Digital X-ray photography real-time image quality evaluation control system, equipment and medium

Abstract

The invention relates to the technical field of image quality evaluation, and discloses a digital X-ray photography real-time image quality evaluation control system and device and a medium, and the system comprises a collection end which is used for collecting X-ray image data of a digital radiographic imaging system DROC interface in real time; the server is provided with a storage module, a classification model, a segmentation model, a detection model and a quality control analysis module, and the optimized classification model is used for identifying a detection part of the X-ray image data transmitted by the acquisition end; the optimized segmentation model is used for segmenting a key anatomical part of a detection part in the X-ray image data so as to obtain a segmentation result; the optimized detection model is used for performing high-precision detection on the region of interest in the segmentation result so as to obtain a detection result of the detection part about the quality control index; the quality control analysis module is used for performing quality control analysis on the detection result so as to obtain a quality control analysis result; and the client is used for automatically generating a quality control report. According to the invention, the quality control efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image quality assessment, and in particular to a digital X-ray photography real-time image quality assessment control system, equipment and medium. Background Art

[0002] Digital radiography (DR) has become a core imaging diagnostic method in modern medicine and is widely used for examinations of all parts of the body. Compared with traditional film imaging, DR has the advantages of fast acquisition speed, large amount of image information, and reusability. It is gradually becoming popular in large hospitals and physical examination institutions. However, due to differences in positioning level and operation standardization among imaging technicians in various hospitals, the quality of DR images varies, and it is difficult to achieve mutual recognition of imaging examination results between hospitals. This mutual recognition problem is particularly significant when patients seek medical treatment across institutions, which leads to a waste of medical resources and reduces the patient's medical experience.

[0003] To solve this problem, hospitals at all levels currently generally conduct quality control on X-ray images to improve the consistency and quality of images. However, traditional image quality control methods mainly rely on manual interpretation, requiring technicians to subjectively evaluate image quality, positioning, exposure and other parameters. This method not only requires a lot of experience, but is also limited by the skill level of the operator, resulting in unstable quality control results. In addition, manual quality control is time-consuming and labor-intensive, and it is difficult to cope with large-scale image data, which increases the burden on medical resources.

[0004] In recent years, some quality control systems based on artificial intelligence have gradually entered this field. For example, some quality control systems have tried to use deep learning models to evaluate image quality in order to automatically detect positioning-related indicators. However, the actual application of these systems still has significant shortcomings, including the low real-time and accuracy of quality assessment, the lack of comprehensiveness for different anatomical parts, and the limited setting of evaluation indicators, which makes it difficult to meet the standard requirements of clinical quality control. Summary of the invention

[0005] The present invention provides a real-time image quality assessment and control system for digital X-ray photography, which aims to achieve real-time and accurate quality assessment of X-ray films through automation technology and deep learning models, so as to solve the problems of low quality control efficiency, unstable assessment and high cost of existing manual quality control.

[0006] The present invention is achieved through the following technical solutions: A digital X-ray photography real-time image quality assessment control system, comprising: The acquisition end is used to collect X-ray image data of the DROC interface of the digital radiography system in real time; The server is provided with a storage module, a classification model, a segmentation model, a detection model and a quality control analysis module, wherein the storage module stores sample images and expert consensus quality control standards, and the classification model, segmentation model and detection model are respectively trained and optimized through the sample images, wherein the optimized classification model is used to identify the detection part of the X-ray image data transmitted from the acquisition end, and the optimized segmentation model is used to segment the key anatomical parts of the detection part in the X-ray image data to obtain the segmentation result; the optimized detection model is used to perform high-precision detection of the region of interest in the segmentation result, so as to obtain the detection result of the detection part with respect to the quality control index; the quality control analysis module performs quality control analysis on the detection result based on the expert consensus quality control standard, so as to obtain the quality control analysis result; The client is used to receive and analyze the X-ray image data, segmentation results, detection results and quality control analysis results, so as to automatically generate a quality control report and display the segmentation results, detection results and analysis results in real time.

[0007] As an optimization, the classification model is a combination model of the ResNetXt-50 model and the OCR recognition model, wherein the specific process of training the classification model is: Input the sample image into the ResNetXt-50 model, output the predicted label through the ResNetXt-50 model, and recognize the text information in the sample image as the actual label through the OCR recognition model, wherein the sample image comes from the X-ray image data of the DROC interface, and the sample image is marked with the name of the body part corresponding to the X-ray image data; The predicted label and the actual label are input into a preset loss function, and the parameters of the ResNet50-32x4d model are tuned according to the loss function.

[0008] As an optimization, the segmentation model is an MSU-Unet model, which includes an encoder and a decoder, wherein the encoder and the decoder both include multiple convolutional layers, and the convolutional layers in the encoder are symmetrically arranged with the convolutional layers in the decoder, and the output end of the convolutional layer in the encoder is connected to the input end of the convolutional layer symmetrically arranged in the decoder by a jump connection; each of the convolutional layers includes a first convolution group composed of two first convolutional blocks of 3*3 convolutions arranged in series, a second convolution group composed of two second convolutional blocks of 7*7 convolutions arranged in series, and a third convolutional block of 1*1 convolution, the input image features of the convolutional layer are respectively obtained by the first convolution group and the second convolution group to obtain the first intermediate features and the second intermediate features, the first intermediate features and the second intermediate features are merged by channel splicing and then pass through the third convolution block to output image features.

[0009] As an optimization, in the encoder, the image features output by the convolutional layer are downsampled through a maximum pooling operation.

[0010] As an optimization, in the decoder, the image features output by the convolution layer are upsampled by deconvolution or bilinearity.

[0011] As an optimization, micro-data enhancement is performed on the sample images for training the segmentation model and the classification model. The specific operation process of the micro-data enhancement includes: performing one or more operations of random brightness adjustment, random saturation adjustment, random contrast adjustment, random translation adjustment, and random cropping adjustment on the sample images.

[0012] As an optimization, the client includes: A receiving and storing module, used for receiving and storing the X-ray image data, segmentation results, detection results and quality control analysis results transmitted by the server; A cropping and display module, used for automatically cropping the area that needs quality control in the segmentation result according to the quality control analysis result, and adaptively displaying it. If there is an abnormality in the analysis result, an alarm is issued in the client to prompt the operator to adjust the shooting parameters in time; A correction module, for manually switching the information of the detection part when it is detected that the detection part identified by the server does not match the actual part, and feeding back the modified information to the server for correction; The indicator display module is used to display all quality control indicators related to the detection part. If any of the quality control indicators is abnormal, an abnormal prompt will be given; An automatic evaluation report generation module is used to mark the test results corresponding to the quality control indicators according to the quality control analysis results. When the test results are within the expert consensus quality control standard range, they are marked as normal; when the test results exceed the expert consensus quality control standard range, they are marked as abnormal and output to the report; The historical query management module is used to convert the report into a document format, save it in a memory, and perform query and tracking in the memory.

[0013] As an optimization, both the server and the client are implemented based on ReviewX software.

[0014] The present invention also discloses a device, comprising: a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor realizes the functions of the aforementioned digital X-ray photography real-time image quality assessment control system when executing the computer program.

[0015] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the functions of the aforementioned digital X-ray photography real-time image quality assessment control system are realized.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: The digital X-ray photography real-time image quality assessment and control system of the present invention is based on a three-layer architecture of the acquisition end, the service end and the client end. By combining the deep learning model with the expert consensus quality control standard, it realizes efficient and automated quality control of DR images, and has the following technical effects: 1. Automated image quality assessment: The server automatically analyzes the image quality through classification, segmentation and detection models, which can accurately identify the key anatomical structures in the image and ensure that the X-ray images of different parts meet the quality control standards. The classification model (ResNetXt-50) ensures the accuracy of image part identification; the segmentation model (MSU_UNET) provides clear anatomical region segmentation to ensure the complete details of key parts; the detection model (YoloV8) monitors the quality indicators of the image in real time. In this way, through the automated image analysis process, the time and energy required for manual intervention are reduced, and the efficiency and accuracy of quality control are significantly improved.

[0017] 2. Real-time and efficient feedback: The real-time quality control capability of this system ensures that quality assessment feedback can be obtained immediately after the image is acquired. The client receives and displays the quality control results pushed by the server. If the image has abnormal conditions such as inaccurate positioning, overexposure or underexposure, the system will issue a prompt on the client interface for the technician to make real-time adjustments. The real-time feedback mechanism reduces the number of image retakes, reduces radiation exposure, and improves the patient's diagnosis and treatment efficiency while ensuring image quality.

[0018] 3. Quality control standardization and consistency: This invention adopts the quality control standards of hospital expert consensus. The algorithm on the server compares the analysis results of the deep learning model with the standards, automatically determines whether the image meets the standards and generates a report. This quality control process based on standardization can effectively reduce the result deviation caused by differences in operator experience, achieve consistency in image quality control, provide support for image mutual recognition among multiple hospitals, and help improve the accuracy of diagnosis and treatment.

[0019] 4. Convenient image data storage and management: The client automatically generates image quality control reports and supports exporting reports in PDF format for long-term storage and follow-up. After completing quality control, technicians can directly call historical quality control data through the client, effectively improving the convenience and systematicness of image quality control management and providing data support for the quality control process.

[0020] 5. System scalability and adaptability: Through the flexibility of the multi-model architecture on the server side, this system can not only be applied to DR image quality control of multiple parts such as the chest, abdomen, cervical spine, lumbar spine and knee joints, but can also adapt to images of other anatomical parts according to new quality control requirements. The tunability of the server-side model and the data parsing function of the client ensure the adaptability of the system in different medical environments, making the present invention not only suitable for large-scale medical institutions, but also for image quality control in grassroots hospitals. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 This is the software architecture diagram of ReviewX; Figure 2 This is an example diagram of an interface displayed by a client of the system of the present invention; Figure 3 for Figure 2 An enlarged schematic diagram of the lower middle part; Figure 4 for Figure 2 Enlarged schematic diagram of the upper middle part. DETAILED DESCRIPTION

[0022] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0023] This embodiment 1 provides a digital X-ray photography real-time image quality assessment control system, such as Figure 1 As shown, including: 1. The acquisition end is used to collect X-ray image data of the DROC interface of the digital radiography system in real time; The acquisition end collects X-ray image data of the DROC interface in real time through a high-resolution acquisition device that supports 2k / 4k, 30hz or 60hz to ensure the accuracy of quality control. During the acquisition process, the acquisition end does not process the original data in any way and transmits the acquired images directly to the server through the USB3.0 or Type-C interface.

[0024] Specific operations include: HDMI IN: Connect the DROC device that needs quality control to obtain X-ray images in real time.

[0025] Output interface: The unprocessed raw images are transmitted to the server via the USB3.0 or Type-C interface for further processing and analysis.

[0026] 2. The server is provided with a storage module, a classification model, a segmentation model, a detection model and a quality control analysis module. The storage module stores sample images and expert consensus quality control standards. The classification model, segmentation model and detection model are trained and optimized through the sample images respectively, wherein the optimized classification model is used to identify the detection part of the X-ray image data transmitted from the acquisition end, and the optimized segmentation model is used to segment the key anatomical parts of the detection part in the X-ray image data to obtain the segmentation result; the optimized detection model is used to perform high-precision detection of the region of interest in the segmentation result, so as to obtain the detection result of the detection part with respect to the quality control index; the quality control analysis module performs quality control analysis on the detection result based on the expert consensus quality control standard, so as to obtain the quality control analysis result; The server is responsible for the storage, processing and quality control analysis of image data, and realizes the automatic classification, segmentation and detection of X-ray images by integrating multiple deep learning models and quality control standards. The main components and processes of the server are as follows: Data collection: The server collects tens of thousands of X-ray images of various parts of the body (such as abdomen, cervical spine, knee joint, chest, lumbar spine, etc.). The data comes from hospitals and is used to train and optimize the performance of subsequent models.

[0027] Classification model: ResNetXt-50 (32x4d) model (tuned) is used. For the classification of X-ray images of different parts, especially the difficult-to-distinguish abdominal and lumbar anteroposterior images, the server introduces the segmentation results as a posteriori reference and uses OCR to recognize the anatomical part information of the DROC interface in real time to improve the accuracy and stability of classification.

[0028] Segmentation model: The tuned MSU_UNET model is used to generate medical imaging data close to reality through micro-data enhancement to ensure that the model accurately segments key anatomical parts in the image. The segmentation results are compared with the quality control standards to ensure that they meet the quality requirements of medical images.

[0029] Detection model: Use the YoloV8 model (also tuned) to perform high-precision detection of the region of interest in the image. The server uses the generated data enhancement method to expand the training samples and improve the detection accuracy and reliability of the model.

[0030] Quality control algorithm: Based on the quality control standards formulated by the consensus of hospital experts, combined with the results of classification, segmentation and detection model reasoning, the server calculates the image quality index in real time and conducts quantitative analysis on the image quality of 9 different parts to ensure that it meets the quality requirements of medical images. It should be noted that quantitative analysis of other parts can be performed in the future as required by medical images.

[0031] 3. The client is used to receive and analyze the X-ray image data, segmentation results, detection results and quality control analysis results, so as to automatically generate a quality control report and display the segmentation results, detection results and analysis results in real time. Figure 2-Figure 4 It should be noted that Figure 2 and Figure 4 The data in are hypothetical exemplary data.

[0032] The client is responsible for receiving and displaying quality control results, and managing quality control reports and historical data. The main functions include data analysis, image display, automatic evaluation report generation and historical query. Specific functional modules include: The receiving storage module realizes data reception and analysis, and is used to receive and store the X-ray image data, segmentation results, detection results and quality control analysis results transmitted by the server; specifically, the image data transmitted by the server is received and stored locally to ensure the consistency and real-time performance of the displayed data.

[0033] The cropping display module is used to adaptively display the segmentation results. If there is an abnormality in the segmentation results, an alarm is issued in the client to prompt the operator to adjust the shooting parameters in time. What is displayed here is the original X-ray image after adaptive cropping and the corresponding model segmentation processing result transmitted from the server by the client, and the adaptive display is performed in the client display image area. The adaptive cropping work has been completed during the communication process between the acquisition end device and the server. The purpose is to crop the unnecessary parts of the X-ray image displayed in DROC (for example: dicom information or other irrelevant areas, only retaining the X-ray foreground area). The image displayed by the client is the original X-ray image transmitted by the server after adaptive cropping and the corresponding segmentation result transmitted by the server (for example: the segmentation result of the lung field area, then the lung field area can be drawn in the original X-ray image in the displayed image). The client only needs to perform initial cropping once when it is installed for the first time, that is, set the cropping area (the area used by DROC to display the image). There is no need to perform any cropping operation in the subsequent use of the client, and all are completed automatically on the server. Of course, the cropping function is also retained, and it is also possible if you want to crop it again on the client.

[0034] A correction module, which is used to support the technician to manually switch the information of the detection part when it is detected that the detection part identified by the server does not match the actual part, and feed back the modified information to the server for correction; The indicator display module is used to display all quality control indicators related to the detection part. If any of the quality control indicators is abnormal, an abnormal prompt will be given; for example, if a quality control indicator is abnormal, a flashing prompt will be given.

[0035] An automatic evaluation report generation module is used to mark the test results corresponding to the quality control indicators according to the quality control analysis results. When the test results are within the expert consensus quality control standard range, they are marked as normal; when the test results exceed the expert consensus quality control standard range, they are marked as abnormal and output to the report; The client automatically generates a quality control evaluation report based on the quality control standards and the indicator data (test results) transmitted by the server and compares it with the standards (expert consensus quality control standards). When the data within the quality control standard range is marked as normal, the data outside the standard is marked as abnormal and output to the report.

[0036] The historical query management module is used to convert reports into document formats and save them to storage, where they can be queried and tracked. For example, while saving the quality control assessment report, the system will automatically extract the patient information in the image for confirmation, and convert the report into PDF format and save it to the local hard disk, supporting technicians to query historical reports and track data in the client interface.

[0037] In this embodiment, both the server and the client are implemented based on ReviewX software.

[0038] ReviewX is a software for X-ray imaging quality assessment. It consists of two parts: the server (ReviewX-S) and the client (ReviewX-C). The operating core of ReviewX-S is based on NVIDIA Jetson AGXOrin, which can provide a real-time computing power foundation for the operation of ReviewX-C. ReviewX-S has 4 module nodes, namely the image collector node (ImageCollector Node), image processor node (ImageProcessorNode), main control node (Controller Node) and device gateway node (DeviceGateway Node). ReviewX-C has 4 functional components, namely the video / information display component, system setting component, image cropping component and report saving / query component.

[0039] For ReviewX-S, when the master node receives the startup signal (Signal), that is, when the core startup program is ready, it passes the Signal to the main mode (Main Mode), and then the Main Mode responds to the ImageCollectorNode and ImageProcessor Node respectively.

[0040] When responding to the ImageCollector Node, the Main Mode calls the image acquisition card software program (HDMI SDK) in the node to respond to the camera (Camera) to capture the screen interface of the digital X-ray image acquisition and processing system (DROC) of the host connected to the current HDMI device in real time. The acquired real-time image responds to the Controller Node through the Main Mode, and the results are packaged and published to ReviewX-C through the DeviceGateway Node based on the ZMQ communication protocol.

[0041] When responding to the ImageProcessor Node, the Main Mode calls the image processing software program (Processor SDK) in the node to process the images obtained by the previous response of the Main Mode and the ImageCollector Node in real time. The processed image results are responded to the Controller Node through the Main Mode, and the results are packaged and published to ReviewX-C through the DeviceGateway Node based on the ZMQ communication protocol.

[0042] For ReviewX-C, all contents of ReviewX-S under real-time processing are obtained by subscribing to the contents packaged and published by ReviewX-S. The parsed HDMI data is passed to the video / information display component and the image cropping component for UI interface display. The video / information display component uses the HDMI data transmitted by ReviewX-S for real-time display, and the image cropping component uses the HDMI data transmitted by ReviewX-S for initial cropping. On the one hand, the cropping is used for the UI main control display unit to display the currently processed image, and at the same time, ReviewX-C feeds back the cropping coordinates to ReviewX-S to guide MainMode to adjust the processing results of ImageProcessor Node. On the other hand, according to the specific requirements for DICOM text information preservation in the report preservation component, the specific area of ​​DICOM text information display in the current HDMI data is cropped to preserve the normal operation of the text recognition function in the report component. The image cropping component is only used for the initialization step, and ideally the user only needs to use it once. If it needs to be adjusted according to personal reading habits or other factors, just repeat the initialization operation. For the system setting component, the user only needs to set the communication IP between the software and the device in this environment, and the default is Ipv4. For the report saving / query component, the quality assessment information of the current assessment image and each assessment detail image are saved. At the same time, according to the results of the text recognition function in the image cropping component, the DICOM information of the current processed image is saved for responsibility tracing. Reports are saved locally in ReviewX-C. In the report saving / query component, historical queries can be performed in ReviewX-C according to the report name.

[0043] Next, we will introduce the structure of each model on the server in detail.

[0044] The classification model is a combination model of the ResNetXt-50 model and the OCR recognition model, wherein the specific process of training the classification model is: Input the sample image into the ResNetXt-50 model, output the predicted label through the ResNetXt-50 model, and recognize the text information in the sample image as the actual label through the OCR recognition model, wherein the sample image comes from the X-ray image data of the DROC interface, and the sample image is marked with the name of the body part corresponding to the X-ray image data; The predicted label and the actual label are input into a preset loss function, and the parameters of the ResNetXt-50 model are tuned according to the loss function.

[0045] ResNetXt-50 is a deep convolutional neural network based on the ResNet50 structure, but it uses a specific configuration - 32x4d. This naming is mainly related to the number of channels in its bottleneck layer. The following is a detailed description of the model architecture and principles: 1. The basic architecture of ResNet50: ResNet50 is a residual network that solves the gradient vanishing problem in deep networks by introducing residual blocks. The main idea is to pass the input directly to the subsequent layers through skip connections, so that even if the gradient vanishes in the intermediate layers, the gradient can still be back-propagated through the residual path.

[0046] ResNet50 has 50 layers, which are mainly composed of convolutional layers and batch normalization layers. A key module in ResNet is the bottleneck block. In ResNet50, the residual block uses three convolutional layers: 1x1 convolution, used to reduce the number of channels. 3x3 convolution, used for feature extraction. 1x1 convolution, used to increase the number of channels. Residual connection adds the input directly to the output, forming a similar skip connection to help the network learn deeper features.

[0047] Meaning of 32x4d: In ResNetXt-50, 32x4d is a modification of the channel configuration of the convolution layer in the residual block. 32: represents the number of convolution groups contained in each group convolution. 4d: represents the cardinality in each group, meaning that the 3x3 convolution in each residual bottleneck block uses a configuration of 4 basic channels (or feature map dimensions). Here d represents the channel (dimension).

[0048] Therefore, ResNetXt-50 uses a group convolution method, which introduces multiple groups in the convolution layer to process different input channels. This method helps to reduce computational complexity while increasing the feature expression ability of the model. This is different from the standard ResNet, which does not use group convolution or uses a smaller number of groups. Specifically: In the traditional ResNet50, 3x3 convolution does not use group convolution. In ResNetXt-50, the 3x3 convolution of the bottleneck block is split into 32 groups, and the convolution kernel of each group only processes the input of 4 channels.

[0049] 2. The main layers of the ResNetXt-50 architecture Convolutional and pooling layers: The network first passes through a 7x7 convolutional layer and a max pooling layer to reduce the size of the input and increase the number of channels.

[0050] Residual blocks: ResNet50 is mainly composed of 4 stages, each stage contains multiple residual blocks. In ResNetXt-50, these residual blocks have been adjusted in terms of the number of channels and the grouping of convolutional layers, with 32 groups performing 3x3 convolution operations.

[0051] Global pooling and fully connected layers: After the last layer of convolution, the network compresses all feature maps into a feature vector through global average pooling. Finally, the feature vector is output through a fully connected layer, which is usually used for classification tasks.

[0052] 3. Principle: The design of ResNetXt-50 mainly follows the following two principles: 1. Residual learning: Residual connections are used to help deep networks learn, so that the network can continue to be effectively trained when the number of layers is increased, avoiding the gradient vanishing problem.

[0053] 2. Group convolution: By using group convolution to reduce the amount of calculation while maintaining high feature expression capabilities. Group convolution can reduce the number of parameters for each convolution operation, thereby reducing computational complexity, and enrich features by processing different channels in groups.

[0054] In some embodiments, the segmentation model is an MSU-Unet model, which includes an encoder and a decoder, wherein the encoder and the decoder both include multiple convolutional layers, and the convolutional layers in the encoder are symmetrically arranged with the convolutional layers in the decoder, and the output end of the convolutional layer in the encoder is connected to the input end of the convolutional layer symmetrically arranged in the decoder by a jump connection; each of the convolutional layers includes a first convolution group composed of two first convolution blocks of 3*3 convolutions arranged in series, a second convolution group composed of two second convolution blocks of 7*7 convolutions arranged in series, and a third convolution block of 1*1 convolution, and the input image features of the convolutional layer are respectively obtained by the first convolution group and the second convolution group to obtain the first intermediate features and the second intermediate features, and the first intermediate features and the second intermediate features are merged by channel splicing and then output through the third convolution block to output image features.

[0055] In the encoder, the image features output by the convolutional layer are downsampled through a maximum pooling operation.

[0056] In the decoder, the image features output by the convolutional layer are upsampled by deconvolution or bilinearity.

[0057] MSU-Unet (Multi-Scale U-Net) is an advanced medical image segmentation model based on the classic U-Net architecture. It improves on U-Net by integrating multi-scale convolutional blocks (convolutional kernels with different receptive fields). This design allows for better feature extraction at different image scales, making it more effective for dense segmentation tasks such as detecting small or detailed structures. The multi-scale approach enhances the network's ability to capture a variety of semantic features, thereby improving the segmentation performance of medical images.

[0058] Architecture level: The overall architecture of MSU_UNet is similar to that of U-Net, both of which consist of an encoder (downsampling path) and a decoder (upsampling path), forming a U-shaped network structure.

[0059] The encoder is responsible for extracting features, gradually reducing the spatial dimensions and increasing the feature channels; the decoder gradually restores the spatial dimensions through upsampling while fusing the jump connections from the encoder.

[0060] Main components: Encoder part: The encoder consists of multiple layers of convolutional blocks, and conv_3_1 modules are used in each convolutional block. conv_3_1 is the core module of MSU_UNet, which combines convolution kernels of different scales, such as 3×3 and 7×7, to capture more contextual information. Specifically, it uses two convolutional blocks: conv_block_3: 3×3 convolution. conv_block_7: 7×7 convolution.

[0061] The results of these convolution operations are combined through channel concatenation (torch.cat) and then 1×1 convolution to reduce the number of channels. The encoder layer starts with shallow features, gradually downsamples the image through the max pooling operation, and processes feature maps of 32, 64, 128, 256, and 512 channels in turn.

[0062] Decoder part: The decoder part restores the feature map to the original resolution through deconvolution (transposed convolution) or bilinear upsampling. After each decoding step, the feature map from the encoder is spliced ​​with the output of the decoder through jump connections (torch.cat) to achieve multi-level feature fusion. In each convolution operation after upsampling, the conv_3_1 module is also used to maintain the multi-scale feature extraction capability. The number of feature channels after upsampling is gradually reduced, and finally a single-channel result map is output through 1×1 convolution.

[0063] Working principle: Multi-scale convolution: By using convolution kernels of different sizes (such as 3×3 and 7×7) in each convolution block, MSU_UNet can effectively capture image features of different sizes and improve the accuracy of segmentation by fusing this multi-scale information.

[0064] Skip connections: Skip connections concatenate the feature maps of each layer in the encoder with the output of the corresponding layer in the decoder, which helps preserve more low-level detail information and helps the network better recover the spatial structure of the image.

[0065] Upsampling: Use deconvolution or bilinear interpolation to gradually restore the resolution and combine the features of the encoder to gradually refine the segmentation result.

[0066] In this embodiment, micro-data enhancement is performed on the sample images for training the segmentation model and the classification model. The specific operation process of the micro-data enhancement includes: performing one or more operations of random brightness adjustment, random saturation adjustment, random contrast adjustment, random translation adjustment, and random cropping adjustment on the sample images.

[0067] A micro-data enhancement method, wherein each m(x) is replaced by m(T(x)), wherein x can be an original image Img1 or a preprocessed image Img2, m is a segmentation model or a classification model, and T is a micro-data enhancement operation process; the micro-data enhancement operation process T includes the following at least one dimension adjustment: 1). Random brightness adjustment: add random brightness x to the input tensor; 2). Random saturation adjustment: x adjusts the saturation of the input tensor by randomly scaling the color channels of the input tensor; 3). Random contrast adjustment: x adjusts the contrast of the input tensor by randomly scaling pixel values; 4). Random translation adjustment: x applies a random translation to the input tensor at a specified ratio of 0.125; 5). Random Crop Adjustment: Random shearing is applied by zeroing a rectangular area within the input tensor. The size and position of the cuts are random.

[0068] In this embodiment, the quality control indicators involved in the expert consensus quality control standard include: Chest Position: Clavicle angle, centering, scapular rotation, exposure range, inhalation detection, chest rotation; 《Chest Lateral Position》: Exposure range, inhalation detection, midaxillary line angle, arms raised, chest rotation; "abdomen": Exposure range; Knee Position: The fibular head overlap, exposure range, and positioning are centered without deviation; Lateral position of the knee: fibular head overlap, knee flexion, and medial and lateral condylar overlap; Cervical Spine Position: Exposure range, mandibular and occipital overlap detection, positive detection, and mandibular left-right asymmetry detection; 《Cervical Spine Correction Functional Position》: Exposure range, mandibular and occipital overlap detection, positive detection, and mandibular left-right asymmetry detection; Lateral position of cervical spine: Exposure detection, mandibular angle overlap detection, cervical spine and mandibular overlap detection; Lumbar Spine Correction: Upper and lower exposure range, pelvic alignment and left and right exposure range, spinal symmetry detection, centerline detection; Lumbar lateral position: Exposure range and lumbar alignment detection.

[0069] It should be noted that this system adopts a three-layer architecture of acquisition end, service end and client end to achieve full-process quality control of X-ray images. The architecture includes a fully automatic workflow from image acquisition to processing, analysis, quality control feedback and storage, which is particularly suitable for real-time quality control needs in medical scenarios. The system architecture design has wide adaptability and can be expanded to different medical devices and medical scenarios, which is one of the key innovations of this invention.

[0070] The present invention integrates a variety of deep learning models on the server side to automatically classify, segment and detect X-ray images of different parts. These include the tuned ResNet50-32x4d classification model, MSU_UNET segmentation model and YoloV8 detection model to ensure accurate analysis of image quality. These models, combined with data enhancement and OCR recognition technology, help to achieve efficient and stable quality control. However, since some models are open source projects, they may not be covered by the patent protection scope, but their unique model tuning and application configuration are important components of the invention.

[0071] The present invention adopts the quality control standards formulated by expert consensus, combined with the quality control data (test results) output by the deep learning model, and realizes the multi-index quantitative evaluation of image quality through the algorithm. This set of post-processing algorithms is based on the expert consensus standards, and realizes real-time and automated image quality evaluation through the integrated analysis of classification, segmentation, and test results. This algorithm can not only accurately control the quality of the current image, but also can be used as a standardized tool for quality control indicators in other types of DR systems, and has the potential for cross-system transplantation.

[0072] The client automatically generates a quality control evaluation report by analyzing the quality control data of the server in real time, and supports export and storage in PDF format. At the same time, the report history management function of the client provides a convenient storage and query method for subsequent quality control tracking and data review. This report management system can be applied to other medical imaging quality control systems and has a high technical protection value.

[0073] Next, the system of the present invention is described in detail through specific examples.

[0074] Case 1: System architecture and data stream processing: Acquisition end: Connect to the DROC (digital radiography system) device through the HDMI interface, collect X-ray images in real time with 2k / 4k resolution, and transmit the images to the server using the USB3.0 or Type-C interface.

[0075] Server: After the image data is transmitted to the server, the server first identifies the shooting part of the image through the classification model, and transmits the image to the segmentation and detection module to obtain the segmentation results and detection results of key anatomical structures and quality control indicators in the image. The quality control algorithm on the server calculates the quality control indicators in real time according to the expert consensus standard and transmits the results to the client.

[0076] Client: The client receives and analyzes the quality control data from the server, automatically generates a quality control report, and displays the quality control results in real time. In abnormal situations (such as overexposure or inaccurate positioning), the client will issue an alarm for the operator to adjust the shooting parameters in time.

[0077] Case 2: Quality control assessment of chest X-ray images: Acquisition process: The acquisition end captures the chest frontal image and transmits it to the server.

[0078] Classification model: The server uses the classification model to identify the image as a chest frontal image.

[0079] Quality control evaluation: The server segmented and detected the image, and calculated quality control indicators including clavicle angle, scapular rotation, exposure range, inhalation detection, etc. If an indicator exceeds the quality control standard range (for example, the clavicle angle exceeds 30°), the server algorithm will generate a quality control warning for the abnormal indicator.

[0080] Report generation: The client receives quality control data and automatically generates a report, prompting the technician to adjust his posture to ensure that the image quality meets the requirements.

[0081] Case 3: Quality control assessment of lumbar spine X-ray images: Acquisition process: The acquisition end acquires the lumbar spine AP images and transmits them to the server end.

[0082] Segmentation model: The server-side segmentation model identifies key anatomical regions of the lumbar spine, such as the centerline of the spine and the pelvis on both sides.

[0083] Quality control evaluation: The server performs tests based on quality control standards such as the symmetry of the spine and pelvis, and calculates key quality control parameters such as upper and lower exposure range, pelvic symmetry, and centerline symmetry. If a parameter (such as whether the pelvis is aligned) does not meet the standard, the server will generate an abnormality alert.

[0084] Report display: The client generates a detailed quality control report, which contains the calculation results of each quality control indicator detected, and displays prompt symbols next to abnormal items to remind technicians to make necessary adjustments.

[0085] Case 4: Generation and management of quality control reports: Real-time report generation: After each image quality control, the client automatically generates a report containing various quality control indicators, distinguishing between qualified items and abnormal items.

[0086] Historical report management: Quality control reports are stored in PDF format on the client's hard disk. Technicians can view and manage saved quality control reports through the client's historical query function, which facilitates long-term tracking and analysis of quality control data.

[0087] Embodiment 2 also discloses a device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the function of a digital X-ray photography real-time image quality assessment control system of embodiment 1 when executing the computer program.

[0088] Embodiment 3 further discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the functions of a digital X-ray photography real-time image quality assessment control system of embodiment 1 are realized.

[0089] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A digital X-ray photography real-time image quality assessment control system, characterized in that: include: The acquisition end is used to collect X-ray image data of the DROC interface of the digital radiography system in real time; The server is provided with a storage module, a classification model, a segmentation model, a detection model and a quality control analysis module, wherein the storage module stores sample images and expert consensus quality control standards, and the classification model, segmentation model and detection model are respectively trained and optimized through the sample images, wherein the optimized classification model is used to identify the detection part of the X-ray image data transmitted from the acquisition end, and the optimized segmentation model is used to segment the key anatomical parts of the detection part in the X-ray image data to obtain the segmentation result; the optimized detection model is used to perform high-precision detection of the region of interest in the segmentation result, so as to obtain the detection result of the detection part with respect to the quality control index; the quality control analysis module performs quality control analysis on the detection result based on the expert consensus quality control standard, so as to obtain the quality control analysis result; The client is used to receive and analyze the X-ray image data, segmentation results, detection results and quality control analysis results, so as to automatically generate a quality control report and display the segmentation results, detection results and analysis results in real time.

2. A digital X-ray photography real-time image quality assessment control system according to claim 1, characterized in that: The classification model is a combination model of the ResNetXt-50 model and the OCR recognition model, wherein the specific process of training the classification model is: Input the sample image into the ResNetXt-50 model, output the predicted label through the ResNetXt-50 model, and recognize the text information in the sample image as the actual label through the OCR recognition model, wherein the sample image comes from the X-ray image data of the DROC interface, and the sample image is marked with the name of the body part corresponding to the X-ray image data; The predicted label and the actual label are input into a preset loss function, and the parameters of the ResNetXt-50 model are tuned according to the loss function.

3. A digital X-ray photography real-time image quality assessment control system according to claim 1, characterized in that: The segmentation model is an MSU-Unet model, which includes an encoder and a decoder, wherein the encoder and the decoder both include multiple convolutional layers, and the convolutional layers in the encoder are symmetrically arranged with the convolutional layers in the decoder, and the output end of the convolutional layer in the encoder is connected to the input end of the convolutional layer symmetrically arranged in the decoder by means of a jump connection; each of the convolutional layers includes a first convolution group consisting of two first convolutional blocks of 3*3 convolutions arranged in series, a second convolution group consisting of two second convolutional blocks of 7*7 convolutions arranged in series, and a third convolutional block of 1*1 convolution, the input image features of the convolutional layer are respectively obtained by the first convolutional group and the second convolutional group to obtain the first intermediate features and the second intermediate features, the first intermediate features and the second intermediate features are merged by channel splicing and then output image features through the third convolutional block.

4. A digital X-ray photography real-time image quality assessment control system according to claim 3, characterized in that: In the encoder, the image features output by the convolutional layer are downsampled through a maximum pooling operation.

5. A digital X-ray photography real-time image quality assessment control system according to claim 3, characterized in that: In the decoder, the image features output by the convolutional layer are upsampled by deconvolution or bilinearity.

6. A digital X-ray photography real-time image quality assessment control system according to claim 1, characterized in that: Micro-data enhancement is performed on the sample images for training the segmentation model and the classification model. The specific operation process of the micro-data enhancement includes: performing one or more operations of random brightness adjustment, random saturation adjustment, random contrast adjustment, random translation adjustment, and random cropping adjustment on the sample images.

7. A digital X-ray photography real-time image quality assessment control system according to claim 1, characterized in that: The client comprises: A receiving and storing module, used for receiving and storing the X-ray image data, segmentation results, detection results and quality control analysis results transmitted by the server; A cropping and display module, used for automatically cropping the area that needs quality control in the segmentation result according to the quality control analysis result, and adaptively displaying it. If there is an abnormality in the analysis result, an alarm is issued in the client to prompt the operator to adjust the shooting parameters in time; A correction module, used for manually switching the information of the detection part when it is detected that the detection part identified by the server does not match the actual part, and feeding back the modified information to the server for correction; The indicator display module is used to display all quality control indicators related to the detection part. If any of the quality control indicators is abnormal, an abnormal prompt will be given; An automatic evaluation report generation module is used to mark the test results corresponding to the quality control indicators according to the quality control analysis results. When the test results are within the expert consensus quality control standard range, they are marked as normal; when the test results exceed the expert consensus quality control standard range, they are marked as abnormal and output to the report; The historical query management module is used to convert the report into a document format, save it in a memory, and perform query and tracking in the memory.

8. A digital X-ray photography real-time image quality assessment control system according to claim 1, characterized in that: The server and the client are both implemented based on ReviewX software.

9. A device, characterized in that: include: A memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the function of a digital X-ray photography real-time image quality assessment control system as described in any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the functions of a digital X-ray photography real-time image quality assessment control system according to any one of claims 1 to 8 are realized.

Images