections fuse the encoder's features into the decoder, further enhancing the discriminative ability of the neural network.

[0099] Let the discriminator be denoted as D. TU Its encoder module is denoted as Its decoder module is denoted as The following sections will elaborate on the identification of global features and local features:

[0100] 1) Determine global features

[0101] The global feature discriminant uses Transunet's encoder to classify whether an image is real or generated; the loss function is denoted as... See equation (2) below:

[0102]

[0103] Where x represents the real image, z represents the low-resolution image, and G represents the generator.

[0104] 2) Identify local features

[0105] The local feature discrimination uses Transunet's decoder to distinguish between real and generated images at the pixel level, and the loss function is denoted as... See the following formula (3):

[0106]

[0107] Based on the improved discriminator, the generator focuses on both global and local features, and the loss function is shown in equation (4):

[0108]

[0109] In the above specific embodiments, the medical image super-resolution reconstruction method based on unsupervised algorithms provided by the present invention obtains a low-resolution medical image to be processed, inputs the low-resolution medical image into a pre-trained resolution reconstruction model, and obtains a high-resolution medical image output by the resolution reconstruction model. The resolution reconstruction model is obtained by training a pre-constructed generator network and discriminator network using sample images, which are generated using a synthetic degradation algorithm. Thus, the method provided by the present invention uses an unsupervised learning synthetic degradation algorithm to generate sample images, eliminating the need for manual annotation and obtaining a large number of high-quality medical training images. Using these data to train a deep learning generation algorithm, high-resolution, clear images can be generated in multiple medical application scenarios, thereby enabling more accurate identification of target types and more precise measurement of target size in medical images. This invention solves the problem of blurred medical images caused by mechanical movement and focusing, making medical images clearer and more realistically restoring the microscopic view. This allows doctors to see the contents more clearly and make accurate judgments. Since training data for medical images is difficult to obtain, an unsupervised algorithm is adopted, eliminating the need for manual annotation, greatly reducing labor costs and improving efficiency. The medical images reconstructed by this invention can improve the recognition effect of downstream tasks, such as object detection, image classification, and image segmentation, solving the technical problem of recognition difficulties caused by insufficient clarity of medical images.

[0110] In addition to the methods described above, this invention also provides a medical image super-resolution reconstruction device based on an unsupervised algorithm, such as... Figure 10 As shown, the device includes:

[0111] Image acquisition unit 1010 is used to acquire low-resolution medical images to be processed;

[0112] The result generation unit 1020 is used to input the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model.

[0113] The resolution reconstruction model is obtained by training a pre-built generator network and a discriminator network using sample images, which are generated using a synthetic degradation algorithm.

[0114] In some embodiments, the synthetic degradation algorithm specifically includes:

[0115] The acquired real medical images are sequentially filtered, sampled, have noise added, and compressed to obtain degraded images, which are then used as the sample images.

[0116] In some embodiments, the expression for the synthetic degradation algorithm is:

[0117] x=D(y=[(y*k)↓ r +n] JPEG

[0118] Where D represents the degradation process, y represents the true image, * represents the filtering operation, and k represents the size of the filtering operator. r This indicates downsampling by a factor of r, where n represents the addition of noise, and JPEG represents image compression.

[0119] In some embodiments, the generator network includes convolutional layers, a basic block, and an upsampling layer; in the generator network, the basic block adopts RRDB, which combines a multi-level residual network and a densely connected network architecture, and is multiplied by a coefficient β before being added to the main branch, where 0 < β < 1.

[0120] In some embodiments, the discriminator network adopts a Transunet architecture, specifically including:

[0121] The encoder downsamples through convolution operations and combines them with a transformer structure, leveraging the transformer's global self-attention mechanism to extract features containing more information.

[0122] The decoder upsamples the data to match the output dimension with the input dimension.

[0123] In some embodiments, Transunet's encoder is used to discriminate global features, and its loss function is... for:

[0124]

[0125] Where x represents the real image, z represents the low-resolution image, and G represents the generator.

[0126] In some embodiments, the Transunet decoder is used to discriminate local features, and its loss function is... for:

[0127]

[0128] Where x represents the real image, z represents the low-resolution image, and G represents the generator.

[0129] In the above specific embodiments, the medical image super-resolution reconstruction device based on unsupervised algorithms provided by the present invention acquires a low-resolution medical image to be processed, inputs the low-resolution medical image into a pre-trained resolution reconstruction model, and obtains a high-resolution medical image output by the resolution reconstruction model. The resolution reconstruction model is obtained by training a pre-constructed generator network and discriminator network using sample images, which are generated using a synthetic degradation algorithm. Thus, the device provided by the present invention uses an unsupervised learning synthetic degradation algorithm to generate sample images, eliminating the need for manual annotation and obtaining a large number of high-quality medical training images. Using these data to train a deep learning generation algorithm, high-resolution, clear images can be generated in multiple medical application scenarios, enabling more accurate identification of target types and more precise measurement of target size in medical images. This solves the technical problem of difficulty in identification due to insufficient clarity in medical images.

[0130] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and model predictions. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The model predictions of the computer device store static and dynamic information data. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0131] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0132] Corresponding to the above embodiments, this invention also provides a computer storage medium containing one or more program instructions. These one or more program instructions are used to execute the method described above.

[0133] The present invention also provides a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and the computer being able to perform the above-described method when the computer program is executed by a processor.

[0134] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0135] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0136] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0137] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0138] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0139] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0140] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0141] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A medical image super-resolution reconstruction method based on an unsupervised algorithm, characterized in that, The method comprises: acquiring a low-resolution medical image to be processed; inputting the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model; wherein the resolution reconstruction model is obtained by training a pre-constructed generator network and discriminator network using sample images, and the sample images are generated using a synthetic degradation algorithm.

2. The method of claim 1, wherein the method is based on an unsupervised algorithm. The synthetic degradation algorithm specifically comprises: filtering, sampling, adding noise, and image compression on the acquired real medical image in sequence to obtain a degraded image, and taking the degraded image as the sample image.

3. The method of claim 2, wherein the method further comprises: The expression of the synthetic degradation algorithm is: x = D(y) = [(y * k) V r + n] JPEG where D denotes a degradation process, y denotes a real image, * denotes a filtering operation, k denotes a size of a filter operator, ↓ r denotes a down-sampling r times, n denotes a noise addition, and JPEG denotes an image compression.

4. The method of claim 1, wherein the method is based on an unsupervised algorithm. The generator network comprises a convolution layer, a Basic Block, and an up-sampling layer; in the generator network, the Basic Block adopts a network architecture combining RRDB, a multi-level residual network, and dense connection, and the Basic Block is multiplied by a coefficient β before the main branch is added, where 0 < β < 1.

5. The method of claim 1, wherein the method further comprises: The discriminator network adopts a Transunet structure, specifically comprising: an encoder that extracts features containing more information through convolution operation down-sampling and combining a transformer structure and global self-attention mechanism of the transformer; a decoder that is up-sampled to match the output dimension and the input.

6. The method of claim 5, wherein the method further comprises: The global features are discriminated using an encoder of Transunet, whose loss function is: wherein x represents a real image, z represents a low-resolution image, and G represents a generator.

7. The method of claim 5, wherein the method further comprises: The decoder using Transunet discriminates local features, and its loss function is: wherein x represents a real image, z represents a low-resolution image, and G represents a generator.

8. An apparatus for medical image super-resolution reconstruction based on an unsupervised algorithm, characterized in that, The device comprises: an image acquisition unit configured to acquire a low-resolution medical image to be processed; a result generation unit configured to input the low-resolution medical image into a pre-trained resolution reconstruction model to obtain a high-resolution medical image output by the resolution reconstruction model; wherein the resolution reconstruction model is obtained by training a pre-constructed generator network and discriminator network using sample images, and the sample images are generated using a synthetic degradation algorithm.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the method according to any one of claims 1-7 when executing the program. 10.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the steps of the method according to any one of claims 1-7 when executed by the processor.

Images