Image quality improvement method, device, computer equipment and storage medium

Feature extraction and integration processing of medical images through neural network models solves the problem of difficult to balance image quality improvement in traditional technology, and improves the resolution and signal-to-noise ratio of medical images, and adapts to image processing of different noise levels.

CN112767260BActive Publication Date: 2024-11-26SHANGHAI UNITED IMAGING INTELLIGENCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011621693.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-30
Publication Date
2024-11-26
Estimated Expiration
2040-12-30

AI Technical Summary

Technical Problem

When improving image quality, traditional medical imaging technology is difficult to find an equalization point between retaining details and removing noise, and different parameters need to be set according to the noise level, resulting in a decrease in image quality and a decrease in signal-to-noise ratio.

Method used

Using the neural network model, including the first processing network, the automatic path selection network and the second processing network, the feature extraction and integration processing of medical images is performed, the image resolution and signal-to-noise ratio are improved, and the neural network model parameters are adjusted through training methods to achieve Best results.

Benefits of technology

It achieves improving image quality while improving medical image resolution and signal-to-noise ratio. It is more effective than traditional methods and can adapt to image processing of different noise levels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112767260B_ABST
    Figure CN112767260B_ABST
Patent Text Reader

Abstract

The present application relates to an image quality improvement method, device, computer equipment and storage medium. The method comprises: obtaining a medical image to be processed; inputting the medical image into a preset neural network model, processing the medical image through the neural network model, and obtaining a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image. The present method can improve the resolution of the medical image while improving the contrast of the medical image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of medical imaging technology, and in particular to an image quality improvement method, apparatus, computer equipment and storage medium. Background Art

[0002] Medical images can display the internal tissues of the human body. They are usually acquired by medical imaging equipment in a non-invasive manner from the human body or a part of the human body to obtain the corresponding medical images of the internal tissues of the human body.

[0003] In order to make medical images more widely used in clinical applications, new acceleration methods have emerged in recent years, but at the same time, they have also brought about different degrees of image quality degradation, such as decreased signal-to-noise ratio and blurred images. In traditional technology, filters are mainly used to reduce the noise caused by data acquisition or image reconstruction, or image enhancement algorithms are used to improve the contrast of medical images. However, in traditional technology, filters often have the same denoising effect on the entire image, and it is difficult to find a balance between retaining image details and removing noise. In the process of using filters for denoising, different parameters need to be set for images with different noise levels to remove different degrees of noise; when using image enhancement algorithms to improve the contrast of medical images, different enhancement parameters need to be set for images with different noise levels to remove different degrees of noise. Summary of the invention

[0004] Based on this, it is necessary to provide an image quality improvement method, device, computer equipment and storage medium that can improve the contrast of medical images while improving the resolution of medical images in order to address the above technical problems.

[0005] A method for improving image quality, the method comprising:

[0006] Acquiring a medical image to be processed;

[0007] The medical image is input into a preset neural network model, and the medical image is processed by the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0008] In one embodiment, the neural network model includes a first processing network, an automatic path selection network, and a second processing network; the step of inputting the medical image into a preset neural network model, and processing the medical image by the neural network model to obtain a processed medical image includes:

[0009] Inputting the medical image into the first processing network, and performing first feature extraction on the medical image through the first processing network to obtain a first feature of the medical image;

[0010] Inputting the first feature of the medical image into the automatic path selection network, and performing second feature extraction on the first feature of the medical image through the automatic path selection network to obtain the second feature of the medical image;

[0011] The second feature of the medical image is input into the second processing network, and the second feature is integrated and convolved by the second processing network to obtain the processed medical image.

[0012] In one embodiment, the automatic path selection network includes a basic subnetwork and a leaf subnetwork; the basic subnetwork is used to perform one or more path processing on the first feature of the medical image; the leaf subnetwork is used to perform feature fusion on the output of the basic subnetwork to obtain the second feature of the medical image; wherein the number of the leaf subnetworks corresponds to the number of the last layer of the basic subnetwork.

[0013] In one embodiment, the training method of the neural network model includes:

[0014] Obtaining sample medical images;

[0015] Inputting the sample medical image into a preset initial neural network model, and obtaining a processed sample medical image through the initial neural network model;

[0016] Obtaining a value of a loss function of the initial neural network model according to the sample medical image and the processed sample medical image;

[0017] According to the value of the loss function of the initial neural network model, the parameters of the initial neural network model are adjusted to obtain the neural network model.

[0018] In one embodiment, the training method of the neural network model includes:

[0019] Acquire a sample medical image and a gold standard image corresponding to the sample medical image; the resolution of the gold standard image is greater than the resolution of the sample medical image; the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image;

[0020] Inputting the sample medical image into a preset initial neural network model, and obtaining a processed sample medical image through the initial neural network model;

[0021] Obtaining a value of a loss function of the initial neural network model according to the processed sample image and the gold standard image;

[0022] According to the value of the loss function of the initial neural network model, the parameters of the initial neural network model are adjusted to obtain the neural network model.

[0023] In one embodiment, acquiring a sample medical image and a gold standard image corresponding to the sample medical image includes:

[0024] Controlling the first scanner to scan the subject once, and determining the obtained scanned image as the sample medical image;

[0025] Control the first scanner to scan the subject multiple times, average the images obtained by the multiple scans, and obtain a gold standard image corresponding to the sample medical image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

[0026] In one embodiment, acquiring a sample medical image and a gold standard image corresponding to the sample medical image includes:

[0027] Controlling the first scanner to scan the subject, and determining the obtained scanned image as the sample medical image;

[0028] The second scanner is controlled to scan the subject, and the obtained scanned image is determined as the gold standard image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

[0029] An image quality improvement device, the device comprising:

[0030] A first acquisition module, used for acquiring a medical image to be processed;

[0031] The first processing module is used to input the medical image into a preset neural network model, and process the medical image through the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0032] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0033] Acquiring a medical image to be processed;

[0034] The medical image is input into a preset neural network model, and the medical image is processed by the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0036] Acquiring a medical image to be processed;

[0037] The medical image is input into a preset neural network model, and the medical image is processed by the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0038] The above-mentioned image quality improvement method, device, computer equipment and storage medium, by inputting the medical image to be processed into a preset neural network model, can process the medical image to be processed through the neural network model to obtain a processed medical image, wherein the resolution of the processed medical image is greater than the resolution of the medical image to be processed, and the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image to be processed. Compared with the traditional image quality improvement method, this method improves the signal-to-noise ratio of the medical image while improving the resolution of the medical image through the neural network model, thereby improving the quality of the obtained processed medical image. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A diagram showing an application environment of a method for improving image quality in an embodiment;

[0040] Figure 2 is a flow chart of a method for improving image quality in one embodiment;

[0041] Figure 3 is a flow chart of a method for improving image quality in one embodiment;

[0042] Figure 4 A schematic diagram of the structure of a neural network model preset in one embodiment;

[0043] Figure 5 A schematic diagram of the structure of a basic sub-network of a neural network model preset in one embodiment;

[0044] Figure 6 is a flow chart of a method for improving image quality in one embodiment;

[0045] Figure 7 is a flow chart of a method for improving image quality in one embodiment;

[0046] Figure 8 FIG. 4 is a structural block diagram of an image quality improvement device in an embodiment. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] The image quality improvement method provided in the embodiment of the present application can be applied to Figure 1 The computer device shown. The computer device includes a processor and a memory connected by a system bus, and a computer program is stored in the memory. When the processor executes the computer program, the steps of the following method embodiment can be executed. Optionally, the computer device may also include a network interface, a display screen and an input device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory, and the non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. Optionally, the computer device can be a server, a personal computer, a personal digital assistant, or other terminal devices, such as a tablet computer, a mobile phone, etc., or a cloud or remote server. The embodiment of the present application does not limit the specific form of the computer device.

[0049] In one embodiment, Figure 2 As shown, a method for improving image quality is provided, and the method is applied to Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0050] S201, obtaining a medical image to be processed.

[0051] The medical image to be processed is a low-quality medical image with low resolution and signal-to-noise ratio. Optionally, the medical image to be processed may be a magnetic resonance imaging (MRI) image or other medical imaging image. Optionally, the medical image to be processed may be a brain image of the scanned object, a knee joint image of the scanned object, or a medical image of other parts of the scanned object.

[0052] Specifically, the computer device acquires the medical image to be processed. Optionally, the computer device may acquire the medical image to be processed from a PACS (Picture Archiving and Communication Systems) server, or may acquire the medical image to be processed in real time from a corresponding medical imaging device. Optionally, after acquiring the medical image to be processed, the computer device may pre-process the medical image to be processed to remove interference caused by the working condition of the medical imaging device on the medical image, wherein the pre-processing includes resampling, resizing, and grayscale normalization.

[0053] S202, inputting the medical image into a preset neural network model, processing the medical image through the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0054] Specifically, the computer device inputs the above-mentioned medical image to be processed into a preset neural network model, and processes the above-mentioned medical image to be processed through the neural network model to obtain a processed medical image, wherein the resolution of the processed medical image is greater than the resolution of the above-mentioned medical image to be processed, and the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the above-mentioned medical image to be processed. Among them, the preset neural network model can be obtained by training a preset initial neural network model using a large number of sample medical images and gold standard images corresponding to the sample medical images. It can be understood that the resolution of the gold standard image corresponding to the sample medical image is greater than the resolution of the sample medical image, and the signal-to-noise ratio of the gold standard image corresponding to the sample medical image is greater than the signal-to-noise ratio of the sample medical image. Optionally, the preset neural network model can be a path automatic selection network model, or it can be a V-Net network, or a DenseNet network. It should be noted here that the preset neural network model in the embodiment of the present application is an automatic path selection network model.

[0055] In the above-mentioned image quality improvement method, the computer device inputs the medical image to be processed into a preset neural network model, and the neural network model can process the medical image to be processed to obtain a processed medical image, wherein the resolution of the processed medical image is greater than the resolution of the medical image to be processed, and the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image to be processed. Compared with the traditional image quality improvement method, this method improves the resolution of the medical image to be processed while also improving the signal-to-noise ratio of the medical image to be processed through the neural network model, thereby improving the quality of the processed medical image.

[0056] In the above-mentioned scenario where the medical image to be processed is input into a preset neural network model, and the medical image to be processed is processed by the preset neural network model to obtain a processed medical image, in one embodiment, the preset neural network model is an automatic path selection network model, and the preset neural network model includes a first processing network, an automatic path selection network, and a second processing network, such as Figure 3 As shown, the above S202 includes:

[0057] S301, inputting a medical image into a first processing network, performing first feature extraction on the medical image through the first processing network, and obtaining a first feature of the medical image.

[0058] Specifically, the computer device inputs the above-mentioned medical image to be processed into the first processing network of the preset neural network model, and performs a first feature extraction on the medical image to be processed through the first processing network of the preset neural network model to obtain the first feature of the medical image to be processed. Optionally, the first processing network can be a feature extraction network. It can be understood that the first feature extraction here is a preliminary feature extraction of the medical image to be processed. Optionally, the first processing network can also process the extracted first feature of the medical image into a suitable size for the automatic path selection network of the neural network model while performing feature extraction on the medical image to be processed.

[0059] S302, inputting the first feature of the medical image into an automatic path selection network, and extracting the second feature of the first feature of the medical image through the automatic path selection network to obtain the second feature of the medical image.

[0060] Specifically, the computer device inputs the first feature of the medical image to be processed obtained above into the automatic path selection network of a preset neural network model, performs second feature extraction on the first feature of the medical image to be processed through the automatic path selection network, and extracts the second feature of the medical image to be processed from the first feature of the medical image to be processed.

[0061] S303, inputting the second feature of the medical image into the second processing network, and performing integration processing and convolution processing on the second feature through the second processing network to obtain a processed medical image.

[0062] Specifically, the computer device inputs the second feature of the medical image to be processed obtained above into the second processing network of the preset neural network model, and integrates and convolves the second feature of the medical image to be processed through the second processing network to obtain the processed medical image. It should be noted here that the second processing network mainly integrates the single-level or multi-level outputs of the automatic path selection network, integrates and convolves the second feature at the feature layer that meets the output requirements of the second processing network, and obtains the processed medical image.

[0063] In this embodiment, the computer device inputs the medical image to be processed into the first processing network of the preset neural network model, and can perform a first feature extraction on the medical image to be processed through the first processing network to obtain the first feature of the medical image to be processed, so that the first feature of the medical image to be processed can be input into the automatic path selection network of the preset neural network model, and the second feature extraction of the first feature of the medical image can be further performed through the automatic path selection network to obtain the second feature of the medical image to be processed, and then the second feature of the medical image can be input into the second processing network of the preset neural network model, and the second feature of the medical image can be integrated and convolved through the second processing network, thereby improving the resolution of the medical image to be processed and the signal-to-noise ratio of the medical image to be processed, thereby improving the quality of the processed medical image.

[0064] In the above-mentioned automatic path selection network that inputs the first feature of the medical image into a preset neural network model, and in the scenario where the second feature of the first feature of the medical image to be processed is extracted through the automatic path selection network, the above-mentioned automatic path selection network includes a basic subnetwork and a leaf subnetwork; the basic subnetwork is used to perform one or more path processing on the first feature of the medical image; the leaf subnetwork is used to perform feature fusion on the outputs of each basic subnetwork to obtain the second feature of the medical image; wherein the number of leaf subnetworks corresponds to the number of the last layers of the basic subnetwork.

[0065] Specifically, Figure 4As shown, the automatic path selection network of the above-mentioned preset neural network model includes a basic subnetwork and a leaf subnetwork, wherein the basic subnetwork is used to perform one or more path processing on the first feature of the medical image to be processed, and the leaf subnetwork is used to perform feature fusion on the output of the basic subnetwork to obtain the second feature of the above-mentioned medical image to be processed; the number of the above-mentioned leaf subnetworks corresponds to the number of the last layer of the above-mentioned basic subnetworks. Optionally, the basic subnetwork can be one or more, and correspondingly, the leaf subnetwork can also be one or more. When there are multiple basic subnetworks, each basic subnetwork has the same structure, and when there are multiple leaf subnetworks, each leaf subnetwork has the same structure. Optionally, as Figure 5 As shown in the figure, the basic sub-network can be composed of a fusion module, a gating module and multiple sub-network modules. The fusion module is responsible for fusing the outputs of multiple basic modules and can be composed of common neural network modules, such as a residual module (resblock), a depthwise separable convolution module, etc.; the gating module is composed of a convolution layer, a global pooling and an activation function. It should be noted here that the one or more path processing of the first feature of the medical image to be processed by the basic sub-network refers to generating multiple weight values ​​between 0 and 1 through the gating module of the basic sub-network, assigning different weights to the outputs of each basic sub-network, and controlling the weights of different paths (i.e., different basic sub-networks) through different weight values; the sub-network module selects various functional network blocks to improve network performance, such as upsample block, downsample block, SE block (Sequeze and Excitation The number of sub-network modules can be adjusted according to actual needs. Each sub-network module is equivalent to a different branch. The weight of the gating module can represent the different branches of each sub-network module. The leaf sub-network only has a fusion module, and the output of the leaf sub-network is the output of the automatic path selection network.

[0066] In this embodiment, since the automatic path selection network of the preset neural network model includes a basic sub-network and a leaf sub-network, the basic sub-network can perform one or more path processing on the first feature of the medical image, and the leaf sub-network can perform feature fusion on the output of the basic sub-network, thereby obtaining the second feature of the medical image to be processed, and then the second feature of the medical image to be processed can be input into the second processing network of the preset neural network model to obtain a processed medical image with higher resolution and signal-to-noise ratio. While improving the resolution of the medical image to be processed, the signal-to-noise ratio of the medical image to be processed is also improved, thereby improving the quality of the processed medical image.

[0067] In the above scenario where the medical image to be processed is input into a preset neural network model, the preset neural network model is a pre-trained neural network model. When the training sample includes the gold standard image, the training method of the neural network model is a supervised training method. When the training sample does not include the gold standard image, the training method of the neural network model is an unsupervised training method. These two training methods will be described in detail below.

[0068] When the training method of the above-mentioned preset neural network model is an unsupervised training method, in one embodiment, Figure 6 As shown, the training method of the above preset neural network model includes:

[0069] S601, obtaining a sample medical image.

[0070] Specifically, the computer device acquires a sample medical image. Optionally, the sample medical image may be a magnetic resonance (MRI) image, or other medical imaging image. Optionally, the computer device may control the first scanner to scan the tester once, and determine the obtained scanned image as the above-mentioned sample medical image, or control the second scanner to scan the tester to obtain a scanned image, and process the obtained scanned image to obtain the above-mentioned sample medical image; wherein the scanning intensity of the second scanner is greater than the scanning intensity of the first scanner, that is, the sample medical image acquired by the computer device is an image with low resolution and signal-to-noise ratio. Exemplarily, taking the sample medical image as a magnetic resonance image as an example, the computer device may control a low-field instrument to scan the tester to obtain a sample medical image with low signal-to-noise ratio and low resolution, or may control a high-field instrument to scan the tester to obtain a high-signal-to-noise ratio high-resolution image, and then process the high-signal-to-noise ratio high-resolution image to obtain the above-mentioned sample medical image, and optionally, the computer device may perform noise reduction and downsampling on the high-signal-to-noise ratio high-resolution image to obtain a sample medical image with low signal-to-noise ratio and low resolution.

[0071] S602, inputting the sample medical image into a preset initial neural network model, and obtaining a processed sample medical image through the initial neural network model.

[0072] Specifically, the computer device inputs the acquired sample medical image into a preset initial neural network model, and obtains a processed sample medical image through the initial neural network model. Optionally, the resolution of the processed sample medical image can be greater than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image can be greater than the signal-to-noise ratio of the sample medical image. Optionally, the resolution of the processed sample medical image can be less than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image can be less than the signal-to-noise ratio of the sample medical image, and then by continuously training the preset initial neural network model, the resolution of the processed sample medical image finally output by the neural network model is greater than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image is greater than the signal-to-noise ratio of the sample medical image. It can be understood that the preset initial neural network model has the same network structure as the above-mentioned neural network model, that is, the preset initial neural network model also includes a first processing network, an automatic path selection network, and a second processing network. For the network structure and principle of the initial neural network model, please refer to the above-mentioned description of the neural network model, which will not be repeated in this embodiment.

[0073] S603, obtaining a value of a loss function of the initial neural network model according to the sample medical image and the processed sample medical image.

[0074] Specifically, the computer device obtains the value of the loss function of the initial neural network model based on the sample medical image and the processed sample medical image. Optionally, the computer device may perform noise reduction and downsampling on the processed sample medical image, and obtain the value of the loss function of the initial neural network model based on the pixel value of each pixel of the sample medical image that has undergone noise reduction and downsampling and the pixel value of each pixel of the processed sample medical image obtained above. Optionally, the loss function of the initial neural network model may be a Dice loss function or a Focal loss function.

[0075] S604, adjusting the parameters of the initial neural network model according to the value of the loss function of the initial neural network model to obtain a neural network model.

[0076] Specifically, the computer device adjusts the parameters of the initial neural network model according to the value of the loss function of the initial neural network model obtained above to obtain the above neural network model. Optionally, the computer device may adjust the parameters of the initial neural network model according to the value of the loss function of the initial neural network model until the value of the loss function of the initial neural network model reaches a stable value or a minimum value, and the initial neural network model when the value of the loss function of the initial neural network model reaches a stable value or a minimum value is determined as the above preset neural network model.

[0077] In this embodiment, the computer device obtains a large number of sample medical images, inputs the sample medical images into a preset initial neural network model, and can obtain processed sample medical images through the initial neural network model, so that the value of the loss function of the initial neural network model can be obtained according to the sample medical images and the processed sample medical images, and then the parameters of the initial neural network model can be adjusted according to the value of the loss function of the initial neural network model. Since the computer device trains the initial neural network model through a large number of sample medical images, the number of sample medical images is large, so the initial neural network model can be trained more accurately, thereby improving the accuracy of the obtained neural network model.

[0078] When the training method of the above-mentioned preset neural network model is a supervised training method, in one embodiment, Figure 7 As shown, the training method of the above preset neural network model includes:

[0079] S701, obtaining a sample medical image and a gold standard image corresponding to the sample medical image; the resolution of the gold standard image is greater than the resolution of the sample medical image; the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

[0080] Specifically, the computer device obtains a sample medical image and a gold standard image corresponding to the sample medical image. The resolution of the gold standard image corresponding to the sample medical image is greater than the resolution of the sample medical image, and the signal-to-noise ratio of the gold standard image corresponding to the sample medical image is greater than the signal-to-noise ratio of the sample medical image. Optionally, the sample medical image may be a magnetic resonance imaging (MRI) image, or other medical imaging image. Optionally, the computer device may control the first scanner to scan the subject once, determine the scanned image obtained as the sample medical image, control the first scanner to scan the subject multiple times, average the images obtained by the multiple scans, and determine the averaged image as the gold standard image corresponding to the sample medical image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image. Optionally, the computer device may also control the first scanner to scan the test subject, and determine the obtained scanned image as the above-mentioned sample medical image; control the second scanner to scan the test subject, and determine the obtained scanned image as the gold standard image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image. Exemplarily, taking the sample medical image as a magnetic resonance image as an example, the computer device may optionally control the low-field instrument (first scanner) to scan the test subject once, and determine the obtained low-signal-to-noise ratio and low-resolution scanned image as the above-mentioned sample medical image, and control the low-field instrument (first scanner) to scan the test subject multiple times, average the images obtained by the multiple scans, and determine the images after the multiple average processing as the gold standard image corresponding to the above-mentioned sample medical image; optionally, the computer device may also control the low-field instrument (first scanner) to scan the test subject, and obtain a sample medical image with a low signal-to-noise ratio and low resolution, and control the high-field instrument (second scanner) to scan the test subject, and obtain a high-signal-to-noise ratio and high-resolution image, and use the high-signal-to-noise ratio and high-resolution image as the gold standard image corresponding to the sample medical image.

[0081] S702, inputting the sample medical image into a preset initial neural network model, and obtaining a processed sample medical image through the initial neural network model.

[0082] Specifically, the computer device inputs the acquired sample medical image into a preset initial neural network model, and obtains the processed sample medical image through the initial neural network model. Optionally, the resolution of the processed sample medical image may be greater than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image may be greater than the signal-to-noise ratio of the sample medical image. Optionally, the resolution of the processed sample medical image may be less than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image may be less than the signal-to-noise ratio of the sample medical image. By continuously training the preset initial neural network model, the resolution of the processed sample medical image finally output by the neural network model is greater than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image is greater than the signal-to-noise ratio of the sample medical image. It can be understood that the preset initial neural network model has the same network structure as the above-mentioned neural network model, that is, the preset initial neural network model also includes a first processing network, an automatic path selection network and a second processing network. For the network structure and principle of the initial neural network model, please refer to the above-mentioned description of the neural network model, which will not be repeated in this embodiment.

[0083] S703, obtaining the value of the loss function of the initial neural network model according to the processed sample medical image and the gold standard image.

[0084] Specifically, the computer device obtains the value of the loss function of the initial neural network model based on the processed sample medical image and the gold standard image. Optionally, the computer device can obtain the value of the loss function of the initial neural network model based on the pixel value of each pixel of the processed sample medical image and the pixel value of each pixel of the gold standard image. Optionally, the loss function of the initial neural network model can be a Dice loss function or a Focal loss function.

[0085] S704, adjusting the parameters of the initial neural network model according to the value of the loss function of the initial neural network model to obtain a neural network model.

[0086] Specifically, the computer device adjusts the parameters of the initial neural network model according to the value of the loss function of the initial neural network model obtained above to obtain the above neural network model. Optionally, the computer device may adjust the parameters of the initial neural network model according to the value of the loss function of the initial neural network model until the value of the loss function of the initial neural network model reaches a stable value or a minimum value, and the initial neural network model when the value of the loss function of the initial neural network model reaches a stable value or a minimum value is determined as the above neural network model.

[0087] In this embodiment, the computer device obtains a large number of sample medical images and gold standard images corresponding to the sample medical images, and inputs the sample medical images into a preset initial neural network model. Through the initial neural network model, a processed sample medical image can be obtained, so that the value of the loss function of the initial neural network model can be obtained according to the processed sample medical image and the gold standard image corresponding to the sample medical image, and then the parameters of the initial neural network model can be adjusted according to the value of the loss function of the initial neural network model. Since the computer device trains the initial neural network model through a large number of sample medical images, the number of sample medical images is large, so the initial neural network model can be trained more accurately, thereby improving the accuracy of the obtained neural network model.

[0088] It should be understood that although Figure 2-7 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2-7 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0089] In one embodiment, Figure 8 As shown, an image quality improvement device is provided, comprising: a first acquisition module and a first processing module, wherein:

[0090] The first acquisition module is used to acquire the medical image to be processed.

[0091] The first processing module is used to input the medical image into a preset neural network model, and process the medical image through the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0092] The image quality improvement device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0093] On the basis of the above embodiment, optionally, the neural network model includes a first processing network, an automatic path selection network, and a second processing network; the first processing module includes: a first feature extraction unit, a second feature extraction unit, and a processing unit, wherein:

[0094] The first feature extraction unit is used to input the medical image into the first processing network, perform first feature extraction on the medical image through the first processing network, and obtain the first feature of the medical image.

[0095] The second feature extraction unit is used to input the first feature of the medical image into the automatic path selection network, and perform second feature extraction on the first feature of the medical image through the automatic path selection network to obtain the second feature of the medical image.

[0096] The processing unit is used to input the second feature of the medical image into the second processing network, and perform integration processing and convolution processing on the second feature through the second processing network to obtain a processed medical image.

[0097] Optionally, the automatic path selection network includes a basic sub-network and a leaf sub-network; the basic sub-network is used to perform one or more path processing on the first feature of the medical image; the leaf sub-network is used to perform feature fusion on the output of the basic sub-network to obtain the second feature of the medical image; wherein the number of leaf sub-networks corresponds to the number of the last layer of the basic sub-network.

[0098] The image quality improvement device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0099] On the basis of the above embodiment, optionally, the above device further includes: a second acquisition module, a second processing module, a third acquisition module and a first training module, wherein:

[0100] The second acquisition module is used to acquire sample medical images.

[0101] The second processing module is used to input the sample medical image into a preset initial neural network model, and obtain a processed sample medical image through the initial neural network model; the resolution of the processed sample medical image is greater than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image is greater than the signal-to-noise ratio of the sample medical image.

[0102] The third acquisition module is used to obtain the value of the loss function of the initial neural network model according to the sample medical image and the processed sample medical image.

[0103] The first training module is used to adjust the parameters of the initial neural network model according to the value of the loss function of the initial neural network model to obtain the neural network model.

[0104] The image quality improvement device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0105] On the basis of the above embodiment, optionally, the above device further includes: a fourth acquisition module, a third processing module, a fifth acquisition module, and a second training module, wherein:

[0106] The fourth acquisition module is used to acquire the sample medical image and the gold standard image corresponding to the sample medical image; the resolution of the gold standard image is greater than the resolution of the sample medical image; the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

[0107] The third processing module is used to input the sample medical image into a preset initial neural network model, and obtain a processed sample medical image through the initial neural network model; the resolution of the processed sample medical image is greater than the resolution of the sample medical image; the signal-to-noise ratio of the processed sample medical image is greater than the signal-to-noise ratio of the sample medical image.

[0108] The fifth acquisition module is used to obtain the value of the loss function of the initial neural network model according to the processed sample image and the gold standard image.

[0109] The second training module is used to adjust the parameters of the initial neural network model according to the value of the loss function of the initial neural network model to obtain a neural network model.

[0110] The image quality improvement device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0111] On the basis of the above embodiment, optionally, the fourth acquisition module includes: a first acquisition unit and a second acquisition unit, wherein:

[0112] The first acquisition unit is used to control the first scanner to scan the subject once, and determine the obtained scanned image as the sample medical image.

[0113] The second acquisition unit is used to control the first scanner to scan the subject multiple times, and average the images obtained by the multiple scans to obtain a gold standard image corresponding to the sample medical image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

[0114] The image quality improvement device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0115] On the basis of the above embodiment, optionally, the fourth acquisition module includes: a third acquisition unit and a fourth acquisition unit, wherein:

[0116] A third acquisition unit is used to control the first scanner to scan the subject, and determine the obtained scanned image as the sample medical image;

[0117] The fourth acquisition unit is used to control the second scanner to scan the subject and determine the obtained scanned image as the gold standard image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

[0118] The image quality improvement device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0119] For the specific definition of the image quality improvement device, please refer to the definition of the image quality improvement method above, which will not be repeated here. Each module in the above-mentioned image quality improvement device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0120] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0121] Acquiring a medical image to be processed;

[0122] The medical image is input into a preset neural network model, and the medical image is processed by the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0123] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0124] Acquiring a medical image to be processed;

[0125] The medical image is input into a preset neural network model, and the medical image is processed by the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image.

[0126] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0127] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0128] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for improving image quality, characterized in that: The method comprises: Acquiring a medical image to be processed; Inputting the medical image into a preset neural network model, processing the medical image through the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image; The neural network model includes a first processing network, an automatic path selection network, and a second processing network; the medical image is input into a preset neural network model, and the medical image is processed by the neural network model to obtain a processed medical image, including: Inputting the medical image into the first processing network, and performing first feature extraction on the medical image through the first processing network to obtain a first feature of the medical image; Inputting the first feature of the medical image into the automatic path selection network, and extracting the second feature of the first feature of the medical image through the automatic path selection network to obtain the second feature of the medical image; Inputting the second feature of the medical image into the second processing network, and performing integration processing and convolution processing on the second feature through the second processing network to obtain the processed medical image; The automatic path selection network includes a basic subnetwork and a leaf subnetwork; the basic subnetwork is used to perform one or more path processing on the first feature of the medical image; the leaf subnetwork is used to perform feature fusion on the output of the basic subnetwork to obtain the second feature of the medical image; The one or more path processings on the first feature of the medical image include: generating multiple weight values ​​through the gating module of the basic sub-network, and assigning different weights to the outputs of each basic sub-network to perform one or more path processings on the first feature of the medical image.

2. The method according to claim 1, characterized in that: The number of the leaf sub-networks corresponds to the number of the last layers of the basic sub-network.

3. The method according to claim 1, characterized in that The training method of the neural network model includes: Obtaining sample medical images; Inputting the sample medical image into a preset initial neural network model, and obtaining a processed sample medical image through the initial neural network model; Obtaining a value of a loss function of the initial neural network model according to the sample medical image and the processed sample medical image; According to the value of the loss function of the initial neural network model, the parameters of the initial neural network model are adjusted to obtain the neural network model.

4. The method according to claim 1, characterized in that The training method of the neural network model includes: Acquire a sample medical image and a gold standard image corresponding to the sample medical image; the resolution of the gold standard image is greater than the resolution of the sample medical image; the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image; Inputting the sample medical image into a preset initial neural network model, and obtaining a processed sample medical image through the initial neural network model; Obtaining a value of a loss function of the initial neural network model according to the processed sample image and the gold standard image; According to the value of the loss function of the initial neural network model, the parameters of the initial neural network model are adjusted to obtain the neural network model.

5. The method according to claim 4, characterized in that The acquiring of a sample medical image and a gold standard image corresponding to the sample medical image comprises: Controlling the first scanner to scan the subject once, and determining the obtained scanned image as the sample medical image; Control the first scanner to scan the subject multiple times, average the images obtained by the multiple scans, and obtain a gold standard image corresponding to the sample medical image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

6. The method according to claim 4, characterized in that The obtaining of a sample medical image and a gold standard image corresponding to the sample medical image comprises: Controlling the first scanner to scan the subject, and determining the obtained scanned image as the sample medical image; The second scanner is controlled to scan the subject, and the obtained scanned image is determined as the gold standard image; wherein the resolution of the gold standard image is greater than the resolution of the sample medical image; and the signal-to-noise ratio of the gold standard image is greater than the signal-to-noise ratio of the sample medical image.

7. An image quality improvement device, characterized in that: The device comprises: A first acquisition module, used for acquiring a medical image to be processed; A first processing module, configured to input the medical image into a preset neural network model, and process the medical image through the neural network model to obtain a processed medical image; the resolution of the processed medical image is greater than the resolution of the medical image; the signal-to-noise ratio of the processed medical image is greater than the signal-to-noise ratio of the medical image; The neural network model includes a first processing network, an automatic path selection network, and a second processing network; the first processing module includes: A first feature extraction unit, configured to input the medical image into the first processing network, and perform first feature extraction on the medical image through the first processing network to obtain a first feature of the medical image; A second feature extraction unit, configured to input the first feature of the medical image into the automatic path selection network, and perform second feature extraction on the first feature of the medical image through the automatic path selection network to obtain a second feature of the medical image; a processing unit, configured to input the second feature of the medical image into the second processing network, and perform integration processing and convolution processing on the second feature through the second processing network to obtain the processed medical image; The automatic path selection network includes a basic sub-network and a leaf sub-network; the basic sub-network is used to perform one or more path processing on the first feature of the medical image; the leaf sub-network is used to perform feature fusion on the output of the basic sub-network to obtain the second feature of the medical image; The one or more path processings on the first feature of the medical image include: generating multiple weight values ​​through the gating module of the basic sub-network, and assigning different weights to the outputs of each basic sub-network to perform one or more path processings on the first feature of the medical image.

8. The device according to claim 7, characterized in that The number of the leaf sub-networks corresponds to the number of the last layers of the basic sub-network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.