DR image quality inspection method, device and computer readable storage medium

Abstract

The present application discloses a DR image quality inspection method, device and computer-readable storage medium. It relates to the field of image processing technology. The method includes: dividing the DR image into multiple image blocks, and determining the absolute positions of the multiple image blocks; determining the relative position coding between the multiple image blocks in different regions according to the absolute positions of the multiple image blocks; generating corresponding first image feature information according to the vector information of the multiple image blocks and the relative position coding between the multiple image blocks through a quality evaluation model; decoding the first image feature information through a quality evaluation model, and outputting the second image feature information corresponding to the DR image; and detecting the DR image according to the second image feature information through a quality evaluation model to determine the quality score of the DR image. This solves the technical problem that the poor effect of the attention mechanism leads to inaccurate scoring of the DR image during image quality evaluation.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a DR image quality inspection method, device and computer-readable storage medium. Background Art

[0002] Digital radiography (DR) is one of the common medical imaging methods, and its image quality directly affects the accuracy of disease diagnosis. However, in practical applications, DR images may have quality problems such as noise, blur, overexposure or underexposure due to acquisition conditions, patient factors or equipment problems, resulting in unreliable diagnostic results. Traditional quality control methods mainly rely on human experience or rule-based image quality evaluation algorithms, which have problems such as strong subjectivity, low efficiency and poor adaptability.

[0003] In recent years, large models have achieved many excellent results in the field of natural language processing due to their pre-training capabilities on large-scale data and strong generalization. At the same time, they have also shown great potential in the field of vision, giving rise to the development of large visual models. Large visual models, with their huge number of parameters, can more accurately capture subtle features in medical images, and have stronger generalization capabilities and stability than traditional convolutional neural networks (CNNs). They have broad application prospects in the field of medical image processing.

[0004] Nevertheless, existing large-scale visual models still face some challenges when processing DR images. One important issue is the poor effect of the attention mechanism. The attention mechanism aims to highlight the key areas in the image in a weighted manner, but in practical applications, it may not be able to effectively focus on small objects or detailed features in the DR image, resulting in inaccurate scores for the DR image when evaluating image quality.

[0005] The publication number is CN118115786A, and the name is a classification method for diabetic retinopathy images with blur and abnormal brightness. The method is based on The pre-trained model is used as the backbone network, which is formed by adding channel and spatial attention modules to the backbone network. model; at the same time, the cross entropy loss function is improved according to the characteristics of DR images, and the image enhancement algorithm is used to improve the quality of the collected DR images. The method comprises the following steps: step S1, using EyePACS and APTOS2019 datasets to construct a DR image dataset; step S2, applying an image enhancement algorithm to the DR images in the dataset to correct the brightness of the DR images, improve the contrast and highlight the characteristics of the lesions; step S3, using The model is based on the attention mechanism to build a Model; Step S4, training using enhanced DR images Model.

[0006] The publication number is CN115089203A, and the name is DR imaging analysis method and DR imaging device. The method includes: obtaining a digital X-ray photography DR image of a target object through a DR imaging device, the DR image including at least one of a DR original image and an image after the DR original image is processed; extracting image features from the DR image; obtaining device parameter features according to device parameters of the DR imaging device, and / or obtaining object parameter features according to object parameters of the target object; determining a DR feature index according to the image features, and the device parameter features and / or the object parameter features, the DR feature index being used to characterize the comprehensive level between the radiation dose of X-rays received by the target object and the image quality of the DR image.

[0007] Currently, no effective solution has been proposed to the technical problem that the attention mechanism in the above-mentioned prior art is not effective, resulting in inaccurate scoring of the DR image during image quality evaluation. Summary of the invention

[0008] The embodiments of the present application provide a DR image quality inspection method, device and computer-readable storage medium to at least solve the technical problems in the prior art that the attention mechanism is not effective, resulting in inaccurate DR image segmentation and low DR image quality score.

[0009] According to one aspect of an embodiment of the present application, a quality inspection method for a DR image is provided, comprising: dividing a DR image into multiple image blocks, and determining the absolute positions of the multiple image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block; determining the relative position codes between the multiple image blocks in different regions according to the absolute positions of the multiple image blocks; generating corresponding first image feature information according to the vector information of the multiple image blocks and the relative position codes between the multiple image blocks through a quality evaluation model; decoding the first image feature information through the quality evaluation model, and outputting second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate multiple position information of the DR image; and detecting the DR image according to the second image feature information through the quality evaluation model to determine the quality score of the DR image.

[0010] According to another aspect of an embodiment of the present application, a quality inspection device for a DR image is also provided, including: a position determination module, used to divide the DR image into multiple image blocks, and determine the absolute positions of the multiple image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block; a coding determination module, used to determine the relative position coding between multiple image blocks in different regions according to the absolute positions of the multiple image blocks; an information generation module, used to generate corresponding first image feature information according to the vector information of the multiple image blocks and the relative position coding between the multiple image blocks through a quality evaluation model; a result output module, used to decode the first image feature information through a quality evaluation model, and output second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate multiple position information of the DR image; and a score determination module, used to detect the DR image according to the second image feature information through a quality evaluation model, and determine the quality score of the DR image.

[0011] According to another aspect of an embodiment of the present application, a DR image quality inspection device is also provided, including: a processor; and a memory, connected to the processor, for providing the processor with instructions for processing the following processing steps: dividing the DR image into multiple image blocks, and determining the absolute positions of the multiple image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block; determining the relative position codes between the multiple image blocks in different regions according to the absolute positions of the multiple image blocks; generating corresponding first image feature information according to the vector information of the multiple image blocks and the relative position codes between the multiple image blocks through a quality evaluation model; decoding the first image feature information through the quality evaluation model, and outputting second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate multiple position information of the DR image; and detecting the DR image according to the second image feature information through the quality evaluation model to determine the quality score of the DR image.

[0012] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0013] In an embodiment of the present application, a DR image is segmented into multiple image blocks and their absolute positions are determined, so as to finely capture the local features of the image, so that small targets or detail features in the DR image can be focused on. Afterwards, the present technical solution uses a quality evaluation model to encode according to relative positions, so as to understand the spatial relationship between image blocks, thereby improving the focus on key areas and segmentation accuracy, enhancing the quality evaluation model's ability to understand image features, and enabling the quality evaluation model to more accurately extract feature information related to disease diagnosis. In addition, the present technical solution detects and scores DR images through a quality evaluation model, provides an objective quality evaluation standard, and improves the image quality score. This solves the technical problem that the attention mechanism existing in the prior art is not effective, resulting in inaccurate scoring of the DR image during image quality evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0015] Figure 1 is a hardware structure block diagram of a computing device for implementing the method according to Embodiment 1 of the present application;

[0016] Figure 2 is a flow chart of a DR image quality inspection method according to the first aspect of Embodiment 1 of the present application;

[0017] Figure 3 It is a module schematic diagram of the DR image quality inspection method according to the first aspect of Example 1 of the present application;

[0018] Figure 4 is a schematic diagram of image blocks of a part of a DR image according to Embodiment 1 of the present application;

[0019] Figure 5A is a first schematic diagram of a DR image according to Example 1 of the present application;

[0020] Figure 5B is a second schematic diagram of a DR image according to Example 1 of the present application;

[0021] Figure 5C is a third schematic diagram of a DR image according to Example 1 of the present application;

[0022] Figure 6 is a sequential flow chart of the DR image quality inspection method according to the first aspect of Embodiment 1 of the present application;

[0023] Figure 7is a schematic diagram of a DR image quality inspection device according to Embodiment 2 of the present application; and

[0024] Figure 8 It is a schematic diagram of the DR image quality inspection device according to Example 3 of the present application. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only embodiments of a part of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] Example 1

[0028] According to this embodiment, a method embodiment of a DR image quality inspection method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0029] The method embodiment provided in this embodiment can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computing device for implementing a DR image quality inspection method. Figure 1As shown, the computing device may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory for storing data, and a transmission device for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0030] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computing device. As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0031] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the quality inspection method of DR images in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the quality inspection method of DR images of the above-mentioned application is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computing device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0032] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computing device. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0033] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of the computing device.

[0034] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing devices described above.

[0035] Under the above operating environment, according to the first aspect of this embodiment, a DR image quality inspection method is provided. The method comprises: Figure 1 The computing device implementation shown in . Figure 2 A schematic diagram showing the process of the method is shown in FIG. Figure 2 As shown, the method includes:

[0036] S202: Segment the DR image into a plurality of image blocks, and determine the absolute positions of the plurality of image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block;

[0037] S204: determining relative position codes between the multiple image blocks in different regions according to the absolute positions of the multiple image blocks;

[0038] S206: Generate corresponding first image feature information according to the vector information of the plurality of image blocks and the relative position encoding between the plurality of image blocks through the quality evaluation model;

[0039] S208: decoding the first image feature information through the quality assessment model, and outputting second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate a variety of position information of the DR image; and

[0040] S210: Detecting the DR image according to the second image feature information using a quality evaluation model to determine a quality score of the DR image.

[0041] Specifically, refer to Figure 3 As shown, the computing device acquires a DR image, such as a DR chest radiograph. The computing device then inputs the DR image into a preprocessing module, which divides the DR image into a plurality of image blocks, such as m image blocks, where image block i is identified as a. i , i=1~m. Then the computing device generates the absolute position of the image block according to the absolute row and column position or index of the image block i through the preprocessing module. Figure 4 The absolute position of each image block in a partial DR image is shown. In this embodiment, the absolute row and column positions of image blocks are used as an example to determine the absolute position c of each image block. i . refer to Figure 4 As shown, image block a1 is the 0th row and 0th column in the partial DR image, then the absolute position c1 of image block a1 is (0,0), image block a2 is the 0th row and 1st column in the partial DR image, then the absolute position c2 of image block a2 is (0,1), and so on, the absolute position c3 of image block a3 is (0,2), the absolute position c4 of image block a4 is (1,0), the absolute position c5 of image block a5 is (1,1), the absolute position c6 of image block a6 is (1,2), the absolute position c7 of image block a7 is (2,0), the absolute position c8 of image block a8 is (2,1), and the absolute position c9 of image block a9 is (2,2).

[0042] Furthermore, the computing device divides the DR image into a plurality of different regions with symmetry as the goal, and the computing device determines the absolute position of each image block in the different regions according to the absolute position of each image block in the different regions through the preprocessing module. i and a j The relative position encoding x i,j , where i=1~m, j=1~m. For example, the relative position encoding x between image blocks a1 and a1 1,1 ; The relative position code x between image blocks a1 and a2 1,2 ; ...; Image block a m and a m The relative position encoding x m,m The relative position coding is a position coding generated according to the relative position between the image blocks. For example, the computing device generates a relative position coding according to the relative position between the image blocks through the preprocessing module. i and a j The absolute position of image block a i and a j The relative position between them is then used to generate the corresponding relative position code x i,j .

[0043] Furthermore, the computing device is pre-set with a quality assessment model, wherein the quality assessment model may be, for example, a Qwen large model. And wherein the quality assessment model includes an encoding module, a decoding module and a quality assessment module. The computing device generates each image block a by using a linear transformation through a preprocessing module i Vector information b i . The computing device then converts the vector information b i and the relative position code x i,j The encoding module is input into the quality assessment model, and the encoding module is used to calculate the quality of the image blocks according to the multiple image blocks a.i Vector information b i And the relative position encoding x between multiple image blocks i,j , generating corresponding image feature information (ie, first image feature information).

[0044] Furthermore, the computing device inputs the first image feature information into the decoding module of the quality assessment model, and the decoding module decodes the image feature information, thereby outputting the second image feature information corresponding to the DR image. For example, the DR image is a DR chest radiograph of the chest in the frontal position, and the output second image feature information corresponding to the DR chest radiograph of the chest in the frontal position is as follows:

[0045]

[0046]

[0047] The parameters left and right refer to the left and right contours. For example, the lung area has left and right lungs. In medical imaging, the left is on the right and the right is on the left, so the x value of the right coordinate is smaller than the x value of the left coordinate. The xs in left and right stores the x coordinate of each point, and the ys stores the y coordinate of each point, one-to-one correspondence. If it is empty, it means there is no left contour or right contour (that is, only half of it is captured); in the external frame coordinates: x, y are the coordinates of the upper left corner, cx, cy are the width and height respectively.

[0048] In addition to the DR chest radiograph, the quality assessment model in this embodiment can also perform encoding and decoding processing on the DR images of the knee joint corresponding to the anteroposterior and lateral views of the knee joint, and the DR images of the pelvis corresponding to the anteroposterior view of the pelvis. For example, the output second image feature information corresponding to the DR image of the anteroposterior view of the knee joint is as follows:

[0049]

[0050]

[0051] For another example, the second image feature information corresponding to the DR image of the lateral position of the knee joint output is as follows:

[0052]

[0053]

[0054] For another example, the second image feature information corresponding to the DR image of the pelvis in the positive position is as follows:

[0055]

[0056] Furthermore, the computing device inputs the second image feature information output by the decoding module into the quality evaluation module of the quality evaluation model, and the quality evaluation module detects and scores the DR image according to the second image feature information, thereby outputting the quality score of the DR image to evaluate the quality of the DR image, wherein the quality of the DR image can be evaluated in terms of noise, blur, overexposure or underexposure.

[0057] As mentioned in the background, existing large visual models still face some challenges when processing DR images. One of the important issues is the poor effect of the attention mechanism. The attention mechanism aims to highlight the key areas in the image in a weighted manner, but in practical applications, it may not be able to effectively focus on small objects or detailed features in DR images, resulting in inaccurate image segmentation.

[0058] In view of the technical problems described above, through the technical solution of the embodiment of the present application, the DR image is segmented into multiple image blocks and their absolute positions are determined, so as to finely capture the local features of the image, so that small targets or detail features in the DR image can be focused on. After that, the technical solution uses the quality evaluation model to encode according to the relative position, so that the spatial relationship between the image blocks can be understood, thereby improving the attention to key areas and the segmentation accuracy, enhancing the quality evaluation model's ability to understand image features, and enabling the quality evaluation model to more accurately extract feature information related to disease diagnosis. In addition, the technical solution detects and scores DR images through a quality evaluation model, provides an objective quality evaluation standard, and improves the quality score of the image. This solves the technical problem that the attention mechanism existing in the prior art is not effective, resulting in inaccurate scoring of the DR image during image quality evaluation.

[0059] Optionally, the operation of determining the relative position coding between multiple image blocks in different areas according to the absolute positions of multiple image blocks includes: identifying the DR image to determine the symmetrical area and the asymmetrical area in the DR image; calculating the relative position coding between the first image blocks in the symmetrical area according to the absolute position using the center line in the symmetrical area; and calculating the relative position coding between the second image block and the other image blocks in the asymmetrical area according to the distance between the second image block and the other image blocks based on the absolute position.

[0060] Specifically, the computing device identifies the DR image through a preprocessing module to determine the symmetrical area and the asymmetrical area in the DR image. Figure 5A A schematic diagram of a DR image is shown. Figure 5AAs shown, the computing device recognizes the DR image through the preprocessing module, wherein the DR image is a DR chest radiograph. In the DR image, the symmetry above the ribs is stronger, and the part above the ribs is taken as the symmetric area Z1; the symmetry below the ribs is weaker, and the part below the ribs is taken as the asymmetric area Z2.

[0061] Further, the computing device determines the midline S of the symmetric region Z1 according to the spine in the DR image through the preprocessing module. Then, the computing device calculates the lateral relative position and the longitudinal relative position between the image blocks (i.e., the first image blocks) in the symmetric region through the preprocessing module, and generates corresponding relative position codes according to the lateral relative position and the longitudinal relative position.

[0062] More specifically, refer to Figure 5B As shown, for example, image block a x and a y are all image blocks in the symmetric region Z1, the computing device determines the image blocks a and b through the preprocessing module. x and a y The absolute position c x and c y . Then the computing device uses the preprocessing module according to the absolute position c x The number of columns of image block a is calculated x Distance to the center line Scol x , and according to the absolute position c y The number of columns of image block a is calculated y Distance to the center line Scol y The distance is in units of image blocks. The computing device then calculates the distance col through the preprocessing module x and col y The difference between them is taken as image block a x and a y Relative horizontal position :

[0063] .

[0064] The computing device then uses the preprocessing module to calculate the absolute position c x The number of rows of raw x and c y The number of rows of raw y The difference between them is taken as image block a x and a y Relative vertical position :

[0065] .

[0066] Furthermore, the computing device is pre-set with weight values ​​d1 and d2 for parameter adjustment, which can be set according to empirical values. The weight value d1 corresponds to the relative position in the longitudinal direction. , the weight value d2 corresponds to the relative position in the horizontal direction . Then calculate the device's relative vertical position and weight value d1 and horizontal relative position And weight value d2 calculate image block a x and a y The relative position encoding x x,y :

[0067] .

[0068] Thus, the computing device calculates the image block a through the preprocessing module according to the above x and a y The relative position encoding x x,y The relative position encoding between all image blocks in the symmetric area is calculated in this way, which will not be repeated here.

[0069] Furthermore, the computing device calculates the relative position coding between the image block in the asymmetric region (ie, the second image block) and the image blocks in any other region (ie, the entire DR image) through the preprocessing module.

[0070] refer to Figure 5C As shown, for example, image block a u is an image block in the symmetric region, image block a t is an image block in an asymmetric region, i.e., two image blocks (image block a and b) used to calculate the relative position encoding. u and a t ) has at least one image block in an asymmetric area.

[0071] The computing device then obtains the image block a through the preprocessing module u and a t The absolute position of image block a u The absolute position c u =(raw u ,col u ), image block a t The absolute position c t =(raw t ,col t ). The computing device then calculates the absolute position c through the preprocessing module u The number of rows of raw u With absolute position c t The number of rows of raw t The difference between u and at Relative vertical position :

[0072] .

[0073] The computing device then calculates the absolute position c through the preprocessing module u Number of columns col u With absolute position c t Number of columns col t The difference between u and a t Relative horizontal position :

[0074] .

[0075] Furthermore, the computing device is pre-set with weight values ​​d1 and d2 for parameter adjustment, which can be set according to empirical values. The weight value d1 corresponds to the relative position in the longitudinal direction. , the weight value d2 corresponds to the relative position in the horizontal direction . Then calculate the device's relative vertical position and weight value d1 and horizontal relative position And weight value d2 calculate image block a u and a t The relative position encoding x u,t :

[0076] .

[0077] Therefore, the technical solution can make full use of the symmetry of medical images and perform relative position encoding by using the midline in the symmetrical area to determine the horizontal and vertical relative positions between image blocks, and adjusting the parameters in combination with the weight value, so as to more accurately reflect the changes in image content caused by the deviation of the positional relationship between different image blocks, which is conducive to accurate image analysis.

[0078] Optionally, the operation of generating corresponding first image feature information according to vector information of multiple image blocks and relative position encoding between the multiple image blocks through a quality assessment model includes: generating corresponding first query vectors, first key vectors and first value vectors from the vector information of multiple image blocks; generating bias items according to the relative position encoding between the multiple image blocks; and generating first image feature information according to the first query vector, first key vector, first value vector and bias item through the quality assessment model.

[0079] Specifically, the computing device processes the image block a through a preprocessing module. i Vector information b iProcessing is performed to generate the corresponding first query vector , first key vector and the first value vector .

[0080] The computing device then uses a preset mapping function through a preprocessing module According to the image block a i and a j The relative position encoding x i,j , generate the corresponding bias term P i,j :

[0081] .

[0082] Further, the computing device uses the encoding module of the quality evaluation model according to the first query vector , first key vector and the first value vector V i And the bias term P i,j , generating image feature information (ie, first image feature information).

[0083] Therefore, the technical solution generates bias items according to relative position encoding through the attention mechanism, thereby not only considering the characteristics of the image blocks themselves, but also fully considering the relative positions between them, thereby improving the model's attention to key areas and segmentation accuracy.

[0084] Optionally, the operation of generating a corresponding first query vector, a first key vector and a first value vector from the vector information of multiple image blocks includes: generating a second query vector, a second key vector and a first value vector based on the vector information of the multiple image blocks; and generating a corresponding first query vector and a first key vector based on the second query vector and the second key vector using a low-rank matrix.

[0085] Specifically, the computing device processes the image block a through a preprocessing module. i Vector information b i Perform linear transformation to generate the corresponding second query vector Q i , the second key vector K i and the first value vector V i The computing device is pre-set with a weight matrix B for linear transformation Q , B K , B V . The weight matrix B Q Used to generate query vector, weight matrix B K Used to generate key vector, weight matrix B V Used to generate a vector of values.

[0086] Thus, the computing device uses the preprocessing module according to the vector information bi and the weight matrix B Q , generate the second query vector Q i :

[0087] Q i =b i B Q .

[0088] The computing device uses the preprocessing module to calculate the vector information b i and the weight matrix B K , generate the second key vector K i :

[0089] K i =b i B K .

[0090] The computing device uses the preprocessing module to calculate the vector information b i and the weight matrix B V , generate the first value vector V i :

[0091] V i =b i B V .

[0092] Furthermore, the computing device uses a preset low-rank matrix through a preprocessing module to respectively process the second query vector Q i and the second key vector K i Processing is performed to generate the corresponding first query vector and the first key vector :

[0093] ;

[0094] .

[0095] Where W Q and W K It is a pre-trained low-rank matrix with rank r, that is, the low-rank matrix W Q and W K The dimension of is r.

[0096] Therefore, the technical solution processes the second query vector and the second key vector through a preset low-rank matrix to generate the first query vector and the first key vector, which not only reduces the demand for computing resources, but also retains key information and improves the computing efficiency of the model.

[0097] Optionally, the operation of generating the first image feature information according to the first query vector, the first key vector, the first value vector and the bias term through the quality assessment model includes: generating the first image feature information according to the following formula:

[0098] ,

[0099] Among them A i represents the first image feature information corresponding to the i-th image block, represents the first query vector corresponding to the i-th image block, represents the first key vector corresponding to the jth image block, V j represents the first value vector corresponding to the jth image block, P i,j represents the bias term corresponding to the i-th image block and the j-th image block, m represents the number of image blocks, and r represents the dimension of the first key vector.

[0100] Specifically, the computing device converts the first query vector , first key vector and the first value vector V i And the bias term P i,j The first query vector is input into the encoding module of the quality evaluation model. , first key vector and the first value vector V j And the bias term P i,j Calculate and generate the first image feature information [WA]=[A1,A2,...,A m ]. Where A i Represents image block a i The image feature information A i :

[0101] .

[0102] Therefore, the present technical solution utilizes the first query vector, the first key vector, the first value vector and the corresponding bias items to generate the first image feature information, so as to more accurately express the correlation between the image blocks.

[0103] Optionally, the operation of detecting the DR image according to the second image feature information through a quality assessment model to determine the quality score of the DR image includes: detecting the DR image according to the second image feature information through a quality assessment model, and outputting a score corresponding to the scoring parameter as the quality score of the DR image.

[0104] Specifically, the computing device inputs the second image feature information into the quality evaluation module of the quality evaluation model, processes the second image feature information through the quality evaluation module and outputs the scores of various scoring parameters of the DR image, thereby serving as the quality score of the DR image. The scoring parameters of the DR image, such as the DR chest radiograph of the chest in the frontal position, and the method of calculating the scores of various scoring parameters may include, for example:

[0105] 1. Midline deviation: The score is calculated based on the difference in distance between the bilateral ribs and the left and right borders of the image;

[0106] 2. Inappropriate FOV: Distinguish between FOV that is too large and too small, and calculate the distance between the upper, lower, left, and right sides of the inspection part and the upper, lower, left, and right boundaries of the image;

[0107] 3. Inappropriate space reserved above the clavicle: Distinguish between too much and too little. Clinically, 2-5 cm is appropriate. The score is calculated based on the distance from the upper end of the clavicle to the upper border of the image.

[0108] 4. Clavicle tilt: The difference in distance between the upper ends of the clavicles on both sides is used to calculate the score;

[0109] 5. Foreign matter detection: Foreign matter is not allowed to appear, otherwise points will be deducted;

[0110] 6. Scapula not opened: The intersection of the scapula and the lung area accounts for the proportion of the scapula area. If it is too large, it means that the scapula is not opened. The score is calculated based on the proportion.

[0111] 7. Insufficient inspiration: The ratio of the intersection of the eighth rib and the lung area to the area of ​​the eighth rib. A small amount of intersection is considered sufficient inspiration, and the score is calculated based on the ratio;

[0112] 8. Too high / too low dose: calculate the average grayscale;

[0113] 9. Heart inversion: The distance between the center of the heart and the two sides of the ribs, if it is to the right, it is right heart position, which is a reverse position;

[0114] 10. Left and right label errors: Check whether the left and right labels are in the correct position;

[0115] 11. The apex of the lung is not fully exposed:

[0116] (1) With a lead collar: whether the top of the lung apex intersects with the lead collar;

[0117] (2) Lead-free neck collar: Check whether the top of the lung apex intersects with the upper edge of the image.

[0118] Thus, the computing device calculates the score of each scoring parameter according to the way of calculating the score of each scoring parameter.

[0119] In addition, in addition to the evaluation parameters corresponding to the DR images of the chest, it also includes evaluation parameters and scoring calculation methods corresponding to DR images of the knee in the AP and lateral positions, and also includes evaluation parameters and scoring calculation methods corresponding to DR images of the pelvis in the AP position.

[0120] The evaluation parameters and scoring calculation methods of DR images of the knee joint in the anteroposterior position are as follows:

[0121] 1. Inappropriate FOV: Based on the proportion of each bone in the image;

[0122] 2. The femoral-shin line is not parallel to the longitudinal axis of the image: calculate the angle between the femoral-shin line and the y-axis;

[0123] 3. The intercondylar notch is not centered: calculate the distance between the intercondylar notch and the center of the lower end of the femur;

[0124] 4. The patella is not centered: calculate the distance between the horizontal coordinate of the patellar center of gravity and the center of the lower end of the femur;

[0125] 5. Asymmetry of left and right gaps: Calculate the difference between the left and right gaps between the femur and tibia;

[0126] 6. Excessive / inadequate overlap of the tibia and fibula: Calculate the ratio of the overlap area of ​​the tibia and fibula to the circular area of ​​the upper end of the fibula;

[0127] 7. Left-right asymmetry: Calculate the distance difference between the overall contour and the two sides of the image (for both legs, there is logic for judging single leg and double legs);

[0128] 8. Left and right label errors: Check whether the left and right labels are in the correct position;

[0129] 9. Foreign matter detection: Foreign matter is not allowed, and points will be deducted if it occurs.

[0130] In addition, the evaluation parameters and scoring calculation methods of DR images of the lateral knee joint are as follows:

[0131] 1. The degree of femoral-tibial overlap is inappropriate: calculate the proportion of the femoral-tibial overlap area to the patella area;

[0132] 2. The degree of femoral-patellar overlap is inappropriate: calculate the proportion of the femoral-patellar overlap area to the patellar mask;

[0133] 3. The degree of tibia-fibular overlap is inappropriate: calculate the ratio of the tibia-fibular overlap area to the fibula mask;

[0134] 4. The angle of the femoral-shin line is inappropriate: calculate the angle between the femoral and shin midlines;

[0135] 5. Left and right label errors: Check whether the left and right labels are in the correct position;

[0136] 6. Foreign matter detection: Foreign matter is not allowed, and points will be deducted if it occurs.

[0137] Evaluation parameters and scoring calculation method of DR images of the pelvis in the frontal position:

[0138] 1. Inappropriate FOV: whether the pelvic contour intersects with the left and right boundaries of the image (x direction);

[0139] 2. The pubic symphysis is not fully exposed: whether the pubic symphysis is exposed (y direction);

[0140] 3. Midline offset: Calculate the distance difference between the pelvic contour and the two sides of the image;

[0141] 4. Left and right label errors: Check whether the left and right labels are in the correct position;

[0142] 5. Foreign matter detection: Foreign matter is not allowed, and points will be deducted if it occurs.

[0143] Therefore, the present technical solution uses preset key indicators as scoring parameters, so that the quality evaluation module can comprehensively evaluate the quality score of the image and improve the accuracy and efficiency of DR image quality evaluation.

[0144] Optionally, the method further comprises: performing a preprocessing operation on the DR image, wherein the preprocessing operation comprises removing noise, processing missing values ​​and standardizing the data format. Thus, the present technical solution performs preprocessing before segmenting the DR image into a plurality of image blocks, thereby improving the accuracy and efficiency of subsequent analysis of the DR image.

[0145] In summary, reference Figure 6 As shown, the specific sequential steps for detecting DR images are as follows:

[0146] S601: Preprocessing the DR image, wherein the preprocessing includes removing noise, processing missing values ​​and standardizing data formats;

[0147] S602: Segment the DR image into a plurality of image blocks, and determine the absolute positions of the plurality of image blocks;

[0148] S603: Identify the DR image and determine the symmetrical area and the asymmetrical area in the DR image;

[0149] S604: Calculate the relative position code between the first image blocks in the symmetric region according to the absolute position using the center line in the symmetric region;

[0150] S605: Calculate the relative position code between the second image block and the other image blocks according to the distance between the second image block and the other image blocks in the asymmetric region based on the absolute position;

[0151] S606: Generate a second query vector, a second key vector and a first value vector according to vector information of the plurality of image blocks;

[0152] S607: Generate a corresponding first query vector and a first key vector according to the second query vector and the second key vector using a low-rank matrix;

[0153] S608: Generate a bias item according to the relative position encoding between the plurality of image blocks;

[0154] S609: Generate first image feature information according to the first query vector, the first key vector, the first value vector and the bias item through the encoding module;

[0155] S610: Decoding the first image feature information through the quality assessment model, and outputting second image feature information corresponding to the DR image;

[0156] S611: Detect the DR image according to the second image feature information through the quality evaluation module, and output a score corresponding to the scoring parameter as the quality score of the DR image.

[0157] In addition, reference Figure 1 As shown, according to the second aspect of this 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 steps of the above method are implemented.

[0158] Therefore, according to this embodiment, the DR image is segmented into multiple image blocks and their absolute positions are determined, so as to finely capture the local features of the image, so that small targets or detail features in the DR image can be focused on. Afterwards, the technical solution uses the quality evaluation model to encode according to the relative position, so as to understand the spatial relationship between the image blocks, thereby improving the attention to key areas and the segmentation accuracy, enhancing the quality evaluation model's ability to understand image features, and enabling the quality evaluation model to more accurately extract feature information related to disease diagnosis. In addition, the technical solution detects and scores DR images through a quality evaluation model, provides an objective quality evaluation standard, and improves the quality score of the image. This solves the technical problem that the attention mechanism existing in the prior art is not effective, resulting in inaccurate scoring of the DR image during image quality evaluation.

[0159] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0160] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0161] Example 2

[0162] Figure 7 FIG. 7 shows a quality inspection device 700 for a DR image according to this embodiment, and the device 700 corresponds to the method according to the first aspect of embodiment 1. Figure 7 As shown, the device 700 includes: a position determination module 710, which is used to divide the DR image into multiple image blocks and determine the absolute positions of the multiple image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block; a coding determination module 720, which is used to determine the relative position coding between multiple image blocks in different regions according to the absolute positions of the multiple image blocks; an information generation module 730, which is used to generate corresponding first image feature information according to the vector information of the multiple image blocks and the relative position coding between the multiple image blocks through a quality evaluation model; a result output module 740, which is used to decode the first image feature information through a quality evaluation model and output second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate a variety of position information of the DR image; and a score determination module 750, which is used to detect the DR image according to the second image feature information through a quality evaluation model and determine the quality score of the DR image.

[0163] Optionally, the coding determination module 720 includes: a first determination submodule, used to identify the DR image and determine the symmetrical area and the asymmetrical area in the DR image; a first calculation submodule, used to calculate the relative position coding between the first image blocks in the symmetrical area according to the absolute position using the midline in the symmetrical area; and a second calculation submodule, used to calculate the relative position coding between the second image block and other image blocks in the asymmetrical area based on the absolute position according to the distance between the second image block and other image blocks.

[0164] Optionally, the information generation module 730 includes: a first generation submodule, used to generate a corresponding first query vector, a first key vector and a first value vector from the vector information of multiple image blocks; a second generation submodule, used to generate a bias item according to the relative position encoding between the multiple image blocks; and a third generation submodule, used to generate first image feature information according to the first query vector, the first key vector, the first value vector and the bias item through a quality assessment model.

[0165] Optionally, the first generation submodule includes: a first generation unit, used to generate a second query vector, a second key vector and a first value vector based on vector information of multiple image blocks; and a second generation unit, used to generate a corresponding first query vector and first key vector based on the second query vector and the second key vector using a low-rank matrix.

[0166] Optionally, the third generating submodule includes: a third generating unit, configured to generate the first image feature information according to the following formula:

[0167] ,

[0168] Among them A i represents the first image feature information corresponding to the i-th image block, represents the first query vector of the i-th image block, represents the first key vector of the jth image block, V j Represents the first value vector of the jth image block, P i,j represents the bias term corresponding to the i-th image block and the j-th image block, m represents the number of image blocks, and r represents the dimension of the first key vector.

[0169] Optionally, the result output module 740 includes: a second determination submodule, configured to detect the DR image according to the second image feature information through a quality assessment model, and output a score corresponding to the scoring parameter as a quality score of the DR image.

[0170] Optionally, the DR image quality inspection device 700 further includes: a processing module, configured to perform a preprocessing operation on the DR image, wherein the preprocessing operation includes removing noise, processing missing values, and standardizing data formats.

[0171] Therefore, according to this embodiment, the DR image is segmented into multiple image blocks and their absolute positions are determined, so as to finely capture the local features of the image, so that small targets or detail features in the DR image can be focused on. Afterwards, the technical solution uses the quality evaluation model to encode according to the relative position, so as to understand the spatial relationship between the image blocks, thereby improving the attention to key areas and the segmentation accuracy, enhancing the quality evaluation model's ability to understand image features, and enabling the quality evaluation model to more accurately extract feature information related to disease diagnosis. In addition, the technical solution detects and scores DR images through a quality evaluation model, provides an objective quality evaluation standard, and improves the quality score of the image. This solves the technical problem that the attention mechanism existing in the prior art is not effective, resulting in inaccurate scoring of the DR image during image quality evaluation.

[0172] Example 3

[0173] Figure 8 FIG. 8 shows a quality inspection device 800 for a DR image according to this embodiment, and the device 800 corresponds to the method according to the first aspect of Embodiment 1. Figure 8 As shown, the device 800 includes: a processor 810; and a memory 820, which is connected to the processor 810 and is used to provide the processor 810 with instructions for processing the following processing steps: dividing the DR image into multiple image blocks, and determining the absolute positions of the multiple image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block; determining the relative position codes between the multiple image blocks in different regions according to the absolute positions of the multiple image blocks; generating corresponding first image feature information according to the vector information of the multiple image blocks and the relative position codes between the multiple image blocks through a quality evaluation model; decoding the first image feature information through the quality evaluation model, and outputting second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate multiple position information of the DR image; and detecting the DR image according to the second image feature information through the quality evaluation model to determine the quality score of the DR image.

[0174] Optionally, the operation of determining the relative position coding between multiple image blocks in different areas according to the absolute positions of multiple image blocks includes: identifying the DR image to determine the symmetrical area and the asymmetrical area in the DR image; calculating the relative position coding between the first image blocks in the symmetrical area according to the absolute position using the center line in the symmetrical area; and calculating the relative position coding between the second image block and the other image blocks in the asymmetrical area according to the distance between the second image block and the other image blocks based on the absolute position.

[0175] Optionally, the operation of generating corresponding first image feature information according to vector information of multiple image blocks and relative position encoding between the multiple image blocks through a quality assessment model includes: generating corresponding first query vectors, first key vectors and first value vectors from the vector information of multiple image blocks; generating bias items according to the relative position encoding between the multiple image blocks; and generating first image feature information according to the first query vector, first key vector, first value vector and bias item through the quality assessment model.

[0176] Optionally, the operation of generating a corresponding first query vector, a first key vector and a first value vector from the vector information of multiple image blocks includes: generating a second query vector, a second key vector and a first value vector based on the vector information of the multiple image blocks; and generating a corresponding first query vector and a first key vector based on the second query vector and the second key vector using a low-rank matrix.

[0177] Optionally, the operation of generating the first image feature information according to the first query vector, the first key vector, the first value vector and the bias term through the quality assessment model includes: generating the first image feature information according to the following formula:

[0178] ,

[0179] Among them A i represents the first image feature information corresponding to the i-th image block, represents the first query vector of the i-th image block, represents the first key vector of the jth image block, V j Represents the first value vector of the jth image block, P i,j represents the bias term corresponding to the i-th image block and the j-th image block, m represents the number of image blocks, and r represents the dimension of the first key vector.

[0180] Optionally, the operation of detecting the DR image according to the second image feature information through a quality assessment model to determine the quality score of the DR image includes: detecting the DR image according to the second image feature information through a quality assessment model, and outputting a score corresponding to the scoring parameter as the quality score of the DR image.

[0181] Optionally, the memory 820 is further configured to provide the processor 810 with instructions for processing the following processing steps: performing a preprocessing operation on the DR image, wherein the preprocessing operation includes removing noise, processing missing values, and standardizing data formats.

[0182] Therefore, according to this embodiment, the DR image is segmented into multiple image blocks and their absolute positions are determined, so as to finely capture the local features of the image, so that small targets or detail features in the DR image can be focused on. Afterwards, the technical solution uses the quality evaluation model to encode according to the relative position, so as to understand the spatial relationship between the image blocks, thereby improving the attention to key areas and the segmentation accuracy, enhancing the quality evaluation model's ability to understand image features, and enabling the quality evaluation model to more accurately extract feature information related to disease diagnosis. In addition, the technical solution detects and scores DR images through a quality evaluation model, provides an objective quality evaluation standard, and improves the quality score of the image. This solves the technical problem that the attention mechanism existing in the prior art is not effective, resulting in inaccurate scoring of the DR image during image quality evaluation.

[0183] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0184] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0185] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0186] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0187] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0188] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0189] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A quality inspection method for digital X-ray DR images, characterized in that: include: Segmenting the DR image into a plurality of image blocks, and determining absolute positions of the plurality of image blocks, wherein the absolute positions are generated based on an absolute row and column position or index of each image block; Determining relative position codes between the plurality of image blocks in different regions according to the absolute positions of the plurality of image blocks; Generate corresponding first image feature information according to the vector information of the plurality of image blocks and the relative position encoding between the plurality of image blocks through a quality assessment model; Decoding the first image feature information by using the quality assessment model, and outputting second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate a variety of position information of the DR image; and The DR image is detected by the quality assessment model according to the second image feature information to determine the quality score of the DR image, wherein The operation of determining the relative position coding between the plurality of image blocks in different regions according to the absolute positions of the plurality of image blocks comprises: Identify the DR image to determine a symmetrical region and an asymmetrical region in the DR image; Calculating the relative position code between the first image blocks in the symmetric region according to the absolute position using the center line in the symmetric region; and The relative position code between the second image block and the other image blocks is calculated based on the absolute position according to the distance between the second image block and the other image blocks in the asymmetric region.

2. The method according to claim 1, characterized in that The operation of generating corresponding first image feature information according to the vector information of the plurality of image blocks and the relative position encoding between the plurality of image blocks through the quality assessment model includes: Generate a corresponding first query vector, a first key vector and a first value vector according to the vector information of the plurality of image blocks; Generate a bias term according to the relative position encoding between the plurality of image blocks; and The first image feature information is generated by the quality assessment model according to the first query vector, the first key vector, the first value vector and the bias item.

3. The method according to claim 2, characterized in that The operation of generating a corresponding first query vector, a first key vector and a first value vector from the vector information of the plurality of image blocks comprises: Generate a second query vector, a second key vector and the first value vector according to the vector information of the plurality of image blocks; and The first query vector and the first key vector are generated according to the second query vector and the second key vector using a low-rank matrix.

4. The method according to claim 2, characterized in that: The operation of generating the first image feature information according to the first query vector, the first key vector, the first value vector, and the bias item through the quality assessment model includes: The first image feature information is generated according to the following formula: ， Among them A i represents the first image feature information corresponding to the i-th image block, represents the first query vector corresponding to the i-th image block, represents the first key vector corresponding to the jth image block, V j represents the first value vector corresponding to the j-th image block, P i,j represents the bias items corresponding to the i-th image block and the j-th image block, m represents the number of the image blocks, and r represents the dimension of the first key vector.

5. The method according to claim 1, characterized in that The operation of detecting the DR image according to the second image feature information by using the quality assessment model to determine the quality score of the DR image includes: The DR image is detected according to the second image feature information by the quality assessment model, and a score corresponding to the scoring parameter is output as the quality score of the DR image.

6. The method according to claim 1, characterized in that Also includes: The DR image is preprocessed, wherein the preprocessing includes removing noise, processing missing values ​​and standardizing data formats.

7. A quality inspection device for digital X-ray DR images, characterized in that: include: a position determination module, configured to segment the DR image into a plurality of image blocks, and determine the absolute positions of the plurality of image blocks, wherein the absolute positions are generated based on the absolute row and column positions or indexes of each image block; A coding determination module, used to determine relative position codes between the plurality of image blocks in different regions according to the absolute positions of the plurality of image blocks; An information generating module, configured to generate corresponding first image feature information according to the vector information of the plurality of image blocks and the relative position codes between the plurality of image blocks through a quality evaluation model; a result output module, configured to decode the first image feature information through the quality assessment model, and output second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate a variety of position information of the DR image; and A score determination module is used to detect the DR image according to the second image feature information through the quality evaluation model to determine the quality score of the DR image, wherein The coding determination module comprises: A first determination submodule is used to identify the DR image and determine the symmetric area and the asymmetric area in the DR image; A first calculation submodule is configured to calculate the relative position code between the first image blocks in the symmetric region according to the absolute position by using the center line in the symmetric region; and The second calculation submodule is used to calculate the relative position code between the second image block and the other image blocks based on the absolute position according to the distance between the second image block and the other image blocks in the asymmetric area.

8. A quality inspection device for digital X-ray DR images, characterized in that: include: processor; as well as A memory, connected to the processor, configured to provide the processor with instructions for processing the following processing steps: Segmenting the DR image into a plurality of image blocks, and determining absolute positions of the plurality of image blocks, wherein the absolute positions are generated based on an absolute row and column position or index of each image block; Determining relative position codes between the plurality of image blocks in different regions according to the absolute positions of the plurality of image blocks; Generate corresponding first image feature information according to the vector information of the plurality of image blocks and the relative position encoding between the plurality of image blocks through a quality assessment model; Decoding the first image feature information by using the quality assessment model, and outputting second image feature information corresponding to the DR image, wherein the second image feature information is used to indicate a variety of position information of the DR image; and The DR image is detected by the quality assessment model according to the second image feature information to determine the quality score of the DR image, wherein The operation of determining the relative position coding between the plurality of image blocks in different regions according to the absolute positions of the plurality of image blocks comprises: Identify the DR image to determine a symmetrical region and an asymmetrical region in the DR image; Calculating the relative position code between the first image blocks in the symmetric region according to the absolute position using the center line in the symmetric region; and The relative position code between the second image block and the other image blocks is calculated based on the absolute position according to the distance between the second image block and the other image blocks in the asymmetric region.

9. 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 claim 1 are implemented.

Images