Method and system for predicting self-adaptive radiotherapy dose of pelvic tumor based on cone beam CT (Computed Tomography)

Abstract

The invention provides a cone beam CT-based pelvic tumor adaptive radiotherapy dose prediction method and system, and is applied to the technical field of data processing. The method comprises the following steps: preprocessing a cone beam CT image of a target user to generate a target cone beam CT image; processing the target cone beam CT image and a preset CT image of the target user to generate preset target region sketching information; processing the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose influence factor; processing the initial image recognition processing model based on the training sample set to generate a target image recognition processing model; processing the target cone beam CT image and the preset target region sketching information based on a target image recognition processing model to generate a target CT image; and processing the target CT image, the dynamic dose influence factor and the physiological state information of the target user based on the target image recognition processing model to generate target dose prediction information.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for predicting adaptive radiotherapy dose of pelvic tumors based on cone beam CT. Background Art

[0002] In the specific field of medical intervention, especially in the treatment of pelvic region-related conditions, radiation intervention is an extremely critical method. However, due to the uniqueness of the internal structure of the pelvic region and the complexity of the surrounding sensitive structures, it has always been difficult to achieve high-precision application of this energy. In addition, during the relevant operation, the abnormal structure in a specific area may change its position and shape due to various factors (such as individual physiological activities, internal organ movement, and changes in the state of certain body fluids, etc.), which undoubtedly makes the entire operation process more complicated. Traditional related operation planning is often difficult to adapt to these dynamic changes, which may result in insufficient treatment of specific areas or unnecessary impact on surrounding normal structures.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] The purpose of this application is to provide a method and system for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT, which at least overcomes the problems existing in the prior art to a certain extent, obtains a target image by preprocessing the cone beam CT image, then processes the image and the preset CT image to generate target area delineation information, and simultaneously processes physiological motion parameters and structural information to obtain dynamic dose influencing factors. The model is trained using a training sample set, and a target CT image is generated based on the target model. Finally, the target CT image, dynamic dose influencing factors and physiological state information are integrated through the target model, and weighted fusion, constraint information acquisition and adjustment factor generation operations are performed to finally achieve the generation of target dose prediction information that helps to accurately perform specific tasks (related to pelvic tumor radiotherapy dose prediction), providing important support for related operations.

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

[0006] According to one aspect of the present application, a method for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT is provided, comprising: acquiring a cone beam CT image of a target user, a preset CT image of the target user, physiological motion parameter information of the target user, target physiological structure information of the target user, physiological state information of the target user, a preset image recognition processing model and a training sample set, wherein the preset CT image of the target user is used to characterize the image acquired when the target user performs dose scheme planning; preprocessing the cone beam CT image of the target user to generate a target cone beam CT image; processing the target cone beam CT image and the preset CT image of the target user to generate a preset target area delineation information. information, wherein the preset target area delineation information is used to characterize the position of the tumor and the target organ in the cone beam CT image; the physiological motion parameter information of the target user and the target physiological structure information of the target user are processed to generate a dynamic dose influence factor; the initial image recognition processing model is processed based on the training sample set to generate a target image recognition processing model; the target cone beam CT image and the preset target area delineation information are processed based on the target image recognition processing model to generate a target CT image; the target CT image, the dynamic dose influence factor and the physiological state information of the target user are processed based on the target image recognition processing model to generate target dose prediction information.

[0007] Another aspect of the present application is a prediction device for adaptive radiotherapy dose of pelvic tumors based on cone beam CT, characterized in that it includes: an acquisition module, used to acquire a cone beam CT image of a target user, a preset CT image of the target user, physiological motion parameter information of the target user, target physiological structure information of the target user, physiological state information of the target user, a preset image recognition processing model and a training sample set, wherein the preset CT image of the target user is used to characterize the image acquired when the target user performs dose scheme planning; a processing module, used to pre-process the cone beam CT image of the target user to generate a target cone beam CT image; and to process the target cone beam CT image and the preset CT image of the target user. Generate preset target area delineation information, wherein the preset target area delineation information is used to characterize the position of the tumor and the target organ in the cone beam CT image; process the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose influence factor; process the initial image recognition processing model based on the training sample set to generate a target image recognition processing model; process the target cone beam CT image and the preset target area delineation information based on the target image recognition processing model to generate a target CT image; process the target CT image, the dynamic dose influence factor and the physiological state information of the target user based on the target image recognition processing model to generate target dose prediction information.

[0008] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the method for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT is implemented.

[0009] The present application provides a method and system for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT. The target image is obtained by preprocessing the cone beam CT image, and then the image and the preset CT image are processed to generate the target area delineation information, and the physiological motion parameters and structural information are processed to obtain the dynamic dose influencing factors. The model is trained using the training sample set, and the target CT image is generated based on the target model. Finally, the target CT image, dynamic dose influencing factors and physiological state information are integrated through the target model. After weighted fusion, constraint information acquisition and adjustment factor generation, the target dose prediction information generation is finally realized, which is helpful for the precise execution of specific tasks (related to the prediction of pelvic tumor radiotherapy dose), and provides important support for related operations.

[0010] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1A flow chart showing a method for predicting adaptive radiotherapy dose of pelvic tumors based on cone beam CT provided in one embodiment of the present application;

[0012] Figure 2 A schematic diagram of the structure of a cone beam CT-based pelvic tumor adaptive radiotherapy dose prediction device provided in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0013] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0014] Combine the following Figure 1 The following describes a method for predicting the dose of adaptive radiotherapy for pelvic tumors based on cone beam CT according to an exemplary embodiment of the present application. In one embodiment, the present application also proposes a method and system for predicting the dose of adaptive radiotherapy for pelvic tumors based on cone beam CT. Figure 1 The flowchart of a method for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT according to an embodiment of the present application is schematically shown. Figure 1 As shown, the method is applied to a server, comprising:

[0015] S101, obtaining a cone beam CT image of a target user, a preset CT image of the target user, physiological motion parameter information of the target user, target physiological structure information of the target user, physiological state information of the target user, a preset image recognition processing model and a training sample set.

[0016] In one embodiment, assuming that the target user is a 55-year-old male user with a pelvic tumor, a special cone beam CT scanning device is used, and the user is scanned in a body position to obtain a series of cone beam CT images of the pelvic area. These images can clearly show the approximate position and morphology of the tumor in the pelvis and its relationship with the surrounding tissues, but there is a certain amount of noise and artifacts, and due to the influence of the user's physiological movements (such as breathing, intestinal peristalsis, etc.), the position of the tumor and organs in the image is slightly blurred or offset. In the early stage of program planning, the user performs a conventional CT scan, and the scanning range covers the entire pelvic area to obtain a preset CT image for accurate planning of dose distribution. The image has high resolution and stability, and provides basic anatomical structure information for subsequent planning. For example, the bone structure is clearly identifiable and can be used as an important reference for positioning and dose calculation.

[0017] By placing a respiratory monitoring sensor (such as a pressure sensor or an optical sensor) on the chest or abdomen of the user, information such as respiratory rate, respiratory amplitude, and respiratory waveform are recorded in real time. When the user is breathing calmly, the respiratory rate is measured to be about 15-20 times / minute, the respiratory amplitude fluctuates between 2-5 cm, and the respiratory waveform shows a typical periodic change. The user's ECG signal is monitored by an electrocardiogram (ECG) device to obtain information such as heart rate and ECG waveform. The user's heart rate is stable at 70-80 times / minute, and the ECG waveform shows a normal sinus rhythm without obvious abnormalities such as arrhythmia. Bone structure information is extracted from preset CT images or through a special bone imaging sequence (such as bone window CT). In the pelvic area, the bone contours of the pelvis are clearly visible, including the ilium, ischium, pubis, etc. These bone structures not only provide a reference for positioning, but also have an important impact on the scattering and absorption of rays. The gray value information of the CT image is converted into tissue density distribution data through a specific algorithm. In the tumor tissue area, the density is relatively high and uneven, and the surrounding normal tissues such as muscle and fat have different density characteristics. This information is crucial for accurately calculating the dose distribution of radiation in the tissue.

[0018] It is clear that the user is 55 years old and male. Age factors affect the repair capacity of tissues and tolerance to radiotherapy. The differences between men and women in physiological structure and hormone levels also affect the tumor. In addition to pelvic tumors, the user also suffers from mild hypertension, but the blood pressure is stable through medication. The nutritional status is good, the weight is moderate, and there is no obvious malnutrition or obesity. The overall physical condition has a certain impact on the risk and response of complications during radiotherapy. Adenocarcinoma was confirmed by pathological biopsy. This pathological type determines the biological behavior of the tumor and its sensitivity to radiotherapy, providing a basis for dose adjustment in subsequent plans. The degree of tumor differentiation is moderately differentiated, indicating that the morphology and function of tumor cells are somewhat different from those of normal tissue cells, and their growth rate and invasiveness are at a medium level, which affects the determination of radiotherapy doses and the choice of strategies. Genetic testing has found that tumor cells have specific gene expression abnormalities, such as changes in the expression of certain genes related to cell proliferation and apoptosis. This information helps to further understand the molecular biological characteristics of tumors and provide a reference for personalization and prediction of radiotherapy effects.

[0019] The preset image recognition processing model uses a deep convolutional neural network (CNN) architecture, such as the U-Net structure, which performs well in medical image segmentation and processing tasks because of its good feature extraction and context information fusion capabilities, and can accurately identify and locate different structures in the image. The encoder of the model consists of multiple convolutional layers and pooling layers alternating. The convolutional layer is used to extract local features of the image, and scans the image through convolution kernels of different sizes (such as 3x3, 5x5, etc.) to capture detailed features such as tumors and organ boundaries and texture information in the image. For example, for cone beam CT images of pelvic tumor users, the convolutional layer can detect subtle differences in grayscale values ​​and textures between tumor tissue and surrounding normal tissue. Pooling layers (such as maximum pooling or average pooling) are used to reduce the resolution of feature maps, reduce the amount of calculation, and retain the main feature information. After each set of convolutional layers, downsampling is performed through the pooling layer to gradually reduce the size of the feature map, realizing a multi-scale abstract representation of image features, which helps the model capture structural information at different levels and gradually understand the image content from the overall morphology to local details. The decoder part is symmetrical with the encoder and gradually restores the size of the feature map through upsampling layers (such as transposed convolution or bilinear interpolation upsampling) to make it the same size as the input image in order to generate the final prediction result.

[0020] During the upsampling process, the convolutional layer is combined with feature fusion to fuse the feature information of different levels in the encoder with the upsampled features, thereby enhancing the ability to restore image details and improving the accuracy of tumor and organ boundary delineation. For example, when processing pelvic CT images, the decoder can fuse feature information such as bone structure and tissue density from different depths of the encoder to accurately restore the anatomical structure of the pelvic region. In order to better fuse the information between the encoder and the decoder, the model uses jump connections. That is, each layer in the decoder is directly connected to the layer of the corresponding resolution in the encoder, and the feature information in the encoder is directly passed to the decoder, so that the model can make full use of features at different levels, avoid information loss caused by multiple convolution and pooling operations, and help to more accurately identify and locate tiny structures and fine boundaries in the image, such as the transition area between tumors and surrounding normal tissues.

[0021] The processing of cone beam CT images first optimizes the quality of the input cone beam CT images. Since cone beam CT images have problems such as noise and artifacts, the model learns the difference features between a large number of high-quality images and low-quality images, uses the filters in the convolution layer to suppress the noise, and repairs the image distortion caused by artifacts, improves the clarity and contrast of the image, and makes the tumor and organ structures in the image clearer and more recognizable, providing better basic data for subsequent recognition and segmentation tasks. Based on the optimized image, the convolutional neural network structure of the model can automatically identify the tumor area, key organs (such as bladder, rectum, femoral head, etc.) and their boundaries in the image. By learning the characteristic patterns of tumors and organs in a large number of training samples, the model can accurately distinguish different tissue structures based on the grayscale value, texture features, shape features and other information of the image, generate segmentation masks for tumor areas and organ areas, clarify the tissue category to which each pixel belongs, and provide accurate anatomical structure information for subsequent dose calculation and intervention plan formulation.

[0022] When the model needs to be processed in conjunction with a preset CT image, the high resolution and stability of the preset CT image are used to further accurately analyze the anatomical structure of the pelvic region. For example, by registering the cone-beam CT image with the preset CT image (registration method based on feature points or image grayscale information), the model can more accurately obtain detailed information about the bone structure, such as the precise position, shape, and joint relationship of each bone in the pelvis, as well as more accurate measurements of tissue density distribution. This information is critical for dose calculation and radiotherapy plan optimization, because the bone structure has a significant impact on the scattering and absorption characteristics of the radiation, and accurate measurement of tissue density helps to accurately calculate the dose distribution of the radiation in different tissues. The anatomical structure information in the preset CT image can be used as a reference to assist the model in more accurately locating tumors and key organs in the cone-beam CT image. Especially when the tumor boundary is unclear or the organ morphology changes due to physiological movement, the relatively stable anatomical structure benchmark provided by the preset CT image can help the model correct and optimize the recognition results, improve the accuracy of tumor and organ positioning, and thus more accurately formulate radiotherapy plans to ensure that the tumor is irradiated with a sufficient dose while minimizing damage to surrounding normal tissues.

[0023] The model design considers the fusion of physiological motion parameter information (such as respiratory motion parameters, heartbeat motion parameters, etc.), and these parameters can be combined with cone beam CT image data as additional input channels or feature vectors. For example, the input node corresponding to the physiological motion parameter is added to the input layer of the convolutional neural network, or the physiological motion parameter is fused with the image features through a specific feature fusion layer. In this way, the model can take into account the dynamic impact of physiological motion on the position of tumors and organs when recognizing and processing images, more accurately predict the dose distribution under different physiological motion states, achieve more dynamic and accurate radiotherapy plan optimization, and improve the adaptability and accuracy of intervention. For the target user's physiological state information (such as age, gender, physical condition, etc.) and tumor-specific characteristics (such as pathological type, degree of differentiation, gene expression characteristics, etc.), if the model has the corresponding processing capabilities, this information can be encoded and integrated into the decision-making process of the model. For example, by constructing a specific feature vector to represent this information and fusing it with the image features in the fully connected layer or decision layer of the model, the model can adjust the dose calculation and intervention plan optimization strategy according to the individual differences of the user and the characteristics of the tumor, and realize truly personalized radiotherapy. This integration approach helps improve the adaptability of the model to different users and tumor conditions, enhance intervention effectiveness, and reduce the risk of complications.

[0024] Image data of pelvic tumor users from multiple medical institutions were collected, including cone beam CT images, CT images, and corresponding clinical intervention information (such as actual radiotherapy dose distribution, intervention effect, etc.). The data covers users of different age groups, genders, pathological types, tumor stages, and intervention stages to ensure the diversity and representativeness of the samples. The image data were annotated in detail, including tumor boundaries, key organ contours, and category labels of tumors and normal tissues. At the same time, according to the intervention results and dose measurement data, the corresponding dose distribution information and intervention effect evaluation indicators (such as tumor control rate, normal tissue complication rate, etc.) were annotated for each sample for supervised learning and performance evaluation during model training and verification. A series of preprocessing operations were performed on the collected image data, such as image normalization, cropping, scaling, etc., so that all samples have a unified format and size for easy model input. At the same time, the annotated information was digitized and sorted to ensure that it accurately corresponds to the image data and provide high-quality training data for the model. Through such a training sample set, the preset image recognition processing model can learn the characteristic patterns and dose response relationships of pelvic tumors and their surrounding tissues in the images, thereby providing accurate tumor identification, dose prediction and intervention plan optimization services for target users in practical applications.

[0025] S102, preprocessing the cone beam CT image of the target user to generate a target cone beam CT image.

[0026] In one embodiment, the scanning parameters of the target cone beam CT scanning device are obtained. The target imaging device parameters are set as follows, and the tube voltage is set to 125kVp. This parameter is mainly used to ensure that the imaging rays have sufficient energy to penetrate the complex internal structures of a specific area of ​​the individual (such as the pelvic area), including bones, various soft tissues, and special structures (such as suspected abnormal tissues, etc.) when performing specific scanning tasks, while maintaining the contrast characteristics of the image as much as possible, providing a basic guarantee for subsequent image analysis. A tube current of 180mA is selected. Taking into account the image quality requirements and the individual radiation exposure risks, this current intensity can not only ensure that there is a relatively sufficient ray intensity during the imaging process, thereby obtaining a clearer image, but also effectively control the radiation dose received by the individual during the scan. In view of the individual's age and overall physical condition, this setting is intended to avoid unnecessary radiation risks caused by excessive tube current, while ensuring that the generated image has sufficient signal strength to meet the requirements of subsequent detailed analysis.

[0027] The scanning time is set at 12 seconds. This setting is to strike a balance between obtaining complete image information of a specific area of ​​an individual (such as the pelvis) and reducing imaging interference (artifacts) caused by individual physiological activities (such as breathing, slight peristalsis of internal organs, etc.). For this individual, a shorter scanning time can reduce the occurrence of motion artifacts caused by long scanning, while ensuring that the image quality obtained can meet the standards required for subsequent planning and evaluation within the tolerable scanning time range. The rotation angle range of the device is set to 360°, with the aim of fully covering the specific area of ​​the individual (pelvis) and collecting image information from all angles in order to construct a complete three-dimensional structural image of the area. Through such a full-range scan, the position, size and spatial relationship of specific internal structures (such as existing abnormal structures) in the area and surrounding tissues and organs (such as bladder, rectum, pelvic bones, etc.) can be accurately presented, providing a comprehensive structural information basis for subsequent related planning. The detector pixel size is 0.4mmx0.4mm. This relatively small pixel size can provide high spatial resolution, which helps to capture the detailed information of tiny structures in a specific area (pelvic cavity) in the image, such as the subtle changes in the edges of special structures, the distribution of small blood vessels, etc. Such detailed information is of great significance for accurately identifying the boundaries of specific structures, judging their characteristics, and formulating subsequent related plans.

[0028] The cone beam CT image of the target user is analyzed for noise characteristics, and the noise type information and noise level distribution information of the cone beam CT image are generated. The noise type information and noise level distribution information of the cone beam CT image are subjected to denoising to generate a denoised cone beam CT image. The original acquired specific imaging (such as cone beam CT) image is deeply analyzed. By observing the random fluctuation of the image gray value distribution, it is found that there are two main interference factors in the image, which are marked as type A and type B noise. Type A noise appears as irregularly distributed fine spots, especially in the area where the signal is relatively weak in the image (which can be compared to the low-dose area). This is mainly due to the statistical characteristics of the ray-related in the imaging process, which causes the signal to change randomly during detection. Type B noise appears as relatively uniform background interference, which is related to the performance of the electronic components and circuit characteristics of the detector. It has a relatively consistent interference level in the entire image area, but it will affect the overall clarity and detail display effect of the image.

[0029] In order to accurately grasp the distribution of interference factors (noise) in the image, the standard deviation of noise in different areas of the image is calculated. The results show that the noise level is relatively high in the edge areas of bone structures (such as ilium, ischium, etc.) near specific areas (such as the pelvic cavity) and special structures (such as suspected abnormal tissue). This is because when the imaging rays pass through the interface of these structures with different densities, complex scattering and absorption processes occur, resulting in increased signal fluctuations, making the noise more significant. In the relatively uniform soft tissue area within the region, the noise level is relatively low, but it cannot be ignored, because even a lower level of noise can mask some subtle structural information and affect the accurate judgment of the relationship between specific structures and surrounding tissues.

[0030] For type A noise, a specific wavelet threshold processing method is used. First, the image is decomposed by wavelet and converted into sub-band representations of different frequencies. In the wavelet domain, type A noise is mainly concentrated in the high-frequency sub-band. By setting a threshold based on statistical characteristics, high-frequency coefficients less than the threshold are regarded as noise and suppressed, while coefficients greater than the threshold are retained to preserve the edge and detail information of the image. Then wavelet reconstruction is performed to obtain a preliminary optimized image, which effectively reduces the impact of type A noise on image quality and makes the details in the image clearer and more discernible, especially the internal structure of special structures and the boundaries with surrounding tissues.

[0031] For type B noise, bilateral filtering technology is used. Bilateral filtering takes into account the similarity of pixel grayscale values ​​while considering the spatial distance of pixels. It can better preserve the edge information of the image while removing type B noise in the smoothed image. Appropriate filtering parameters are selected and adjusted according to the actual situation of noise in the image. Weighted average processing is performed on each pixel to minimize the blurring effect on image details and edges in the process of removing type B noise. After this series of processing, optimized specific imaging (such as cone beam CT) images are obtained, the image quality is significantly improved, and the contours of specific structures and surrounding tissues are clearer, laying a good foundation for further analysis and processing.

[0032] The denoised cone beam CT image is processed to generate metal artifact area information, and the metal artifact area information is processed based on a preset metal artifact correction algorithm to generate a first cone beam CT image, wherein the first cone beam CT image is used to characterize the cone beam CT image after metal artifact removal. In the specific imaging (such as cone beam CT) image after preliminary optimization, a specific area (such as the area affected by the existing metal implant or other high-density foreign body) is searched. Due to the existence of certain specific objects in individuals (such as metal fixtures implanted due to previous injuries), these objects appear as obvious high-density areas in the image, and the grayscale value is much higher than the surrounding tissue. By setting a higher grayscale threshold (such as 2500HU), the pixels in the specific area are initially screened, and then morphological operations are used (such as first performing an expansion operation to connect adjacent pixels belonging to the specific area, and then performing an erosion operation to remove the isolated small areas generated by the expansion) to accurately determine the range and shape of the specific area, and provide accurate target area information for subsequent correction. Based on a preset specific area correction algorithm (such as an iterative interpolation algorithm based on prior knowledge), the specific area is corrected using the individual's previous records or previously acquired related information (such as metal implant shape and position information) as prior knowledge. Through multiple iterative calculations, the algorithm gradually estimates the reasonable grayscale values ​​of pixels in the specific area based on the grayscale distribution characteristics of normal tissues around the specific area to fill the information missing area caused by the specific area, making the grayscale value transition of the image near the area more natural and smooth. After correction, a specific imaging (such as cone beam CT) image after the first stage of processing is obtained, the impact of the specific area is significantly improved, the structural integrity of the specific structure and surrounding tissue in the image is restored, and the risk of analysis errors caused by the specific area is reduced. More accurate image data is provided for subsequent image geometry correction, and the usability and analysis value of the entire image are improved.

[0033] The first cone beam CT image is processed based on the scanning parameters of the target cone beam CT scanning device to generate a second cone beam CT image, wherein the second cone beam CT image is used to characterize the cone beam CT image after the geometric distortion correction is completed, and the image quality processing is performed on the second cone beam CT image to generate the target cone beam CT image. According to the detailed scanning parameters of the acquired target imaging device (including the detector pixel size, the precise position of the rotation center, the distance from the ray source to the detector, etc.), the image geometric distortion model is constructed. Due to the existence of slight mechanical errors in the device during the scanning process (such as the non-uniformity of the detector rotation, the slight offset of the relative position between the ray source and the detector, etc.), the image is distorted in spatial geometry, which is manifested as straight line bending, circular structure deformation, etc. Using calibration data and precise geometric transformation formulas, the coordinate mapping relationship of each pixel in the ideal geometric position is calculated, and the specific imaging (such as cone beam CT) image after the first stage of processing is geometrically corrected. For example, for the bone structure in the image, correction is performed to restore it to its true shape and position, ensuring that the anatomical structure in the image accurately corresponds to the structure of a specific area of ​​the actual human body (such as the pelvis) in space, providing a reliable geometric basis for subsequent related planning.

[0034] After careful geometric correction, the specific imaging (such as cone beam CT) images after the second stage processing are obtained, and the geometric shapes and positional relationships of specific structures, organs, and bone structures in the images are more accurate and realistic. In the subsequent related planning process, accurate geometric information is crucial for accurately determining the relative positional relationship between specific structures and surrounding key tissues (such as bladder, rectum, etc.), which helps relevant personnel to accurately plan the operation range, angle, and parameter distribution to avoid unnecessary impact on normal tissues, while ensuring that specific structures can receive reasonable attention and treatment, thereby improving the accuracy and effectiveness of the overall planning. Adaptive contrast enhancement is performed on the specific imaging (such as cone beam CT) images after the second stage processing. This method automatically adjusts the enhancement strategy according to the grayscale distribution characteristics of different regions in the image, so that the contrast between the specific structure and the surrounding normal tissue is optimized. When the grayscale value changes little in the specific structure area, the grayscale range is appropriately stretched to highlight the difference between the specific structure and the surrounding tissue; in the tissue transition area, a smooth transition enhancement method is used to avoid unnatural grayscale jumps. This can make specific structures more eye-catching in the image, facilitate accurate identification of their boundaries and ranges, provide clear image basis for subsequent precise analysis and planning, and improve the accuracy and reliability of planning.

[0035] The intelligent edge sharpening algorithm based on multi-scale analysis is used to sharpen the boundaries of specific structures and organs in combination with the scale characteristics of different structures in the image. The algorithm detects the edge information of the image at different scales, and then performs weighted sharpening according to the importance and continuity of the edge, which can not only enhance the clarity of the boundaries of specific structures and organs, but also avoid noise amplification or edge artifacts caused by over-sharpening. In the formulation of relevant plans, clear and accurate boundaries help to accurately calculate the volume of specific structures and determine their spatial relationship with surrounding key tissues, thereby achieving more accurate parameter calculations and personalized planning schemes, improving the accuracy of planning, reducing the risk of mistreatment of surrounding normal tissues, effectively improving related results and reducing the occurrence of potential problems. After comprehensive image quality processing, the target imaging (such as cone beam CT) image is finally generated. The image has reached a high level in interference suppression, removal of specific area influences, geometric correction and image quality optimization, providing high-quality and reliable image data support for subsequent related planning, which helps to improve the accuracy and effectiveness of the entire process.

[0036] S103, processing the target cone beam CT image and the preset CT image of the target user to generate preset target area delineation information.

[0037] In one embodiment, feature extraction processing is performed on the target cone beam CT image and the preset CT image of the target user respectively to generate a number of anatomical feature points, wherein the anatomical feature points include bone landmarks and boundary points of the target organ. In the target cone beam CT image, the bone landmarks of the pelvis are first found. For example, the anterior superior iliac spine appears as an obvious high-density raised area in the image, and its grayscale value is significantly different from the surrounding tissue. By setting a suitable grayscale threshold and morphological operations (such as corrosion, expansion, etc.), the position coordinates of the anterior superior iliac spine can be accurately identified. Similarly, the ischial tuberosity also has a unique shape and grayscale feature, and its edge contour can be extracted using an edge detection algorithm, and then its center position is determined as a bone landmark. These bone landmarks have relatively stable positions and features in the pelvic cone beam CT images of different individuals, and can be used as an important reference for subsequent image registration.

[0038] For target organs (such as the bladder and rectum), the extraction of their boundary points needs to combine the physiological structure of the organs and the image features. When the bladder is filled, it appears as a low-density area that is approximately elliptical. Through image segmentation algorithms (such as methods based on region growing or threshold segmentation), the low-grayscale area inside the bladder is expanded to the surrounding area until a boundary with a large difference in grayscale value from the surrounding tissue is encountered, thereby determining the boundary points of the bladder. The boundary of the rectum is relatively complex. Since it is surrounded by adipose tissue, the texture analysis algorithm can be used to identify the transition area between fat and rectal wall and determine the boundary points of the rectum. The accurate extraction of these boundary points is crucial for accurately delineating the target area and evaluating the organ dose.

[0039] In the preset CT image, the above-mentioned bone landmark recognition method for cone beam CT images is also used to accurately find the key bone landmarks of the pelvis, such as the iliac crest and pubic symphysis. Since CT images have high resolution and contrast, and the bone structure is clearer, the extraction of these landmarks is relatively easier and more accurate. For example, the pubic symphysis appears as a clear linear high-density structure in the CT image, and its position can be accurately determined by simple edge detection and feature matching algorithms. For target organs such as the bladder and rectum, the high resolution advantage of the preset CT image is used to further refine the extraction of boundary points. The bladder boundary can be optimized based on its elliptical shape characteristics through a more accurate contour fitting algorithm, so that the boundary points more accurately reflect the true shape of the bladder. The rectal boundary can be combined with the tissue density information in the CT image, and the multi-threshold segmentation method can be used to better distinguish the rectal wall from the surrounding tissues, improve the accuracy of the boundary points, and provide more reliable basic data for the subsequent generation of accurate spatial transformation mapping tables.

[0040] A number of anatomical feature points are processed to generate a preset spatial transformation mapping table, wherein the preset spatial transformation mapping table is used to characterize the distance error between the minimized feature points of the target cone beam CT image and the preset CT image of the target user. The target cone beam CT image is image registered based on the preset spatial transformation mapping table to generate preset target area delineation information, which is used to characterize the position of the tumor and the target organ in the cone beam CT image. Specifically, the anatomical feature points (bone landmark points and target organ boundary points) extracted from the target cone beam CT image and the preset CT image are matched. For example, the anterior superior iliac spine in the cone beam CT image is paired with the corresponding point in the preset CT image, and the spatial distance error between them is calculated. For each matched feature point pair, the distance error is calculated by the coordinate difference, and these error values ​​are recorded in the preset spatial transformation mapping table. Assuming that in a certain case, after matching, it is found that the coordinates of a bone landmark point in the cone beam CT image are (x1, y1, z1), and the coordinates of the corresponding point in the preset CT image are (x2, y2, z2), then the distance error between them is By calculating the distance error of all matching feature point pairs, a complete preset space transformation mapping table is generated. This table reflects the difference in feature points between the two images and provides a key quantitative basis for subsequent image registration.

[0041] According to the distance error information in the preset spatial transformation mapping table, select the appropriate spatial transformation model (such as rigid body transformation, affine transformation or nonlinear transformation, etc.). If the distance error is small and the relative position relationship of the feature points is basically consistent, select the rigid body transformation model, whose parameters include translation vector and rotation matrix, and determine the values ​​of these parameters by minimizing the distance error between the feature points. If there is a certain degree of scaling or deformation between the images, it is necessary to use affine transformation or a more complex nonlinear transformation model, and solve the model parameters through the optimization algorithm so that the transformed cone beam CT image matches the preset CT image in the feature point position as much as possible, thereby achieving accurate image registration.

[0042] Based on the information in the preset spatial transformation mapping table, the target cone beam CT image is spatially transformed. For example, if a rigid body transformation model is used, each pixel in the cone beam CT image is translated and rotated accordingly according to the calculated translation vector and rotation matrix, so that the anatomical structure in the image (especially the tumor and target organ) is aligned with the preset CT image in spatial position. During the registration process, adjustments are continuously made based on the matching of feature points to ensure the accuracy of image registration. After registration, the positions of the tumor and target organ in the cone beam CT image and the corresponding structures in the preset CT image are optimally matched in space, providing an accurate image basis for subsequent target area delineation.

[0043] On the registered target cone beam CT image, using professional radiotherapy planning system software, experienced radiotherapists or physicists manually or in combination with automatic delineation algorithms outline the tumor and target organs according to clinical guidelines and oncology knowledge. For the tumor target area, the boundary range of the tumor is accurately determined based on its morphology, grayscale characteristics and relationship with surrounding tissues in the image, and the delineation information of the tumor target volume (GTV) is generated, including parameters such as the volume, shape, and position coordinates of the tumor. For target organs (such as the bladder and rectum), their contours are also accurately outlined, the volume of organs at risk (OAR) is determined, and their position information in the image is recorded. These preset target area delineation information will serve as an important basis for the formulation of radiotherapy plans, and will be used to calculate radiotherapy dose distribution, evaluate intervention effects, and conduct real-time monitoring and adjustments during the intervention process to ensure the accuracy and safety of radiotherapy, maximize tumor control rates, and reduce damage to normal tissues.

[0044] S104: Process the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose impact factor.

[0045] In one embodiment, the physiological motion parameter information of the target user is processed to generate characteristic parameter information of different breathing cycles. A high-precision breathing monitoring device (such as a breathing belt sensor or an optical plethysmograph) is used to perform real-time breathing monitoring on the target user. Within a period of time (such as 5 minutes), the breathing signal is continuously collected to obtain breathing waveform data. The waveform data reflects the change of the movement amplitude of the chest or abdomen during the breathing process of the target user over time. The collected breathing waveform data is analyzed to determine the breathing cycle. By setting a suitable threshold (such as the peak and valley values ​​of the breathing waveform), the breathing cycle is divided into an inspiratory phase, an expiratory phase, and an apnea phase (if any). The duration of each breathing phase is calculated. For example, in a certain breathing cycle, the duration of the inspiratory phase is 2 seconds, the duration of the expiratory phase is 3 seconds, and the duration of the apnea phase is 0.5 seconds (if any). At the same time, the characteristics of each breathing phase are analyzed, such as the rising slope of the inspiratory phase (reflecting the inspiratory speed), the falling slope of the expiratory phase (reflecting the expiratory speed), etc. These characteristic parameters help to gain a deeper understanding of the breathing pattern of the target user.

[0046] The target user's ECG signal is collected through an electrocardiogram (ECG) device to ensure that the electrode placement is accurate in order to obtain a high-quality ECG waveform. The collected ECG signal is filtered to remove noise and interference, improve the clarity and stability of the signal, and facilitate subsequent analysis. The heart rate, that is, the number of heartbeats per minute, is calculated by analyzing the ECG waveform cycle over a period of time (such as 1 minute). The fluctuation of the heart rate is observed to evaluate the stability of the heart rate. For example, among multiple heart rate data measured continuously, the standard deviation of the heart rate is calculated. If the standard deviation is small (such as less than 5 times / minute), it indicates that the heart rate is relatively stable; conversely, if the standard deviation is large, there is an irregular heartbeat. At the same time, key information is extracted from the ECG waveform, such as identifying the P wave (representing atrial depolarization), the QRS complex (representing ventricular depolarization) and the T wave (representing ventricular repolarization), and determining the cardiac systole (from the beginning of the QRS complex to the end of the T wave) and diastole (from the end of the T wave to the beginning of the next QRS complex). This information is crucial for understanding the physiological movement state of the heart.

[0047] The target physiological structure information of the target user is processed based on the preset time series information to generate dynamic change information of the structure under different physiological motion states. Advanced imaging technology (such as four-dimensional CT or real-time magnetic resonance imaging) is used to dynamically image the target physiological structure (such as tumors in the pelvic cavity, bladder, rectum, and surrounding bones, etc.) while the target user is performing physiological motion (breathing and heartbeat). A series of images are continuously acquired over a period of time (such as a respiratory cycle or multiple cardiac cycles) to ensure that the temporal resolution of the image is high enough to capture the subtle changes of the structure during physiological motion. The acquired dynamic image sequence is processed, and the position and morphological changes of the target physiological structure at different time points are tracked through image registration and analysis algorithms. For example, for tumor tissue, its displacement during respiratory motion is observed, and the displacement vector of the tumor center relative to the reference position (such as the initial position or average position) is calculated at different respiratory phases (end of inspiration, end of expiration, etc.). For organs such as the bladder and rectum, in addition to position changes, the dynamic changes of their shape and volume are also analyzed. Under the influence of cardiac movement, the tiny displacements and deformations of organs and tissues in the pelvic cavity are observed, such as the impact of blood flow and pressure changes caused by heart beating on surrounding tissues, to generate dynamic change information of the structure under different physiological motion states. This information includes the displacement trajectory, deformation degree, volume change curve, etc. of the structure, which intuitively reflects the dynamic characteristics of the target physiological structure during physiological motion.

[0048] The characteristic parameter information of different respiratory cycles, cardiac motion parameters and dynamic change information of structures under different physiological motion states are processed to generate dynamic dose influence factors, where the dynamic dose influence factors are used to characterize the difference in dose distribution between static state and dynamic motion state. Select a dose calculation model that is suitable for considering the influence of physiological motion, such as a dose calculation algorithm based on Monte Carlo simulation or an improved analytical dose calculation model. According to the individual characteristics of the target user (such as body size, tissue density distribution, etc.) and intervention equipment parameters (such as radiation energy, field shape, etc.), set the relevant parameters in the dose calculation model to ensure that the model can accurately simulate the propagation of radiation in different tissues and the dose deposition process.

[0049] The information of respiratory cycle characteristic parameters, cardiac motion parameters and dynamic changes of structures under different physiological motion states are input into the dose calculation model. Based on this information, the model simulates the irradiation of the target physiological structure by radiation in a static state (assuming that the physiological motion is zero) and a dynamic motion state, and calculates the dose distribution. For example, in a static state, the theoretical dose distribution of the tumor and surrounding normal tissue is calculated when there is no influence of physiological motion; in a dynamic motion state, the dose distribution is recalculated by considering the changes in the position and morphology of the structure caused by respiratory and cardiac motion. Then, by comparing the difference in dose distribution under the two states, a dynamic dose influence factor is generated. The dynamic dose influence factor can be expressed as the ratio of the dose at a point in a dynamic motion state to the dose at the same point in a static state under the same irradiation conditions, or presented in the form of a dose difference value (such as dynamic dose minus static dose). For example, for a specific point in the tumor, the calculated dose in a static state is 2Gy, and the dose in a dynamic motion state is 2.5Gy. The dynamic dose influence factor is 2.5Gy / 2Gy=1.25 (or the dose difference is 0.5Gy). This factor quantitatively characterizes the degree of influence of physiological movement on dose distribution, and provides an important basis for subsequent radiotherapy plan optimization and dose adjustment. It helps doctors to more accurately predict and control the dose distribution in the actual intervention process, improve the intervention effect of radiotherapy, and reduce damage to normal tissues.

[0050] S105, processing the initial image recognition processing model based on the training sample set to generate a target image recognition processing model.

[0051] In one embodiment, multiple data groups are extracted from the training sample set, wherein each data group contains a preset number of data samples, and at least one data sample includes target feature data. A large number of relevant imaging data and clinical information of pelvic tumor users are collected as a training sample set. The imaging data include cone beam CT images, preset CT images, and corresponding precisely annotated target area delineation information (annotated by experienced radiation oncologists to clarify the boundaries and ranges of tumors and key organs). At the same time, the user's physiological motion parameter information (such as respiratory cycle characteristic parameters, cardiac motion parameters, etc.), physiological state information (such as age, gender, physical condition, tumor-specific characteristics, etc.) and dose distribution data during the actual intervention process are also collected. These data are integrated and preprocessed to ensure that the data format is unified, the quality is reliable, and each sample contains sufficient information for model training.

[0052] Extract multiple data groups from the training sample set according to certain rules. For example, each data group contains 10 data samples. During the extraction process, ensure that at least one data sample includes target feature data, which can be specific image features related to pelvic tumors (such as tumor shape, size, grayscale pattern, etc.), physiological motion features (such as tumor displacement characteristics caused by respiratory motion, characteristics of the impact of cardiac motion on pelvic organs, etc.) or clinically relevant features (such as the correlation between the user's age, gender and tumor intervention response, etc.). For example, a data sample in a data group is a cone beam CT image of a 55-year-old male user with pelvic tumors. The tumor in the image presents an irregular shape and uneven grayscale (target feature data), and also contains other sample information such as the user's respiratory cycle characteristic parameters, age, gender, and corresponding target area delineation information and actual intervention dose distribution data.

[0053] Based on the preset clinical prior knowledge information and the preset anatomical structure restriction information, multiple data sets are processed to generate target data sets. The extracted data sets are screened and corrected using the preset clinical prior knowledge information. For example, based on the known growth patterns and common metastasis patterns of pelvic tumors, data samples that do not conform to the actual clinical situation (such as data where the tumor location is obviously inconsistent with anatomical common sense) are excluded. At the same time, for some data samples with special clinical characteristics (such as data where the user also suffers from other serious diseases that affect the effect of radiotherapy), special marking or processing is performed based on clinical experience so that the model can learn how to deal with these special situations during the training process. For example, if clinical prior knowledge indicates that users of certain age groups and gender combinations have low tolerance for radiotherapy, for data samples belonging to such situations, corresponding marking information is added to the data group to prompt the model to consider dose calculation and target area delineation more carefully when processing these samples.

[0054] The data set is further optimized based on the preset anatomical structure restriction information. Ensure that the image data in the data set conforms to the anatomical structure characteristics of the pelvic region, such as checking whether the relative position relationship between the tumor position and the surrounding key organs (bladder, rectum, femoral head, etc.) is reasonable. If it is found that the tumor position in a data sample does not match the anatomical structure (such as the tumor seems to pass through the anatomical structure that should not be passed under normal circumstances), the sample is corrected or excluded. In addition, based on the anatomical structure information, the target area delineation information is verified and adjusted to ensure that the range and shape of the target area are anatomically reasonable. For example, if the target area delineation in a data sample exceeds the normal anatomical range of the pelvis, it is corrected according to the anatomical structure restriction information to make it conform to the actual anatomical boundary. Through such processing, the target data set is generated, which is guaranteed in terms of clinical rationality and anatomical accuracy, providing a high-quality data foundation for subsequent model training.

[0055] The initial image recognition processing model is trained based on the target data set to generate a trained image recognition processing model. The initial image recognition processing model is trained using the target data set. The model uses a deep learning architecture (such as a convolutional neural network) with multiple hidden layers to learn complex features and patterns in the data. During the training process, the image data (such as cone beam CT images, preset CT images) in the target data set is used as the input of the model, and the corresponding target area delineation information and other related information (such as physiological motion parameters, physiological state information, etc.) are used as output labels or auxiliary information. The model calculates the predicted target area and dose distribution and other results through forward propagation, and then compares them with the real label information to calculate the loss function (such as the cross entropy loss function for classification tasks, the mean square error function for dose prediction tasks, etc.). According to the value of the loss function, the back propagation algorithm is used to adjust the parameters of the model (such as convolution kernel weights, fully connected layer weights, etc.), and through multiple iterative training, the model gradually learns the accurate relationship between the image data and the target area delineation and dose distribution, continuously optimizes the performance of the model, and generates a trained image recognition processing model.

[0056] The trained image recognition processing model is processed based on the validation set to generate test results. If the data sample containing the target feature data in the test result is a match between the predicted image area and the target label information, the trained image recognition processing model is used as the target image recognition processing model. Prepare an independent validation set. The data source and preprocessing method of the validation set are similar to those of the training sample set, but it does not participate in the model training process. Input the validation set data into the trained image recognition processing model to obtain the model's prediction results for the image data in the validation set, including the predicted target range, dose distribution, and features related to physiological movement. Then, compare these prediction results with the real label information in the validation set (such as accurate target delineation, actual measured dose distribution, etc.) to generate test results. The test results measure the performance of the model through a variety of evaluation indicators, such as the accuracy of target delineation (such as the Dice coefficient, which is used to measure the overlap between the predicted target and the actual target. The higher the Dice coefficient, the better the accuracy, and the ideal value is 1), the error indicators of dose prediction (such as the mean absolute error, which reflects the average difference between the predicted dose and the actual dose. The smaller the error, the better), and evaluation indicators related to physiological motion (such as the correlation between the predicted dose changes caused by physiological motion and the actual measured values).

[0057] After obtaining the test results, carefully analyze the test results of the data samples containing the target feature data. If the test results of these data samples indicate that the predicted image area matches the target area label information, that is, the predicted target area overlaps the real target area to a high degree (for example, the Dice coefficient reaches a certain threshold, such as above 0.8), the dose prediction error is within an acceptable range (such as the mean absolute error is less than a certain value, such as 0.2Gy), and the prediction characteristics related to physiological movement are consistent with the actual situation, then it is considered that the model performs well when processing samples with target feature data. At this time, the trained image recognition processing model is used as the target image recognition processing model for subsequent processing of pelvic tumor imaging data of new users to achieve accurate target area identification, dose prediction, and radiotherapy plan optimization. If the test results do not meet the requirements, it is necessary to further analyze the reasons, such as insufficient model training, problems with the data group, or the need to adjust the model structure, etc., and take corresponding improvement measures according to the specific situation (such as increasing the number of training iterations, optimizing the data group, improving the model architecture, etc.), retrain the model and verify it until a target image recognition processing model that meets clinical needs is obtained. Such a target model determination process ensures that the finally selected model has high accuracy and reliability, and can provide effective technical support for adaptive radiotherapy of pelvic tumors, improve the accuracy and intervention effect of radiotherapy, and reduce damage to the user's normal tissues.

[0058] S106, processing the target cone beam CT image and the preset target area delineation information based on the target image recognition processing model to generate a target CT image.

[0059] In one embodiment, the target cone beam CT image and the preset target area delineation information are processed based on the target image recognition processing model to generate initial recognition information of the target tumor area and initial recognition information of the target organ area. The target cone beam CT image and the preset target area delineation information are input into the model, and the model identifies the target area by analyzing the grayscale, texture and other features in the CT image. For example, in the cone beam CT image of the pelvis, the model detects an area with uneven local grayscale and a shape that is approximately spherical, which is the tumor area. At the same time, the model identifies a cystic structure in the front of the pelvis and determines it to be the bladder (target organ). Generate initial recognition information of the target tumor area, such as the tumor is roughly located on the right side of the pelvis and has a diameter of about 4 cm; generate initial recognition information of the target organ area (bladder), such as the bladder is located in the front of the pelvis and is approximately 5 cm long, 3 cm wide and 3 cm high.

[0060] Obtain the equipment parameter information of the target radiation instrument, where the equipment parameter information includes the target radiation energy, target field shape and target dose rate. Assuming that a linear accelerator is used, the target radiation energy: for example, 10MV (megavolts). Radiation of this energy can effectively penetrate the pelvic tissue to reach the tumor site. Target field shape: According to the location of the tumor and surrounding organs, the field shape is an irregular polygon. The field shape is adjusted by a multi-leaf collimator so that it can cover the tumor area as much as possible while avoiding important organs such as the bladder and rectum. Target dose rate: set to 200MU / min (monitoring unit / minute). This dose rate can control the intervention time within a reasonable range while ensuring the intervention effect.

[0061] Based on the initial identification information of the target tumor region and the initial identification information of the target organ region, the target cone beam CT image is segmented to generate several tissue region information, wherein the tissue region information includes the tumor tissue region, the normal tissue region and the target organ tissue region. According to the initial identification information of the tumor region and the organ region obtained above, the target cone beam CT image is segmented. For example, a method based on threshold and region growing is adopted. Taking the initial identification position of the tumor region as the seed point, the tumor region is continuously expanded by comparing the gray value of the pixel until a boundary with a large difference in gray value is encountered to determine the tumor tissue region. For the normal tissue region and the target organ tissue region (such as bladder, rectum, etc.), a similar method is also used to combine their anatomical features for segmentation. Tissue region information marked with different colors is obtained, red represents the tumor tissue region, green represents the normal tissue region (such as pelvic muscle, fat, etc.), and blue represents the bladder tissue region.

[0062] Process several tissue region information to generate preset tissue density values, where different tissue region information corresponds to different preset tissue density values. Tumor tissue density value: Usually, tumor tissue density is slightly higher than normal soft tissue. Assume that the preset tissue density value of the tumor tissue region is 1.1 g / cm 3 Normal tissue density value: The density of pelvic muscle tissue is relatively high, assuming 1.05g / cm 3 ; The density of pelvic fat tissue is low, assumed to be 0.9 g / cm 3 Target organ density value: The density of urine in the bladder is low, and the density of the bladder tissue area (excluding the urine part) is assumed to be 1.02g / cm 3. The equipment parameter information and preset tissue density value of the target radiation instrument are processed to generate a target CT image, wherein the target CT image is used to characterize the preset dose value and preset dose distribution information of different tissue regions. By performing similar calculations on each point in the image, a target CT image is obtained, wherein different grayscales or colors represent the preset dose value and dose distribution information of different tissue regions. For example, a high-dose area (near the center of the tumor) is displayed as white in the image, and a low-dose area (such as normal tissue away from the tumor) is displayed as black or dark.

[0063] In another embodiment, the scheme includes a mathematical calculation formula for obtaining a preset tissue density value, and the mathematical calculation formula is: P tissue =P base ×(1+k1×f1(x,y,z)+k2×f2(G)+k3×f3(M)+k4×f4(P)); where P tissue Represents the preset tissue density value, P base represents the basic tissue density reference value, k1, k2, k3, k4 represent weight coefficients, which are used to adjust the influence of different factors on the tissue density value; f1(x, y, z) represents the function related to the spatial position of the target tissue in the image, f2(G) represents the function related to the grayscale feature G of the target tissue, f3(M) represents the function related to the metabolic state M of the user, and f4(P) represents the function related to the physiological feature P of the user.

[0064] Assuming the reference value of basic tissue density P tissue =1.0g / cm 3, which is a basic tissue density benchmark value for subsequent calculations. Weight coefficients k1 = 0.1, k2 = 0.2, k3 = 0.3, k4 = 0.4. These coefficients are used to adjust the degree of influence of different factors on tissue density values. Spatial position correlation function f1 (x, y, z): Assume that the coordinates of the tumor center in the image are (x = 3, y = 4, z = 5), and f1 (3, 4, 5) = 0.5. This function value reflects the influence of the spatial position of the target tissue (tumor) in the image on tissue density. Grayscale feature correlation function f2 (G): Assume that the grayscale feature G of the target tissue (tumor) is 120 (grayscale value), and f2 (120) = 0.3. This function value reflects the influence of the grayscale feature of the target tissue on tissue density. Metabolic state correlation function f3 (M): Assume that the user's individual metabolic state M = 0.8 (can be some quantitative metabolic indicator), and f3 (0.8) = 0.4. This function value reflects the impact of the user's individual metabolic state on tissue density. Physiological characteristic related function f4(P): Assume that the user's physiological characteristic P = 0.6 (which can be a quantitative indicator related to physiology), and f4(0.6) = 0.2. This function value reflects the impact of the user's physiological characteristics on tissue density.

[0065] According to the formula P tissue =P base ×(1+k1×f1(x,y,z)+k2×f2(G)+k3×f3(M)+k4×f4(P)) to calculate: P tissue =1.0×(1+0.1×0.5+0.2×0.3+0.3×0.4+0.4×0.2)=1.31g / cm 3 In this example of pelvic tumor radiotherapy, the preset tissue density value calculated by the above formula and assumed parameters is 1.31g / cm 3 This value can be used for subsequent radiotherapy dose calculations and other operations. For example, when calculating the dose distribution in the target CT image, different tissue density values ​​will affect the dose deposition of the radiation in the tissue.

[0066] S107 , processing the target CT image, the dynamic dose influencing factor and the physiological state information of the target user based on the target image recognition processing model to generate target dose prediction information.

[0067] In one embodiment, a target CT image and a dynamic dose influence factor are weighted and fused based on a target image recognition processing model to generate target feature information of different tissue regions, wherein the target feature information of different tissue regions is used to characterize the dose response values ​​of different tissue regions under the influence of physiological motion. The target CT image is a CT image of the pelvis, which contains different tissue regions such as tumor tissue, bladder, rectum, surrounding muscles and bones. These tissues present different grayscale values ​​on the CT image and can be used to distinguish different tissue types. Consider the influence of physiological motion (such as breathing and heartbeat) on dose distribution. For example, for a tumor in the pelvis, the tumor will move up and down during breathing, and its dynamic dose influence factor can be obtained by previous measurement and calculation. At the end of inspiration, the tumor moves up 2mm, and the dynamic dose influence factor of this position is 1.2 (meaning that the dose will increase by 20% compared to the static state); at the end of exhalation, the tumor returns to the initial position, and the dynamic dose influence factor is 1.0. The model weightedly fuses the tissue information in the target CT image with the corresponding dynamic dose influence factor. For tumor tissue, it is assumed that the grayscale value of the CT image at a certain position corresponds to a tissue density value of 1.1 g / cm 3 , combined with the dynamic dose influence factor of 1.2 at the end of inspiration, after weighted calculation of the model (the specific weight is determined by model training), the target characteristic information (dose response value) of the position under the influence of physiological movement is obtained. For example, the dose response value of the position is calculated to be 2.2Gy (Gray).

[0068] Obtain dynamic dose constraint information for different tissue regions. Dynamic dose constraint information for different tissue regions: According to clinical intervention requirements, tumor tissue needs to receive a sufficiently high radiation dose to achieve the intervention effect. For example, the minimum dose constraint for tumor tissue is set at 60Gy. Endangered organs (such as bladder and rectum): In order to protect normal tissue, the bladder and rectum have upper dose constraints. For example, the maximum dose constraint for the bladder is 40Gy, and the maximum dose constraint for the rectum is 35Gy. These constraint information is based on clinical experience and research, and is intended to minimize damage to normal tissue while killing tumor cells.

[0069] The target user's physiological status information is processed to generate a dose adjustment factor. The target user's physiological status information includes age, gender, body mass index (BMI), basal metabolic rate, etc. People of different ages and genders have different physical functions and tolerance to intervention. For example, the body recovery ability of the elderly is weaker than that of the young, and there are also differences between men and women in physiological structure and metabolism. BMI reflects the degree of fatness or thinness of a person. A BMI of 25 indicates that the target user is overweight. Different BMI values ​​affect the absorption and scattering characteristics of body tissues to radiation, which in turn affects the distribution of radiotherapy doses. The basal metabolic rate represents the energy metabolism rate of the human body when it is awake and extremely quiet, and is not affected by muscle activity, ambient temperature, food, and mental stress. The basal metabolic rate is 1300 kcal / day. This value can reflect the target user's physical energy consumption and is also related to the body's physiological functions.

[0070] Based on the above physiological status information, a specific algorithm is used to calculate the dose adjustment factor. For example, let age be A (unit: years) and the basic dose adjustment factor be F. age .

[0071] The formula is: F age =1+0.005×(A-50) (here 50 years old is a reference age and the coefficient 0.005 is the assumed age impact weight).

[0072] For a 60-year-old man, F age =1+0.005×(60-50)=1.05.

[0073] The gender dose adjustment factor F is assumed to be male. gender The gender dose adjustment factor for women is 1.02 (a value based on empirical assumptions, indicating that men require 2% more dose adjustment than the baseline in this case), and the gender dose adjustment factor for women is 0.98 (indicating that women require 2% less dose adjustment than the baseline).

[0074] The following is an example of the BMI-related algorithm. Let BMI be B and the dose adjustment factor be F. BMI .

[0075] The formula is: F BMI =1+0.01×(B-22) (here 22 is a reference BMI value and the coefficient 0.01 is the assumed BMI impact weight).

[0076] For BMI = 25, F BMI =1+0.01×(25-22)=1+0.03=1.03.

[0077] Assume that the basal metabolic rate is M (unit: kcal / day), and the dose adjustment factor is F metabolism .

[0078] The formula is: F metabolism =1+(M-1200) / 10000 (here 1200 kcal / day is a reference basal metabolic rate, and the denominator 10000 is to make the impact moderate).

[0079] For M = 1300 kcal / day, F metabolism =1+(1300-1200) / 10000=1.01.

[0080] Then, these factors are combined to obtain the total dose adjustment factor F: F = F age ×F gender ×F BMI ×F metabolism =1.05×1.02×1.03×1.01=1.11. In this example, the calculated dose adjustment factor is 1.11. The role of this factor is: when predicting radiotherapy doses, one cannot rely solely on conventional calculation methods, but also needs to take into account the individual physiological state of the target user. For example, without considering these physiological states, it is calculated that the radiotherapy dose required for a certain tissue area is 100Gy, but due to factors such as the age, gender, BMI and basal metabolic rate of the target user, the actual dose required should be 100Gy×1.1=110Gy. In this way, the radiotherapy dose can be adjusted more accurately to achieve better intervention effects while minimizing damage to normal tissues.

[0081] The target feature information of different tissue regions, the dynamic dose constraint information of different tissue regions, and the dose adjustment factors are processed to generate target dose prediction information. The model integrates this information to generate target dose prediction information. For tumor tissue, after considering the dynamic dose effect, dose constraint, and dose adjustment factor, it is predicted that the final radiation dose required is 66.6Gy (60Gy×1.11), and it is ensured that during the intervention, the dose received by the surrounding normal tissues (such as the bladder and rectum) does not exceed its maximum dose constraint value. Through such comprehensive processing, the radiotherapy plan can more accurately determine the radiation dose of each tissue region, improve the intervention effect and reduce side effects.

[0082] The server preprocesses the cone beam CT image, including obtaining scanning parameters, analyzing and processing noise, reducing noise, determining and correcting metal artifact areas, correcting geometric distortion, and processing image quality to obtain the target cone beam CT image. Then the target cone beam CT image and the preset CT image are processed to generate preset target area delineation information, which involves extracting anatomical feature points, generating a spatial transformation mapping table, and performing image registration. Then, the physiological motion parameters and physiological structure information are processed to obtain dynamic dose influencing factors, including analyzing respiratory and cardiac motion parameters, as well as the dynamic changes of physiological structures under different motion states.

[0083] The training sample set is used to train the initial image recognition processing model to obtain the target model. The process includes data group extraction, screening based on clinical and anatomical knowledge, model training and verification, and processing the target cone beam CT image and preset target area delineation information based on the target model to generate the target CT image. Finally, the target model is used to process the target CT image, dynamic dose influencing factors and physiological state information to obtain the target dose prediction information, including weighted fusion, obtaining dose constraint information and generating dose adjustment factors, which is helpful for accurate pelvic tumor radiotherapy dose prediction.

[0084] In one embodiment, if Figure 2 As shown, the present application also provides a prediction device for adaptive radiotherapy dose of pelvic tumors based on cone beam CT, comprising:

[0085] An acquisition module 201 is used to acquire a cone beam CT image of a target user, a preset CT image of a target user, physiological motion parameter information of a target user, target physiological structure information of a target user, physiological state information of a target user, a preset image recognition processing model and a training sample set, wherein the preset CT image of a target user is used to characterize an image acquired when the target user performs a dose scheme planning;

[0086] The processing module 202 is used to pre-process the cone beam CT image of the target user to generate a target cone beam CT image; process the target cone beam CT image and the preset CT image of the target user to generate preset target area delineation information, wherein the preset target area delineation information is used to characterize the position of the tumor and the target organ in the cone beam CT image; process the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose influence factor; process the initial image recognition processing model based on the training sample set to generate a target image recognition processing model; process the target cone beam CT image and the preset target area delineation information based on the target image recognition processing model to generate a target CT image; process the target CT image, the dynamic dose influence factor and the physiological state information of the target user based on the target image recognition processing model to generate a target dose prediction information.

[0087] Each embodiment in the present application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the prediction method, electronic device, electronic device, and readable storage medium for evaluating the adaptive radiotherapy dose of pelvic tumors based on cone beam CT, since they are basically similar to the above-mentioned prediction method for adaptive radiotherapy dose of pelvic tumors based on cone beam CT, the description is relatively simple, and the relevant parts can be referred to the partial description of the above-mentioned prediction method for adaptive radiotherapy dose of pelvic tumors based on cone beam CT.

Claims

1. A method for predicting adaptive radiotherapy dose of pelvic tumors based on cone beam CT, characterized in that: include: Acquire a cone beam CT image of a target user, a preset CT image of the target user, physiological motion parameter information of the target user, target physiological structure information of the target user, physiological state information of the target user, a preset image recognition processing model, and a training sample set, wherein the preset CT image of the target user is used to characterize an image acquired when the target user performs dose scheme planning; Preprocessing the cone beam CT image of the target user to generate a target cone beam CT image; Processing the target cone beam CT image and the preset CT image of the target user to generate preset target area delineation information, wherein the preset target area delineation information is used to characterize the position of the tumor and the target organ in the cone beam CT image; Processing the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose impact factor; Processing the initial image recognition processing model based on the training sample set to generate a target image recognition processing model; Processing the target cone beam CT image and the preset target area delineation information based on the target image recognition processing model to generate a target CT image; The target CT image, the dynamic dose influencing factor and the physiological state information of the target user are processed based on the target image recognition processing model to generate target dose prediction information.

2. The method according to claim 1, characterized in that Preprocessing the cone beam CT image of the target user to generate a target cone beam CT image includes: Obtaining scanning parameters of a target cone beam CT scanning device; Performing noise characteristic analysis on the cone beam CT image of the target user to generate noise type information of the cone beam CT image and noise level distribution information of the cone beam CT image; Performing noise reduction processing on the noise type information of the cone beam CT image and the noise level distribution information of the cone beam CT image to generate a noise-reduced cone beam CT image; Processing the denoised cone beam CT image to generate metal artifact area information; Processing the metal artifact region information based on a preset metal artifact correction algorithm to generate a first cone beam CT image, wherein the first cone beam CT image is used to represent the cone beam CT image after metal artifact removal; Processing the first cone-beam CT image based on the scanning parameters of the target cone-beam CT scanning device to generate a second cone-beam CT image, wherein the second cone-beam CT image is used to represent the cone-beam CT image after geometric distortion correction is completed; Image quality processing is performed on the second cone-beam CT image to generate a target cone-beam CT image.

3. The method according to claim 1, characterized in that Processing the target cone beam CT image and the preset CT image of the target user to generate preset target area delineation information includes: Performing feature extraction processing on the target cone beam CT image and the preset CT image of the target user respectively to generate a plurality of anatomical feature points, wherein the anatomical feature points include bone landmark points and boundary points of target organs; Processing a plurality of anatomical feature points to generate a preset spatial transformation mapping table, wherein the preset spatial transformation mapping table is used to characterize a distance error between minimized feature points of a target cone beam CT image and a preset CT image of a target user; The target cone beam CT image is subjected to image registration processing based on the preset spatial transformation mapping table to generate preset target area delineation information.

4. The method according to claim 3, characterized in that Processing the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose impact factor includes: Processing the physiological motion parameter information of the target user to generate characteristic parameter information and cardiac motion parameters of different respiratory cycles; Processing the target physiological structure information of the target user based on the preset time series information to generate dynamic change information of the structure under different physiological motion states; The characteristic parameter information of the different respiratory cycles, the cardiac motion parameters and the dynamic change information of the structure under different physiological motion states are processed to generate a dynamic dose impact factor, wherein the dynamic dose impact factor is used to characterize the difference value of the dose distribution under the static state and the dynamic motion state.

5. The method according to claim 1, characterized in that Processing the initial image recognition processing model based on the training sample set to generate a target image recognition processing model includes: Extracting multiple data groups from the training sample set, wherein each data group includes a preset number of data samples, and at least one data sample includes target feature data; Processing the plurality of data sets based on the preset clinical prior knowledge information and the preset anatomical structure restriction information to generate a target data set; Training the initial image recognition processing model based on the target data group to generate a trained image recognition processing model; Processing the trained image recognition processing model based on the verification set to generate a test result; If the data sample containing the target feature data in the test result represents that the predicted image area matches the target area label information, the trained image recognition processing model is used as the target image recognition processing model.

6. The method according to claim 4, characterized in that The target cone beam CT image and the preset target area delineation information are processed based on the target image recognition processing model to generate a target CT image, including: Processing the target cone beam CT image and the preset target area delineation information based on the target image recognition processing model to generate initial recognition information of the target tumor area and initial recognition information of the target organ area; Acquiring equipment parameter information of a target radiation instrument, wherein the equipment parameter information includes target radiation energy, target radiation field shape, and target dose rate; Performing image segmentation processing on the target cone beam CT image based on the initial identification information of the target tumor region and the initial identification information of the target organ region to generate a plurality of tissue region information, wherein the tissue region information includes a tumor tissue region, a normal tissue region and a target organ tissue region; Processing a plurality of tissue region information to generate a preset tissue density value, wherein different tissue region information corresponds to different preset tissue density values; Processing the equipment parameter information of the target radiation instrument and the preset tissue density value to generate a target CT image, wherein the target CT image is used to characterize the preset dose values ​​and preset dose distribution information of different tissue regions; The method includes a mathematical calculation formula for obtaining a preset tissue density value, and the mathematical calculation formula is: P tissue =P base ×(1+k1×f1(x,y,z)+k2×f2(G)+k3×f3(M)+k4× f4(P)); Among them, P tissue Represents the preset tissue density value, P base represents the basic tissue density reference value, k1, k2, k3, k4 represent weight coefficients, which are used to adjust the influence of different factors on the tissue density value; f1(x, y, z) represents the function related to the spatial position of the target tissue in the image, f2(G) represents the function related to the grayscale feature G of the target tissue, f3(M) represents the function related to the metabolic state M of the user, and f4(P) represents the function related to the physiological feature P of the user.

7. The method according to claim 6, characterized in that The target CT image, the dynamic dose influencing factor and the physiological state information of the target user are processed based on the target image recognition processing model to generate target dose prediction information, including: Based on the target image recognition processing model, the target CT image and the dynamic dose influence factor are subjected to weighted fusion processing to generate target feature information of different tissue regions, wherein the target feature information of tissues in different regions is used to characterize dose response values ​​of different tissue regions under the influence of physiological motion; Obtain dynamic dose constraint information for different tissue regions; Processing the physiological status information of the target user to generate a dosage adjustment factor; The target feature information of different tissue regions, the dynamic dose constraint information of different tissue regions and the dose adjustment factor are processed to generate target dose prediction information.

8. A prediction device for adaptive radiotherapy dose of pelvic tumors based on cone beam CT, characterized in that: The device comprises: an acquisition module, used to acquire a cone beam CT image of a target user, a preset CT image of a target user, physiological motion parameter information of a target user, target physiological structure information of a target user, physiological state information of a target user, a preset image recognition processing model and a training sample set, wherein the preset CT image of a target user is used to characterize an image acquired when the target user performs a dose scheme planning; A processing module is used to pre-process the cone beam CT image of the target user to generate a target cone beam CT image; process the target cone beam CT image and the preset CT image of the target user to generate preset target area delineation information, wherein the preset target area delineation information is used to characterize the position of the tumor and the target organ in the cone beam CT image; process the physiological motion parameter information of the target user and the target physiological structure information of the target user to generate a dynamic dose influence factor; process the initial image recognition processing model based on the training sample set to generate a target image recognition processing model; process the target cone beam CT image and the preset target area delineation information based on the target image recognition processing model to generate a target CT image; process the target CT image, the dynamic dose influence factor and the physiological state information of the target user based on the target image recognition processing model to generate target dose prediction information.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; The first processor is configured to execute the method for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT as described in any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the method for predicting the adaptive radiotherapy dose of pelvic tumors based on cone beam CT as described in any one of claims 1 to 7 is implemented.

Images