Collimator adjustment method and related apparatus

Abstract

Embodiments of the present application provide a collimator adjustment method and related equipment. The collimator adjustment method comprises: obtaining a medical image and an examination protocol; calculating a target position of a collimator based on the medical image and the examination protocol; and controlling the collimator to adjust according to the target position. The technical solution of the embodiments of the present application can automatically calculate the target position of the collimator according to the medical image and the current examination protocol of the collimator, and control the collimator to adjust accordingly. Since the image features of the low-density region are relatively simple, the robustness and accuracy of the segmentation algorithm for the low-density region are much higher than that of directly segmenting the complex ROI, and therefore the present method can achieve stable, reliable and highly accurate automatic collimation effect.

Description

Technical Field

[0001] This application relates to the fields of computer and communication technology, and more specifically, to a collimator adjustment method and related equipment. Background Technology

[0002] Currently, collimator adjustment is mainly done manually. Operators usually need to observe the X-ray image while operating the collimator control device for adjustment, which is not only inconvenient but also subject to additional radiation dose. If adjustment is performed during surgery, it will also waste surgical time.

[0003] Existing automatic adjustment methods mostly rely on directly identifying complex regions of interest. However, due to the diversity of human anatomy, the identification of regions of interest is extremely difficult and the accuracy is not high, resulting in unsatisfactory automatic adjustment effects. Summary of the Invention

[0004] The embodiments of this application provide a collimator adjustment method and related equipment, which can at least to some extent overcome the problem that existing automatic adjustment methods rely heavily on direct identification of complex regions of interest, resulting in unsatisfactory automatic adjustment effects.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] According to one aspect of the embodiments of this application, a collimator adjustment method is provided, comprising: acquiring medical images and an examination protocol; calculating a target position of the collimator based on the medical images and the examination protocol; and controlling the collimator to adjust according to the target position.

[0007] According to one aspect of the embodiments of this application, a collimator adjustment device is provided, the collimator adjustment device comprising: an examination protocol acquisition module for acquiring medical images and an examination protocol; a target position determination module for calculating a target position of the collimator based on the medical images and the examination protocol; and a target position adjustment module for controlling the collimator to adjust according to the target position.

[0008] According to one aspect of the embodiments of this application, a computer program product is provided, including one or more computer programs that, when executed by one or more processors, implement the steps of the collimator adjustment method as described in the above embodiments.

[0009] According to one aspect of the embodiments of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the collimator adjustment method as described in the above embodiments.

[0010] According to one aspect of the embodiments of this application, an electronic device is provided, including: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the collimator adjustment method as described in the above embodiments.

[0011] In some embodiments of this application, the target position of the collimator can be automatically calculated based on medical images and the current collimator inspection protocol, and the collimator can be controlled to make corresponding adjustments. Compared with the traditional manual collimator adjustment method, this method greatly improves the accuracy and efficiency of adjustment. It can avoid the problem of poor imaging effect caused by the subjectivity and lag of manual adjustment, ensuring that the collimator is in the optimal position when acquiring medical images, thereby improving the quality of medical images, providing doctors with clearer and more accurate diagnostic basis, and also helping to reduce unnecessary radiation doses and protect the health of patients. Since the image features of low-density areas are relatively simple, the robustness and accuracy of the segmentation algorithm for them are much higher than those for directly segmenting complex ROIs. Therefore, this method can achieve stable, reliable and highly accurate automatic collimation effect.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings: Figure 1 A schematic diagram of an exemplary system architecture to which the technical solutions of the embodiments of this application can be applied is shown.

[0014] Figure 2 A flowchart illustrating a collimator adjustment method is shown.

[0015] Figure 3 It shows according to Figure 2 A flowchart illustrating a specific implementation of step S200 in the collimator adjustment method shown in the corresponding embodiment.

[0016] Figure 4 It shows according to Figure 3 A flowchart illustrating a specific implementation of step S210 in the collimator adjustment method shown in the corresponding embodiment.

[0017] Figure 5 It shows according to Figure 4 A flowchart illustrating a specific implementation of step S214 in the collimator adjustment method shown in the corresponding embodiment.

[0018] Figure 6 The diagram illustrates the neural network architecture of the low-density region segmentation model provided in this embodiment.

[0019] Figure 7 A diagram illustrating the connected domain adjustment process provided in an embodiment of this application is shown.

[0020] Figure 8 It shows according to Figure 3 A flowchart illustrating a specific implementation of step S220 in the collimator adjustment method shown in the corresponding embodiment.

[0021] Figure 9 A schematic diagram of a collimator adjustment device provided in an embodiment of this application is shown.

[0022] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0023] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0024] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0025] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0026] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0027] Figure 1 A diagram illustrating the implementation environment of the collimator adjustment method provided in one embodiment, such as... Figure 1 As shown, in this implementation environment, there are collimator 100, human tissue 200, detector 300, data processing unit 400, display unit 500 and control unit 600.

[0028] The X-ray source generates X-rays with a certain energy and intensity under the action of a high-voltage electric field. The X-ray exit of the X-ray source is connected to a collimator 100, which includes an opaque shield, a semi-transparent shield, and a spectral filter. The opaque shield is typically composed of multiple blades made of high-density materials such as lead or tungsten, used to block X-rays outside the field of view. The semi-transparent shield is typically made of materials such as aluminum or stainless steel, and has finger-shaped or wedge-shaped forms, used to block low-density areas in the image. The spectral filter is made of metal materials such as aluminum or copper. A detector 300 receives the X-rays to generate an image, which is then transmitted to a data processing unit 400. After image processing, the image is transmitted to a display unit 500 for display. Simultaneously, the data processing unit 400, connected to the collimator, can transmit commands to adjust the shields, filters, and other components on the collimator 100 to the desired positions. The user can also control the collimator 100 through a control unit 600 connected to the data processing unit 400.

[0029] When automatically adjusting the collimator 100, the data processing unit 400 first acquires the medical image and the examination protocol in the collimator 100; based on the medical image and the examination protocol in the collimator 100, it calculates the target position of the collimator 100; and controls the collimator 100 to adjust according to the target position.

[0030] The detector 300, data processing unit 400, display unit 500 and control unit 600 can be connected by wired, wireless or other communication methods, and the present invention does not impose any restrictions.

[0031] The implementation details of the technical solutions in the embodiments of this application are described in detail below: Figure 2 A flowchart of a collimator adjustment method according to an embodiment of this application is shown. This collimator adjustment method can be executed by a server, which may be... Figure 1 The server shown. (Refer to...) Figure 2 As shown, the collimator adjustment method includes at least the following: S100, Acquisition of medical images and examination protocols.

[0032] S200, based on the medical image and the examination protocol, calculate the target position of the collimator.

[0033] S300, control the collimator to adjust according to the target position.

[0034] The embodiments of this application can automatically calculate the target position of the collimator based on medical images and the current collimator inspection protocol, and control the collimator to make corresponding adjustments. Compared with the traditional manual collimator adjustment method, this method greatly improves the accuracy and efficiency of adjustment. It can avoid the problem of poor imaging effect caused by the subjectivity and lag of manual adjustment, ensuring that the collimator is in the optimal position when acquiring medical images, thereby improving the quality of medical images, providing doctors with clearer and more accurate diagnostic basis, and also helping to reduce unnecessary radiation doses and protect the health of patients. Since the image features of low-density areas are relatively simple, the robustness and accuracy of the segmentation algorithm for them are much higher than those for directly segmenting complex ROIs. Therefore, this method can achieve stable, reliable and highly accurate automatic collimation effect.

[0035] In S100, this is typically used in medical imaging equipment such as X-ray machines and CT scanners. These devices generate corresponding medical image data when scanning patients. Taking an X-ray machine as an example, the X-ray detector receives X-rays passing through the human body and converts them into electrical or digital signals. After a series of processing steps, an analyzable medical image is formed. This image can be a two-dimensional planar image or three-dimensional volumetric data, depending on the type of imaging equipment and scanning mode. The acquired medical image serves as the basis for subsequent analysis and calculation of the collimator target position.

[0036] The examination protocol refers to the organ position and type of examination, such as coronary angiography or lower extremity angiography. Different examination protocols correspond to the imaging of different organs, and the types of images that need to be generated are also different. Different types of images require different collimator baffles. Collimators generally contain various baffles, such as opaque baffles (used to limit the irradiation range of X-rays) and semi-transparent baffles (which can reduce the irradiation intensity of specific areas).

[0037] In S200, image recognition technology is used to identify key regions in images, such as specific organs and tissues of the human body. For example, in a chest X-ray image, the lung region is identified. After identifying the key region, its shape, size, location, and other features are analyzed.

[0038] The analyzed medical image features are combined with the acquired examination protocol. For example, for imaging the lung region, a wedge filter is typically used to generate an organ mask image. In this case, the target position to which the wedge filter needs to be adjusted to optimize lung imaging needs to be calculated based on the specific location and shape of the lung region. The calculation process may involve various methods such as geometric calculations and data analysis to ensure that the calculated target position accurately meets the requirements of medical image acquisition.

[0039] Specifically, in some embodiments, the specific implementation of step S200 can be found in [reference needed]. Figure 3 . Figure 3 It is based on Figure 2 According to the detailed description of step S200 in the collimator adjustment method shown in the corresponding embodiment, step S200 in the collimator adjustment method may include the following steps: S210, based on the inspection protocol, the medical image is converted into a corresponding low-density area masking image.

[0040] S220, Calculate the target position of the collimator based on the low-density region masking image.

[0041] In this embodiment, medical images are converted into low-density area masking images based on an examination protocol, and the collimator target position is calculated accordingly, making collimator adjustment more precise and targeted. This method can effectively identify low-density areas in medical images, such as lungs and background areas, and then optimize the masking effect on these areas by reasonably adjusting the collimator baffle, further improving the quality of medical images, reducing unnecessary radiation to low-density areas, and ensuring clear imaging of high-density areas, thus providing a more reliable basis for medical diagnosis.

[0042] In S210, the preprocessed medical images are analyzed according to the acquired examination protocol. For example, if the examination protocol is a lung CT scan, the analysis focuses only on low-density features in body tissue regions such as the lungs. Image processing algorithms are used to enhance or mark low-density regions in the images. For example, a threshold segmentation algorithm is used to initially identify potential low-density regions based on the density differences of different tissues. These initially identified regions are then adjusted and optimized, taking into account the current occlusion range and effect of the collimator baffle, highlighting low-density regions requiring further processing.

[0043] After the above processing, the marked low-density regions are used to generate mask images. Mask images are typically binary images, where low-density regions are represented by a specific value (e.g., 1), and other regions are represented by another value (e.g., 0). For example, in a mask image of low-density regions of the lungs, the lung region is represented by 1, and the non-lung region by 0. Such mask images clearly show the location and shape of low-density regions in medical images, providing an intuitive data basis for subsequent calculation of the collimator target position.

[0044] Specifically, in some embodiments, the specific implementation of step S210 can be found in [reference needed]. Figure 4 . Figure 4 It is based on Figure 3 According to the detailed description of step S210 in the collimator adjustment method shown in the corresponding embodiment, step S210 in the collimator adjustment method may include the following steps: S212, preprocess and segment the medical image to obtain the global image and multiple corresponding image sub-blocks.

[0045] S214, based on the inspection protocol, the global image and the corresponding multiple image sub-blocks are input together into the corresponding low-density region segmentation model to obtain the corresponding low-density region masking image.

[0046] In this embodiment, a mask image is generated by preprocessing and segmenting the medical image and combining it with a specific low-density region segmentation model input from the examination protocol. This approach makes the identification and extraction of low-density regions more accurate and comprehensive. It fully utilizes the global information and local details of the medical image, while considering the impact of the examination protocol on image analysis. This provides high-quality mask image data for the subsequent accurate calculation of the collimator target position, thereby further optimizing the collimator adjustment effect, improving the quality of medical images, reducing unnecessary radiation, and providing more reliable image evidence for medical diagnosis.

[0047] In S212, a single raw medical image is processed into two different forms of data for use by subsequent neural network models.

[0048] The global image is a low-resolution version of the original image obtained by scaling it up, which represents the overall image context. Multiple image sub-blocks are small, high-resolution image patches divided from the original image, used to preserve local detail information of the image.

[0049] Specifically, in some embodiments, the specific implementation of step S212 can be found in the following embodiments. This embodiment is based on... Figure 4 According to the detailed description of step S212 in the collimator adjustment method shown in the corresponding embodiment, step S212 in the collimator adjustment method may include the following steps: Medical images are normalized to grayscale to obtain normalized images.

[0050] The normalized image is scaled and segmented to obtain a global image and multiple corresponding image sub-blocks.

[0051] In this embodiment, by sequentially performing grayscale normalization and scaling segmentation operations on the medical images, the original medical images can be effectively processed, making them more suitable for the input requirements of low-density region segmentation models. Grayscale normalization eliminates the differences in grayscale values ​​under different imaging conditions, providing a unified standard for subsequent processing; scaling segmentation, while preserving global information, acquires rich local detail information. Combined with the inspection protocol input to the segmentation model, it can more accurately generate low-density region masking images, thereby providing strong support for accurately calculating the collimator target position. Ultimately, it achieves optimized collimator adjustment, improves medical image quality, reduces radiation dose, and provides higher-quality image data for medical diagnosis.

[0052] The grayscale range of raw medical images may vary due to differences in imaging equipment and shooting parameters. To eliminate these differences and ensure a uniform standard for subsequent processing, grayscale normalization is performed on medical images. Specifically, a histogram is first used to determine the window width and level, then the data is truncated and normalized, mapping it to the 0-1 range to obtain a normalized image. After grayscale normalization, different medical images become comparable at the grayscale level, which is helpful for subsequent image processing and analysis.

[0053] To accommodate subsequent processing and model input requirements, the normalized image is scaled. Typically, the image is scaled to a fixed size, such as a square image with sides of 512 pixels. If the original image is rectangular, it is scaled proportionally and appropriately padded. This ensures that the main information of the image is not lost while meeting the model's input size requirements. The scaled image is the global image, which preserves the overall structure and macroscopic features of the medical image.

[0054] After obtaining the global image, it is segmented to capture local details. The global image is divided into multiple smaller image blocks according to certain rules. For example, a 128×128 pixel block size is used, with a stride of 64 pixels. Starting from the top left corner of the global image, 128×128 pixel blocks are successively extracted at stride intervals until the entire global image is covered. These resulting image blocks contain information from different local regions of the global image, and the stride setting allows for partial overlap between adjacent blocks, preventing information loss. These image blocks, along with the global image, will serve as input data for the subsequent low-density region segmentation model.

[0055] In S214, the segmentation model needs to use both global image and multiple image sub-blocks for comprehensive judgment. The inspection protocol defines the prerequisite for this operation: the system performs segmentation based on the valid image regions determined by the inspection protocol, ensuring the effectiveness of the processing.

[0056] Specifically, in some embodiments, the specific implementation of step S214 can be found in [reference needed]. Figure 5 . Figure 5 It is based on Figure 4 The detailed description of step S214 in the collimator adjustment method shown in the corresponding embodiment, wherein the low-density region segmentation model in the collimator adjustment method includes a global network and a local network, and step S214 may include the following steps: S2142, Based on the inspection protocol, call the corresponding low-density region segmentation model.

[0057] S2144, Input the global image into the global network to obtain a global feature map with the same size as the global image.

[0058] S2146, Each of the image sub-blocks is sequentially connected to the global feature map and then input into the local network to obtain a complete probability map.

[0059] S2148, perform post-processing on the complete probability map to obtain the corresponding low-density region masking image.

[0060] In this embodiment, please refer to the following: Figure 6 By employing a low-density region segmentation model incorporating both global and local networks, this method fully integrates global information and local details from medical images. Combined with examination protocols, it more accurately identifies and segments low-density regions, generating high-quality masked images. This provides more accurate data support for subsequent collimator target position calculations, enabling more precise collimator adjustment, further improving medical image quality, effectively reducing radiation dose to patients, and providing more reliable and clear image evidence for medical diagnosis.

[0061] In S2142, based on the currently acquired inspection protocol, the most suitable model is selected from multiple pre-trained low-density region segmentation models. Different inspection protocols may imply different masking objects, and different types of mask maps should be generated.

[0062] Therefore, by matching examination protocols with corresponding segmentation models, the accuracy of low-density region segmentation can be improved. These pre-trained models are trained using a large amount of medical image data under different examination protocols to adapt to various real-world scenarios.

[0063] In S2144, the pre-processed segmented global image is input into the global network of a low-density region segmentation model. The global network typically employs an architecture capable of effectively extracting global image features, such as an encoder-decoder structure based on a convolutional neural network (CNN). In the global network, a series of convolutional layers first extract features from the global image. The convolutional kernels slide across the image, extracting feature information at different scales and orientations. As the convolutional layers deepen, the spatial resolution of the image gradually decreases, but the feature dimension gradually increases, enabling the network to capture the macroscopic structure and overall features of the image. After processing by the encoder, the decoder performs upsampling to restore the spatial resolution of the image, ultimately obtaining a global feature map of the same size as the global image. This global feature map contains the overall structure and layout information of the medical image, providing a foundation for subsequent fusion with local features.

[0064] In step S2146, each image sub-block is sequentially connected to a feature map of the same size and location cropped from the global feature map. Specifically, this connection can be performed along the channel dimension, allowing each image sub-block to incorporate global feature information. The connected feature data is then input into a local network. This local network, also based on a CNN architecture, focuses on extracting local detail features from the image sub-blocks. Through a series of convolution, pooling, and activation operations, the local network further extracts subtle features from the image sub-blocks. These features complement the global features, helping to more accurately identify low-density regions. After processing by the local network, a complete probability map is output, where each pixel value represents the probability that the corresponding location belongs to a low-density region.

[0065] Specifically, in some embodiments, the specific implementation of step S2146 can be found in the following embodiments. This embodiment is based on... Figure 5 According to the detailed description of step S2146 in the collimator adjustment method shown in the corresponding embodiment, step S2146 in the collimator adjustment method may include the following steps: Based on the spatial position of the image sub-block in the original image, the corresponding feature sub-block is cropped from the global feature map.

[0066] At the data level, the image sub-block and its corresponding feature sub-block are connected to form a fused sub-block.

[0067] The fused sub-block is input into the local network to generate the corresponding local probability map.

[0068] The complete probability map is obtained by stitching together the various local probability maps in sequence.

[0069] In this embodiment, corresponding feature blocks are cropped from the global feature map based on their spatial location in the original image, and then connected at the data level to form fused blocks. A local network is then used to generate a local probability map, which is then stitched together to obtain a complete probability map. This method accurately fuses global and local information. It allows the low-density region segmentation model to fully utilize features from different layers of medical images, more accurately identifying low-density regions. The generated complete probability map provides strong support for the subsequent generation of high-precision low-density region masking images. This, in turn, provides accurate data for calculating the collimator target position, enabling precise collimator adjustment, ultimately significantly improving medical image quality, reducing unnecessary radiation exposure for patients, and providing more valuable image data for medical diagnosis.

[0070] Specifically, during preprocessing segmentation, the spatial location of each image sub-block within the original medical image is known. Based on this spatial location information, a feature sub-block of the same size as the image sub-block is cropped from the corresponding location in the global feature map. For example, if an image sub-block is located in the upper left corner of the original medical image, then a feature sub-block of the same size is cropped from the corresponding upper left corner of the global feature map. The resulting cropped feature sub-block contains the relevant feature information of that image sub-block from a global perspective.

[0071] The cropped feature blocks are concatenated with their corresponding image blocks at the data level, typically along the channel dimension. Assuming the image block has C1 channels and the feature block has C2 channels, the resulting fused block will have C1 + C2 channels. This concatenation method allows the fused block to incorporate both local details from the image block and global information from the global feature map, providing rich data for subsequent local networks to more accurately analyze the features of that region.

[0072] The fused sub-blocks are input into a local network. This local network, based on a convolutional neural network architecture, is designed to deeply mine local features within the fused sub-blocks. Through a series of convolutions, pooling, and activation operations, the local network extracts and processes features from the fused sub-blocks, further refining and highlighting features related to low-density regions. After processing by the local network, a corresponding local probability map is output, where each pixel value represents the probability that the location belongs to a low-density region.

[0073] The local probability maps are stitched together according to the order of the image sub-blocks in the original medical image. For example, if the image sub-blocks are obtained sequentially from left to right and from top to bottom during preprocessing segmentation, the local probability maps are stitched together in the same order. In this way, all local probability maps are stitched together into a complete probability map, which reflects the probability distribution of each pixel belonging to a low-density region in the entire medical image.

[0074] In S2148, the complete probability map is post-processed to convert it into a binary low-density region mask image. First, a threshold is set, for example, 0.5. For each pixel in the probability map, if its probability value is greater than or equal to the threshold, it is classified as a low-density region, and the corresponding pixel value in the mask image is set to 1; otherwise, it is classified as a non-low-density region, and the pixel value is set to 0. After thresholding, some isolated small pixels or discontinuous regions may remain, possibly due to noise or segmentation errors. To remove this noise, morphological processing methods such as erosion and dilation operations are used. Erosion removes isolated small pixels, while dilation fills in small holes, making the segmented low-density regions more continuous and complete. The final low-density region mask image clearly identifies the location and shape of low-density regions in the medical image, providing accurate data for subsequent collimator target position calculations.

[0075] Specifically, in some embodiments, the specific implementation of step S2148 can be found in the following embodiments. This embodiment is based on... Figure 5 According to the detailed description of step S2148 in the collimator adjustment method shown in the corresponding embodiment, step S2148 in the collimator adjustment method may include the following steps: The complete probability map is binarized to obtain a binarized image.

[0076] The binarized image is post-processed to obtain a low-density region masking image.

[0077] In this embodiment, a suitable threshold, such as 0.5, is set to binarize the complete probability map. For each pixel in the complete probability map, if its probability value is greater than or equal to the set threshold, the pixel is assigned a value of 1, indicating that the location belongs to a low-density region; if the probability value is less than the threshold, the pixel is assigned a value of 0, indicating that the location does not belong to a low-density region. In this way, continuous probability values ​​are converted into discrete binary images, initially distinguishing between low-density and non-low-density regions. However, due to factors such as noise or segmentation errors, the binarized image may contain some isolated small pixels or discontinuous regions, requiring further processing.

[0078] Post-processing of binarized images typically employs morphological operations such as erosion and dilation. Erosion uses a specific structuring element (such as a square or circle) to slide across the image. If all pixels within the area covered by the structuring element have a value of 1, the central pixel remains 1; otherwise, it becomes 0. This operation can remove isolated small pixels and reduce the impact of noise. Dilation, on the other hand, is the opposite of erosion. If any pixel within the area covered by the structuring element has a value of 1, the central pixel becomes 1. Dilation can fill small holes, making low-density areas more continuous and complete. After post-processing operations such as erosion and dilation, a final low-density region mask is obtained. This mask clearly and accurately identifies the location and shape of low-density regions in medical images, providing precise data for subsequent collimator target location calculations.

[0079] It should be noted that the training methods for the aforementioned low-density region segmentation model specifically include: A medical image sample set was acquired, containing multiple medical image samples. Each sample was manually delineated and labeled by medical experts with corresponding low-density region annotations (also known as gold standard mask images) as labels. These low-density regions mainly include organ regions and background regions.

[0080] It should be noted that, in order to improve the robustness and generalization ability of the model, the above-mentioned medical image sample set will specifically include images of various complex situations that may occur in real clinical scenarios, such as changes in organ shape and position due to different patient body size or positioning, artifacts in the image, and image noise generated under different exposure conditions.

[0081] During training, a sample set containing multiple medical image samples can be obtained first, and then the multiple samples can be divided into a training set, a validation set, and a test set according to a predetermined ratio (e.g., 8:1:1).

[0082] Before training on the training set, the medical image samples in the training set are preprocessed. The preprocessing strictly follows the processing method during model inference, specifically including grayscale normalization, which linearly transforms the pixel values ​​of the image to the range of [0,1]; and resizing and block division, which scales the normalized image to a specific size (e.g., 512x512) to obtain the global image and divides it into multiple fixed-size (e.g., 128x128) image sub-blocks.

[0083] To further expand the training data and prevent model overfitting, data augmentation is performed only on image samples in the training set. Specifically, a series of random transformations are applied to the data images, including geometric and color transformations. Geometric transformations include, for example, horizontal or vertical flipping, small-angle rotation, slight scaling or shearing distortion, which simulate imaging morphology under different shooting angles and patient body shapes. Color transformations include, for example, adding Gaussian noise to simulate electronic noise, applying blurring to simulate slight motion artifacts, and adjusting image brightness and contrast.

[0084] In some embodiments, generative adversarial networks (GANs) can also be used to generate image samples that are difficult to identify in clinical settings but are not readily available in large quantities, such as images of anatomical structures that are adhered or partially obscured by other tissues or medical devices. This helps the algorithm model to more accurately identify low-density areas in various complex situations and reduces false and false identifications.

[0085] After obtaining the enhanced training set, the low-density region segmentation model is trained based on this training set, updating its internal network parameters and weights. The specific training process is as follows: medical image samples from the training set are input one by one into the low-density region segmentation model to obtain the predicted segmentation results of the low-density regions output by the model, i.e., the complete probability map.

[0086] The complete probability map output by the model is compared with the labeled segmentation result corresponding to the sample, and the difference between the two is calculated, which is the loss function. For segmentation tasks, commonly used loss functions can be cross-entropy loss or Dessian similarity coefficient loss. Then, an optimization algorithm (such as stochastic gradient descent SGD or Adam) is used to minimize the loss function, and the parameters and weights in the model are updated by the backpropagation algorithm. The above process is repeated until the value of the loss function converges or reaches the preset number of training epochs, marking the completion of one round of training.

[0087] After each training round, the performance of the low-density region segmentation model with updated parameters is evaluated based on the validation set. Based on the loss function value calculated on the validation set, hyperparameters such as the learning rate are automatically adjusted to achieve better results in the next training round. When the model's loss function on the validation set meets predetermined conditions, the model parameters and weights for that stage are retained. These predetermined conditions could include the loss function no longer decreasing over multiple rounds or reaching a minimum point.

[0088] It should be noted that, in this embodiment, the validation set does not need to be enhanced.

[0089] After all training and validation processes are completed, a final test is performed on all remaining candidate models based on the test set. The test set is an independent dataset that was not used in training and validation. By comparing the performance metrics of each model on the test set, including loss value and segmentation accuracy, the weights of the best-performing model are selected and output, and this model is determined as the final trained low-density region segmentation model.

[0090] Once the trained low-density region segmentation model is obtained, it can be deployed in practical applications to automatically, quickly, and accurately segment low-density regions in new medical images.

[0091] In S220, a series of precise geometric or statistical calculations are performed based on the low-density region masking image to determine the adjustment parameters of each baffle.

[0092] Specifically, in some embodiments, the specific implementation of step S220 can be found in [reference needed]. Figure 8 . Figure 8 It is based on Figure 3 According to the detailed description of step S220 in the collimator adjustment method shown in the corresponding embodiment, step S220 in the collimator adjustment method may include the following steps: S222, determine the image type of the low-density region masking image.

[0093] S224, if the image type is an organ masking image, then the target position of the collimator is determined by analyzing the geometric intersection relationship between the connected components and the image edges in the organ masking image.

[0094] S226, if the image type is a background mask image, the target position of the collimator is determined by accumulating and statistically analyzing the pixel values ​​of the mask image in the row and column directions.

[0095] In this embodiment, by distinguishing the types of low-density region masking images, different analysis methods are used for organ masking images and background masking images to determine the target position of the collimator, making the collimator adjustment more targeted and precise. This differentiated processing based on image type can fully utilize the information contained in the masking images to accurately determine the position of each baffle of the collimator, adapting to the imaging needs of different regions, further improving the quality of medical images, reducing excessive radiation to specific areas, and providing clearer and more accurate image data for medical diagnosis.

[0096] In S222, the type of the masked image is determined by feature analysis. For example, if the main area identified in the masked image has the shape features of an organ, such as the shape of a lung, it can be identified as an organ masked image; if the main area identified in the masked image is a background area that does not contain human tissue, it is identified as a background masked image.

[0097] The image type of the low-density region masking image can also be determined based on the inspection protocol. For example, if the collimator's wedge filter is movable while the others are not, the image type of the low-density region masking image is an organ masking image; if the finger filter and the opaque mask are movable, the image type of the low-density region masking image is a background masking image.

[0098] S224 represents the adjustment path for organs (such as the lungs). Because organs are irregularly shaped and typically require grayscale suppression using a semi-transparent wedge filter rather than complete occlusion, a geometric analysis approach is employed. The system analyzes the contour, location, and relationship to the image boundaries of the organ region to calculate the optimal coverage.

[0099] Specifically, in some embodiments, the specific implementation of step S224 can be found in the following embodiments. This embodiment is based on... Figure 8 According to the detailed description of step S224 in the collimator adjustment method shown in the corresponding embodiment, in which the collimator includes a wedge filter, step S224 may include the following steps: Determine the geometric intersection relationship between connected components and image edges in the organ masking image.

[0100] If the geometric intersection relationship is of the first type, the intersection points are connected to form a target straight line, and the position of the target straight line is adjusted according to the overlap ratio between the target straight line and the connected region to determine the target position of the wedge filter; in the first type of relationship, the connected region and two adjacent edges of the image have at least one intersection point.

[0101] If the geometric intersection relationship is a second type of relationship, then the connected component is virtually translated to the adjacent edge of the edge with the intersection point to make it a first type of relationship, and the target position of the wedge filter is determined by the first type of relationship; in the second type of relationship, the connected component has an intersection point with at most one edge of the image.

[0102] In this embodiment, please refer to the following: Figure 7By meticulously analyzing the geometric intersection relationships between connected components and image edges, and employing specific methods to determine the target position of the wedge filter for different relationships, the collimator can more accurately adjust the occlusion of organ regions. This precise adjustment helps optimize the imaging effect of organs in medical images, avoids overexposure of organ regions, improves image quality, provides doctors with clearer and more accurate diagnostic information, and reduces unnecessary radiation to patients.

[0103] Specifically, connected component analysis is performed on the organ masking image to identify the connected components of the organ regions in the image, and the intersection of each connected component with the image edges is determined. Here, the image edges refer to the four boundaries of the masked image.

[0104] A connection is considered a first-type relation when a connected component intersects with at least one of its two adjacent edges in the image. Figure 7 710. For example, in a lung masking image, if a connected component of the lung intersects with the left and top edges of the image, connecting these two intersections forms a target straight line. The overlap ratio between the triangle formed by the target straight line and the connected component is calculated. If the overlap ratio is less than a threshold T1 (e.g., set to 80%), the target straight line is moved in a direction perpendicular to it, moving one pixel at a time, until the overlap ratio exceeds the threshold T1. The determined position of the target straight line at this point is the target position of the wedge filter. This adjustment allows the wedge filter to more accurately mask organ regions, optimizing the imaging effect. If multiple connected components satisfying the first type of relationship exist in the image, the same operation is performed on each connected component.

[0105] When a connected component intersects with at most one edge of the image, it is classified as a second-type relation, such as... Figure 7 720 in the example. For instance, a connected component may have only one intersection with the right side of the image, or none at all edges. In such cases, as... Figure 7 As shown, the connected component is virtually translated towards the adjacent edge of the edge with intersection. Specifically, if it intersects with the right edge, it is translated towards the adjacent edge below or above; if it does not intersect with any edge, it can be translated in any adjacent edge direction. The maximum distance of the virtual translation is T2 (unit pixel, typically 1 / 5 of the image side length in the direction of movement). If, after translating T2 pixels, the connected component becomes a first-type relationship with at least one intersection with each of the two adjacent edges, the target position of the wedge filter is determined according to the first-type relationship method. If the first-type relationship requirement is still not met after translation (e.g., the intersection point does not meet the condition), no processing is performed. By using this virtual translation method, the second-type relationship is transformed into a first-type relationship for processing, further improving the accuracy of collimator adjustment and ensuring effective occlusion of organ regions.

[0106] S226 is the adjustment path for the background. The background area is characterized by a complete lack of useful information and needs to be masked as much as possible using opaque baffles or finger filters. Therefore, statistical analysis is a more efficient and robust approach. The system quickly locates large blank areas by calculating the sum of row and column pixels, thereby determining the occlusion boundaries.

[0107] Specifically, in some embodiments, the specific implementation of step S226 can be found in [reference needed]. Figure 8 . Figure 8 It is based on Figure 3 According to the detailed description of step S226 in the collimator adjustment method shown in the corresponding embodiment, in which the collimator includes an opaque shield and a finger-shaped filter, step S226 may include the following steps: The pixel accumulation values ​​for each row and each column of the masked image are calculated respectively to obtain the row accumulation vector and the column accumulation vector.

[0108] In the row accumulation vector and column accumulation vector, regions that are continuously background values ​​are detected.

[0109] If the start or end position of the region where the continuous background value coincides with the edge of the image, the target position of the opaque mask is determined based on the boundary of the region.

[0110] If the region with continuous background values ​​is located in the non-edge part of the image and its width is greater than a preset threshold, then the target position of the finger filter is determined based on the center position of the region.

[0111] In this embodiment, by accumulating and statistically analyzing the pixel values ​​in the row and column directions of the background mask image, and determining the target positions of the opaque masking plate and the finger-shaped filter based on different statistical results, the collimator can accurately mask the background area. This precise adjustment based on the characteristics of the background area helps reduce ineffective radiation to the background area, improves the quality of medical images, and avoids interference from the bright parts of the background area in medical diagnosis, providing doctors with clearer and more accurate image information.

[0112] Specifically, for background masking images, since they do not contain human information, this portion of the radiation dose is wasted and should be masked as much as possible using a collimator. The background area masking image Bmask is a binary image, where 0 represents the background and 1 represents the non-background area, so the background value is 0.

[0113] In some other embodiments, 1 can be used as the background and 0 as the non-background area, in which case the background value is 1.

[0114] Let the size of Bmask be M x N, where M is the number of rows and N is the number of columns. `sum(Bmask(i,:))` represents the summation of pixels in the i-th row of Bmask, and `sum(Bmask(:,j))` represents the summation of pixels in the j-th column of Bmask. By summing the pixel values ​​of each row, i.e., `Bmask_Msum = sum(Bmask(i,:))`, i = 1, 2, ..., M, we obtain the row-accumulated vector `Bmask_Msum`; by summing the pixel values ​​of each column, i.e., `Bmask_Nsum = sum(Bmask(:,j))`, j = 1, 2, ..., N, we obtain the column-accumulated vector `Bmask_Nsum`. Here, `Bmask(i,j)` represents the pixel value in the i-th row and j-th column of the background mask image.

[0115] In the obtained row-accumulated vector Bmask_Msum and column-accumulated vector Bmask_Nsum, find regions where the background value is 0 consecutively. For example, in the row-accumulated vector, find a continuous interval m1 to m2 where all element values ​​are 0, which determines a region where the background value is consecutively 0; similarly, perform a similar detection in the column-accumulated vector.

[0116] It should be noted that the specific method for determining the target location of the opaque barrier is as follows: If the region of continuous background values ​​starts at the first element of the vector, then the ending position of that region is the target position of the first blade of the opaque shield.

[0117] If the region of continuous background values ​​ends at the last element of the vector, then the starting position of that region is the target position of the second blade of the opaque shield.

[0118] In this embodiment, by accurately determining the relationship between the continuous background value region and the vector position, the target position of the opaque shield blades is clearly defined, further improving the accuracy of the collimator's adjustment of the background region occlusion. This precise adjustment can more effectively reduce the radiation of the background region and avoid interference from the background region on the main part of the medical image, thereby significantly improving the quality of the medical image and providing clearer and more reliable image data for medical diagnosis.

[0119] Taking the row accumulation vector Bmask_Msum as an example, if Bmask_Msum(1)=0 and all elements from m1=1 to m2 are 0, that is, the region with consecutive background values starts from the first element of the vector, which means there is a continuous background area at the top of the image. At this time, the target position of the upper blade (which can be regarded as the first blade) of the opaque baffle is m2. In actual adjustment, by controlling the upper blade of the opaque baffle to move to this position, the background area at the top of the image can be effectively blocked. Similarly, for the column accumulation vector Bmask_Nsum, if Bmask_Nsum(1)=0 and all elements from 1 to n2 are 0, the target position of the left blade (regarded as the first blade) of the opaque baffle is n2, so as to block the background area on the left side of the image.

[0120] Still taking the row accumulation vector Bmask_Msum as an example, if Bmask_Msum(M)=0 and all elements from m1 to M are 0, that is, the region with consecutive background values ends at the last element of the vector, which indicates that there is a continuous background area at the bottom of the image. At this time, the target position of the lower blade (which can be regarded as the second blade) of the opaque baffle is m1. In actual adjustment, controlling the lower blade of the opaque baffle to move to this position can effectively block the background area at the bottom of the image. Similarly, for the column accumulation vector Bmask_Nsum, if Bmask_Nsum(N)=0 and all elements from n1 to N are 0, the target position of the right blade (regarded as the second blade) of the opaque baffle is n1, thereby blocking the background area on the right side of the image.

[0121] In the row accumulation vector, if there is a region with consecutive background values, the starting position m1 of the region is greater than 1 and the ending position m2 is less than M, and the width m2-m1 of this region is greater than a preset threshold T3 (T3 is determined according to the width of the finger filter), then the target position of the center of the finger filter is (m1 + m2) / 2, and the angle is vertical. Similarly, in the column accumulation vector, if there is a similar region with consecutive background values, its starting position n1>1 and ending position n2<N, and the width n2-n1 is greater than the preset threshold T3, the target position of the center of the finger filter is (n1 + n2) / 2, and the angle is horizontal. In this way, for the background area located in the non-edge part of the image with a large width, by determining the target position of the finger filter, this area can be effectively blocked to avoid its interference to medical images.

[0122] In S300, after the target positions of each baffle of the collimator are calculated, the control system will generate corresponding control signals. These signals contain the specific position information that each baffle needs to move to, for example, the translation distance required for the opaque baffle, the rotation angle required for the translucent baffle, etc. The control signals are sent to the driving device of the collimator through the signal transmission line inside the equipment.

[0123] After receiving a control signal, the collimator's drive mechanism moves each baffle according to the signal instructions. The drive mechanism may employ various methods, such as motors or hydraulic systems, to achieve precise baffle movement. For example, a motor might drive a lead screw to move an opaque baffle along a track to the target position; or a motor might drive a rotating shaft to adjust the angle of a semi-transparent baffle until the calculated target position is reached. During adjustment, a feedback mechanism may also be included to monitor the baffle's position in real time, ensuring it accurately reaches the target position, thus completing the collimator adjustment and placing it in the optimal state for acquiring high-quality medical images.

[0124] The following describes an embodiment of the apparatus described in this application, which can be used to execute the collimator adjustment method in the above embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the collimator adjustment method described above.

[0125] Figure 9 A block diagram of a collimator adjustment device according to one embodiment of this application is shown.

[0126] Reference Figure 9 As shown, a collimator adjustment device 910 according to an embodiment of this application includes: a check protocol acquisition module 910, a target position determination module 920, and a target position adjustment module 930.

[0127] The examination protocol acquisition module 910 is used to acquire medical images and examination protocols; the target position determination module 920 is used to calculate the target position of the collimator based on the medical images and the examination protocol; and the target position adjustment module 930 is used to control the collimator to adjust according to the target position.

[0128] In some embodiments of this application, the target location determination module 920 specifically includes: a medical image conversion submodule, used to convert the medical image into a corresponding low-density region masking image based on the examination protocol; and a target location calculation submodule, used to calculate the target location of the collimator based on the low-density region masking image.

[0129] In some embodiments of this application, the medical image conversion submodule specifically includes: an image segmentation unit, used to preprocess and segment the medical image to obtain a global image and multiple corresponding image sub-blocks; and a region segmentation unit, used to input the global image and multiple corresponding image sub-blocks together into a corresponding low-density region segmentation model based on the inspection protocol to obtain a corresponding low-density region masking image.

[0130] In some embodiments of this application, the image segmentation unit specifically includes: a grayscale normalization subunit, used to normalize the grayscale of the medical image to obtain a normalized image; and a scaling segmentation subunit, used to scale and segment the normalized image to obtain a global image and multiple corresponding image sub-blocks.

[0131] In some embodiments of this application, the low-density region segmentation model includes a global network and a local network; The region segmentation unit specifically includes: a model invocation subunit, used to invoke the corresponding low-density region segmentation model based on the inspection protocol; a global network subunit, used to input the global image into the global network to obtain a global feature map with the same size as the global image; a local network subunit, used to sequentially connect each of the image sub-blocks to the global feature map and then input them into the local network to obtain a complete probability map; and a data post-processing subunit, used to perform data post-processing on the complete probability map to obtain the corresponding low-density region masking image.

[0132] In some embodiments of this application, the local network subunit is specifically used to perform the following: based on the spatial position of the image sub-block in the original image, cropping the corresponding feature sub-block from the global feature map; connecting the image sub-block and its corresponding feature sub-block at the data level to form a fused sub-block; inputting the fused sub-block into the local network to generate a corresponding local probability map; and stitching the local probability maps in sequence to obtain a complete probability map.

[0133] In some embodiments of this application, the data post-processing subunit is specifically used to perform: binarizing the complete probability map to obtain a binarized image; and post-processing the binarized image to obtain a low-density region masking map.

[0134] In some embodiments of this application, the target location calculation submodule specifically includes: an image type determination unit, used to determine the image type of the low-density region masking image; an organ masking image unit, used to determine the target location of the collimator by analyzing the geometric intersection relationship between connected components and image edges in the organ masking image if the image type is an organ masking image; and a background masking image unit, used to determine the target location of the collimator by accumulating and statistically analyzing the pixel values ​​in the row and column directions of the masking image if the image type is a background masking image.

[0135] In some embodiments of this application, the collimator includes a wedge filter, and the organ masking image unit specifically includes: a geometric intersection subunit, used to determine the geometric intersection relationship between connected components in the organ masking image and image edges; a first-type relationship execution subunit, used to connect the intersection points to form a target straight line if the geometric intersection relationship is a first-type relationship, and adjust the position of the target straight line according to the overlap ratio between the target straight line and the connected component to determine the target position of the wedge filter; in the first-type relationship, the connected component and two adjacent edges of the image each have at least one intersection point; a second-type relationship execution subunit, used to virtually translate the connected component to the adjacent edge of the edge with the intersection point if the geometric intersection relationship is a second-type relationship, so that it becomes a first-type relationship, and determine the target position of the wedge filter with the first-type relationship; in the second-type relationship, the connected component and at most one edge of the image have an intersection point.

[0136] In some embodiments of this application, the collimator includes an opaque mask and a finger-shaped filter; the background mask image unit specifically includes: an accumulation vector calculation subunit, used to calculate the pixel accumulation value of each row and each column of the mask image respectively, to obtain a row accumulation vector and a column accumulation vector; a continuous detection subunit, used to detect regions that are continuously background values ​​in the row accumulation vector and column accumulation vector; an edge coincidence subunit, used to determine the target position of the opaque mask based on the boundary of the region if the start or end position of the region that is continuously background values ​​coincides with the edge of the image; and a center position subunit, used to determine the target position of the finger-shaped filter based on the center position of the region if the region that is continuously background values ​​is located in the non-edge part of the image and its width is greater than a preset threshold.

[0137] In some embodiments of this application, the edge-overlapping sub-unit is specifically used to perform: if the region of continuous background values ​​starts at the first element of the vector, then the end position of the region is the target position of the first blade of the opaque mask; if the region of continuous background values ​​ends at the last element of the vector, then the start position of the region is the target position of the second blade of the opaque mask.

[0138] The embodiments of this application can automatically calculate the target position of the collimator based on medical images and the current collimator inspection protocol, and control the collimator to make corresponding adjustments. Compared with the traditional manual collimator adjustment method, this method greatly improves the accuracy and efficiency of adjustment. It can avoid the problem of poor imaging effect caused by the subjectivity and lag of manual adjustment, ensuring that the collimator is in the optimal position when acquiring medical images, thereby improving the quality of medical images, providing doctors with clearer and more accurate diagnostic basis, and also helping to reduce unnecessary radiation doses and protect the health of patients. Since the image features of low-density areas are relatively simple, the robustness and accuracy of the segmentation algorithm for them are much higher than those for directly segmenting complex ROIs. Therefore, this method can achieve stable, reliable and highly accurate automatic collimation effect.

[0139] Figure 10 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0140] It should be noted that, Figure 10 The computer system of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0141] like Figure 10 As shown, the computer system includes a Central Processing Unit (CPU) 1801, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 1802 or programs loaded from storage portion 1808 into Random Access Memory (RAM) 1803, such as performing the methods described in the above embodiments. Various programs and data required for system operation are also stored in RAM 1803. The CPU 1801, ROM 1802, and RAM 1803 are interconnected via bus 1804. An Input / Output (I / O) interface 1805 is also connected to bus 1804.

[0142] The following components are connected to I / O interface 1805: an input section 1806 including a keyboard, mouse, etc.; an output section 1807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1808 including a hard disk, etc.; and a communication section 1809 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1809 performs communication processing via a network such as the Internet. A drive 1810 is also connected to I / O interface 1805 as needed. Removable media 1811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1810 as needed so that computer programs read from them can be installed into storage section 1808 as needed.

[0143] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1809, and / or installed from removable medium 1811. When the computer program is executed by central processing unit (CPU) 1801, it performs various functions defined in the system of this application.

[0144] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0145] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0146] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0147] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.

[0148] This application also provides a computer program product storing at least one instruction, which is loaded and executed by the processor as described above. Figures 1-8 The method described in the illustrated embodiment can be found in the following document for a detailed execution process. Figures 1-8 The specific details of the illustrated embodiments will not be elaborated here.

[0149] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0150] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0151] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0152] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A collimator adjustment method, characterized in that, include: Obtain medical images and examination protocols; Based on the medical images and the examination protocol, calculate the target position of the collimator; The collimator is controlled to adjust according to the target position; Specifically, calculating the target position of the collimator based on the medical image and the examination protocol includes: Based on the aforementioned inspection protocol, the medical image is converted into a corresponding low-density region masking image. Based on the low-density region masking image, calculate the target position of the collimator; The calculation of the collimator's target position based on the low-density region masking image specifically includes: The image type of the low-density region masking image is determined, and the method of determining the image type includes at least one of determining it based on the inspection protocol and determining it through feature analysis of the masking image; If the image type is an organ masking image, the target position of the collimator is determined by analyzing the geometric intersection relationship between the connected components and the image edges in the organ masking image; If the image type is a background mask image, the target position of the collimator is determined by accumulating and statistically analyzing the pixel values ​​of the mask image in the row and column directions. Specifically, the determination through feature analysis of the masking image includes: if the main identified area in the masking image has the shape features of an organ, it is determined to be an organ masking image; if the main identified area in the masking image is a background area that does not contain human tissue, it is determined to be a background masking image. The determination based on the inspection protocol specifically includes: if the collimator's wedge filter is movable while the others are not, then the image type of the low-density area masking image is an organ masking image; if the finger filter and the opaque mask are movable, then the image type of the low-density area masking image is a background masking image.

2. The collimator adjustment method as described in claim 1, characterized in that, The process of converting the medical image into a corresponding low-density region masking image based on the examination protocol specifically includes: Medical images are preprocessed and segmented to obtain a global image and multiple corresponding image sub-blocks; Based on the inspection protocol, the global image and its corresponding multiple image sub-blocks are input together into the corresponding low-density region segmentation model to obtain the corresponding low-density region masking image.

3. The collimator adjustment method as described in claim 2, characterized in that, The preprocessing and segmentation of medical images to obtain a global image and multiple corresponding image sub-blocks specifically includes: Medical images are normalized in grayscale to obtain normalized images; The normalized image is scaled and segmented to obtain a global image and multiple corresponding image sub-blocks.

4. The collimator adjustment method as described in claim 2, characterized in that, The low-density region segmentation model includes a global network and a local network; Based on the inspection protocol, the global image and its corresponding multiple image sub-blocks are input together into the corresponding low-density region segmentation model to obtain the corresponding low-density region masking image, specifically including: Based on the aforementioned inspection protocol, the corresponding low-density region segmentation model is invoked. The global image is input into the global network to obtain a global feature map with the same size as the global image. Each of the image sub-blocks is sequentially connected to the global feature map and then input into the local network to obtain a complete probability map. The complete probability map is post-processed to obtain the corresponding low-density region masking image.

5. The collimator adjustment method as described in claim 4, characterized in that, The step of sequentially concatenating each of the image sub-blocks with the global feature map and then inputting them into the local network to obtain a complete probability map specifically includes: Based on the spatial position of the image sub-block in the original image, the corresponding feature sub-block is cropped from the global feature map; At the data level, the image sub-block and its corresponding feature sub-block are connected to form a fused sub-block; The fused sub-block is input into the local network to generate the corresponding local probability map; The complete probability map is obtained by stitching together the various local probability maps in sequence.

6. The collimator adjustment method as described in claim 4, characterized in that, The step of post-processing the complete probability map to obtain a low-density region masking map specifically includes: The complete probability map is binarized to obtain a binarized image; The binarized image is post-processed to obtain a low-density region masking image.

7. The collimator adjustment method as described in claim 1, characterized in that, The collimator includes a wedge filter. The determination of the collimator's target location by analyzing the geometric intersection relationship between connected components and image edges in the organ mask image specifically includes: Determine the geometric intersection relationship between connected components and image edges in the organ masking image; If the geometric intersection relationship is of the first type, the intersection points are connected to form a target straight line, and the position of the target straight line is adjusted according to the overlap ratio between the target straight line and the connected region to determine the target position of the wedge filter; in the first type of relationship, the connected region and two adjacent edges of the image have at least one intersection point; If the geometric intersection relationship is a second type of relationship, then the connected component is virtually translated to the adjacent edge of the edge with the intersection point to make it a first type of relationship, and the target position of the wedge filter is determined by the first type of relationship; in the second type of relationship, the connected component has an intersection point with at most one edge of the image.

8. The collimator adjustment method as described in claim 1, characterized in that, The collimator includes an opaque shield and a finger-shaped filter; The step of determining the target position of the collimator by accumulating and statistically analyzing the pixel values ​​in the row and column directions of the mask image specifically includes: Calculate the pixel accumulation value for each row and each column of the masked image to obtain the row accumulation vector and column accumulation vector; In the row accumulation vector and column accumulation vector, detect regions that are continuously background values; If the start or end position of the region where the continuous values ​​are background coincides with the edge of the image, then the target position of the opaque mask is determined based on the boundary of the region. If the region with continuous background values ​​is located in the non-edge part of the image and its width is greater than a preset threshold, then the target position of the finger filter is determined based on the center position of the region.

9. The collimator adjustment method as described in claim 8, characterized in that, If the start or end position of the region where the continuous values ​​are background coincides with the edge of the image, then the target position of the opaque mask is determined based on the boundary of the region, specifically including: If the region of continuous background values ​​starts at the first element of the vector, then the ending position of the region is the target position of the first blade of the opaque baffle. If the region of continuous background values ​​ends at the last element of the vector, then the starting position of that region is the target position of the second blade of the opaque shield.

10. A collimator adjustment device, characterized in that, The collimator adjustment device includes: The examination protocol acquisition module is used to acquire medical images and examination protocols; The target location determination module is used to calculate the target location of the collimator based on the medical image and the examination protocol; A target position adjustment module is used to control the collimator to adjust according to the target position; The target location determination module specifically includes: a medical image conversion submodule, used to convert the medical image into a corresponding low-density region masking image based on the examination protocol; and a target location calculation submodule, used to calculate the target location of the collimator based on the low-density region masking image. The target location calculation submodule specifically includes: an image type determination unit, used to determine the image type of the low-density region masking image, wherein the image type is determined by at least one of determination based on the inspection protocol and determination by feature analysis of the masking image; an organ masking image unit, used to determine the target location of the collimator by analyzing the geometric intersection relationship between connected components and image edges in the organ masking image if the image type is an organ masking image; and a background masking image unit, used to determine the target location of the collimator by accumulating and statistically analyzing the pixel values ​​in the row and column directions of the masking image if the image type is a background masking image. Specifically, the determination through feature analysis of the masking image includes: if the main identified area in the masking image has the shape features of an organ, it is determined to be an organ masking image; if the main identified area in the masking image is a background area that does not contain human tissue, it is determined to be a background masking image. The determination based on the inspection protocol specifically includes: if the collimator's wedge filter is movable while the others are not, then the image type of the low-density area masking image is an organ masking image; if the finger filter and the opaque mask are movable, then the image type of the low-density area masking image is a background masking image.

11. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the collimator adjustment method as described in any one of claims 1 to 9.

12. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the collimator adjustment method as described in any one of claims 1 to 9.

Images