Intelligent digital image processing system for urinary surgery

Abstract

The invention provides an intelligent digital image processing system for urinary surgery. The intelligent digital image processing system is suitable for accurate diagnosis and dynamic monitoring of diseases such as kidney tumors and bladder tumors. The system comprises a data acquisition module, a multi-modal image fusion module, an image segmentation module, a dynamic lesion tracking module, a three-dimensional modeling module, an intelligent diagnosis support module and a federal learning module. The data acquisition module receives CT, MRI and other image data; the multi-modal image fusion module generates a high-quality comprehensive image through a registration algorithm; the image segmentation module extracts a lesion area by using deep learning; the dynamic lesion tracking module calculates lesion volume, surface area and change trend thereof; the three-dimensional modeling module generates a three-dimensional visual model of the lesion area; and the intelligent diagnosis support module outputs a diagnosis report containing lesion features, development trends and treatment suggestions. According to the invention, through multi-modal fusion and dynamic monitoring, the accuracy of lesion area extraction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital image information processing, and in particular to an intelligent digital image processing system for urology. Background Art

[0002] Urological diseases (such as kidney tumors, bladder tumors, prostate cancer, etc.) are major diseases with a high incidence worldwide. Their diagnosis and treatment rely on the precise analysis of medical images. Currently, commonly used imaging examination methods in clinical practice include computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, etc. These technologies provide a basis for the discovery and evaluation of lesions. However, traditional imaging diagnostic methods have significant deficiencies in the following aspects:

[0003] Multimodal imaging (such as CT and MRI) has its own advantages in terms of anatomical information and tissue function information, but their independent use is often difficult to fully describe the characteristics of the lesion:

[0004] CT images provide high-resolution anatomical information, but have poor ability to identify soft tissue and lesion properties;

[0005] MRI images have advantages in soft tissue contrast and lesion functional imaging, but the resolution is relatively low. Traditional imaging diagnostic methods usually rely on manual comparative analysis of multimodal images, which is not only time-consuming and easily affected by human factors, but also has low fusion efficiency and accuracy.

[0006] During the treatment of urological diseases (such as kidney tumors and bladder tumors), dynamic monitoring of the lesion area is required to determine the development trend of the lesion and the treatment effect. Existing methods mainly rely on:

[0007] Static lesion feature analysis: such as volume and shape, but cannot capture its dynamic change characteristics;

[0008] Manual annotation: Doctors need to manually measure volume and shape changes, which is cumbersome and lacks standardization. This not only increases the diagnostic burden, but also may affect the accurate assessment of lesion change trends.

[0009] Medical imaging data is scattered across different hospitals or medical institutions. Due to privacy regulations and technical limitations, cross-institutional sharing of raw data faces the following challenges:

[0010] Data silo problem: Each hospital accumulates data independently, making it difficult to integrate and train high-performance models;

[0011] Privacy protection issues: The sensitivity of patient data leads to security risks in cross-institutional data transmission.

[0012] Existing technologies usually rely on centralized data sharing or model training methods in multi-institutional collaboration, which can easily trigger privacy risks and cause waste of computing resources.

[0013] Although artificial intelligence technology has made some progress in the field of medical imaging, existing systems still face the following problems:

[0014] The segmentation accuracy is not stable enough: Especially in complex lesion areas, the existing segmentation algorithms are difficult to achieve high accuracy and high robustness;

[0015] Low module integration: Existing image processing systems usually develop modules such as segmentation and 3D modeling independently, lacking integrated process support;

[0016] Lack of clinical explainability: Most AI systems output a single indicator or prediction result, failing to provide doctors with detailed and intuitive diagnostic basis.

[0017] The diagnosis and treatment of urological diseases requires processing a large amount of complex imaging data. Traditional imaging analysis methods usually have the following problems:

[0018] Large manual workload: Doctors need to check the image data frame by frame and mark the lesion area;

[0019] Diagnosis takes a long time: From image acquisition to diagnostic report generation, the process is lengthy and difficult to meet the clinical needs of rapid diagnosis.

[0020] Therefore, we proposed an intelligent digital image processing system for urology to solve the above problems. Summary of the invention

[0021] The purpose of the present invention is to solve the technical problems raised in the above background technology, and to provide an intelligent digital image processing system for urology.

[0022] The above-mentioned purpose of the present invention is achieved like this:

[0023] On the one hand, the present invention provides an intelligent digital image processing system for urology, the system comprising:

[0024] Data acquisition module: receiving multimodal imaging data from medical imaging equipment ,in: represents the pixel value of the three-dimensional image at time t; are the spatial coordinates of the image, indicating the positions in the horizontal, vertical and depth directions respectively; t is the time point, indicating the time of image acquisition;

[0025] Data preprocessing module: input image data Perform noise removal, image enhancement and standardization to generate preprocessed images , calculated according to the following formula:

[0026] ,

[0027] in:

[0028] : Standardized image pixel value;

[0029] : The pixel mean in the local window is calculated as ;

[0030] : The standard deviation of pixels within the local window is calculated as ;

[0031] : The number of pixels in the local window;

[0032] :Window The intensity value of each pixel;

[0033] Multimodal image fusion module: realizes the registration of multimodal images through the following objective function to generate aligned comprehensive image data :

[0034] ,

[0035] in:

[0036] : Fusion error, the smaller the error, the more accurate the registration;

[0037] : pixel value of CT image after preprocessing;

[0038] : pixel values ​​of the preprocessed MRI image, adjusted by the calibrated offset;

[0039] : The spatial offset between CT and MRI images, aligned in horizontal, vertical and depth directions respectively;

[0040] : The number of pixels in the local window;

[0041] Image segmentation module: Extract lesion area through convolutional neural network (CNN) , calculated according to the following formula:

[0042] ,

[0043] in:

[0044] : Prediction results of lesion area;

[0045] : Activation function, usually ReLU or sigmoid activation;

[0046] : Convolution kernel weight matrix, used to extract image features;

[0047] *: convolution operation; : Comprehensive image data after multi-modal fusion alignment;

[0048] : Bias term, used to adjust the linear offset of feature extraction results;

[0049] Dynamic lesion tracking module: according to the lesion area output by the image segmentation module , calculate the volume of the lesion area And predict the future volume by the following formula :

[0050] ,

[0051] in:

[0052] : Lesion volume at time t;

[0053] :time The lesion area;

[0054] : The volume of each voxel;

[0055] : Adjustment coefficient, used to correct the predicted volume change;

[0056] : Lesion volume at the previous time point;

[0057] 3D modeling module: using the lesion area provided by the dynamic lesion tracking module , generate a three-dimensional mesh model of the lesion area, and the interpolation formula is:

[0058] ,

[0059] in:

[0060] : The voxel value of the interpolation point;

[0061] : Isosurface threshold, used to define the boundary of the lesion area;

[0062] and are the voxel gray values ​​at both ends of the interpolation point;

[0063] : Interpolation weight, which indicates the position ratio of the interpolation point between the two end voxels;

[0064] Intelligent diagnosis support module: Combines the results of the above modules to generate an imaging diagnosis report, including lesion area analysis, dynamic trend prediction and treatment recommendations.

[0065] On the other hand, the image segmentation module optimizes the parameters of the model through the following loss function and :

[0066] ,

[0067] in: : Loss function, used to measure the error between the segmentation result and the true label; : No. The true label of each pixel (0 or 1); : No. The predicted probability of a pixel is given by calculate; : The number of pixels in the local window.

[0068] On the other hand, the dynamic lesion tracking module calculates the lesion volume growth rate by the following formula:

[0069] ,

[0070] in:

[0071] GrowthRate: The growth rate of the lesion volume, which indicates the ratio of the current volume to the volume at the previous moment;

[0072] :time The volume of the lesion; ;

[0073] :time The lesion area is defined by the output of the image segmentation module ;

[0074] : the volume of a single voxel; ;

[0075] : The spatial offset between CT and MRI images;

[0076] :time The lesion volume is calculated in the same way as .

[0077] On the other hand, the multimodal image fusion module introduces a weighted optimization objective function:

[0078] ,

[0079] in:

[0080] : Fusion error, used to evaluate the registration accuracy between CT and MRI images;

[0081] : Pixel weighting factor, indicating pixel weights in the registration process;

[0082] : pixel value of CT image after preprocessing;

[0083] : Pixel values ​​of MRI images after preprocessing, adjusted by registration offset;

[0084] : The spatial offset between CT and MRI images, used for registration;

[0085] : The number of pixels in the local window.

[0086] On the other hand, the 3D modeling module generates the optimal surgical path in combination with preoperative planning :

[0087] ,

[0088] in:

[0089] : Optimal path, indicating the optimal navigation path in surgical planning;

[0090] : Path set, representing all possible paths;

[0091] : The number of nodes in the path;

[0092] :node The weight of a path to measure its importance in path optimization;

[0093] :Path Node The loss value of risk or cost.

[0094] On the other hand, the imaging diagnosis report generated by the system includes:

[0095] lesion size, location, and volume;

[0096] Prediction of lesion development trend.

[0097] On the other hand, the federated learning module realizes multi-center data sharing through the following formula:

[0098] ,

[0099] in:

[0100] : Global model parameters, used for parameter optimization in the image segmentation module, in the After iterations, the weighted average of the local model parameters is obtained, including and ;

[0101] : No. The local model parameters of each data center are obtained from the local training results of the image segmentation module;

[0102] : No. The number of samples in each data center;

[0103] : The number of pixels in the local window.

[0104] On the other hand, the data preprocessing module uses a filtering algorithm to remove image noise, and the calculation formula is:

[0105] ,

[0106] in: : Pixel value of the image after filtering;

[0107] : The original pixel value within the filter window;

[0108] : The total number of pixels within the filter window; : The first The spatial coordinates of the pixels.

[0109] On the other hand, the dynamic lesion tracking module combines the triangular patch data output by the 3D modeling module to calculate the surface area of ​​the lesion area. :

[0110] ,

[0111] in:

[0112] :time Surface area of ​​the lesion;

[0113] Z: the number of triangular patches, indicating how many triangles the surface of the lesion area consists of;

[0114] : No. The area of ​​a triangular patch is calculated from the mesh data output by the 3D modeling module;

[0115] ;

[0116] ;

[0117] : The cross product of two edge vectors;

[0118] : The modulus of the cross product, which means the area of ​​the triangle is twice;

[0119] The lesion shape change rate is further calculated:

[0120] ,

[0121] Among them: ShapeRate: lesion shape change rate, indicating the change ratio of the current surface area relative to the surface area at the previous moment; , The time and of the lesion surface area.

[0122] On the other hand, the system supports remote diagnosis function, and imaging diagnosis reports can be transmitted to other medical institutions via the network for remote consultation.

[0123] Compared with the prior art, the present invention has the following beneficial effects:

[0124] The present invention combines the high-resolution anatomical information of CT images with the soft tissue contrast advantage of MRI images through multimodal image fusion technology to generate high-quality comprehensive image data. Combined with image segmentation and three-dimensional modeling modules, the system can accurately extract the lesion area and calculate its volume, surface area and its change trend (such as volume growth rate, shape change rate). This full process realizes a seamless connection from static features to dynamic monitoring, significantly improving the diagnostic accuracy of urological diseases (such as kidney tumors, bladder tumors, etc.), and providing a comprehensive and reliable reference for clinical practice.

[0125] Through modular design, the system integrates data acquisition, preprocessing, multimodal fusion, dynamic lesion tracking, 3D modeling and diagnostic report generation, greatly simplifying the clinical image analysis process. Doctors do not need to manually mark or measure lesion characteristics. The system automatically outputs lesion volume, surface area and trend analysis data, and generates a 3D intuitive display, which improves diagnostic efficiency and significantly reduces the workload of doctors, meeting the clinical needs of rapid diagnosis.

[0126] This invention solves the privacy protection problem in medical data sharing by introducing a federated learning module, and realizes distributed collaborative optimization across institutions. Hospitals only need to share model parameters without transmitting original image data, ensuring patient privacy. In addition, federated learning optimizes model performance by integrating multi-institutional data, significantly improving the stability and accuracy of image segmentation, and providing technical support for the widespread deployment and continuous improvement of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0128] Figure 1 The system block diagram of an intelligent digital image processing system for urology. DETAILED DESCRIPTION

[0129] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0130] The following is combined with Figure 1 The specific embodiments of the present invention are described in detail with reference to the accompanying drawings and examples.

[0131] The present invention provides an intelligent digital image processing system for urology, which aims to realize the automatic processing and accurate diagnosis of patient medical image data. The functions of the system are completed by the collaboration of multiple modules, specifically including the following steps:

[0132] 1. Data acquisition module: First, the multimodal imaging data of the patient's urinary system is collected through medical imaging equipment (such as CT, MRI, etc.). represents, where: It is the image in time Time, location The pixel value of Represents three-dimensional coordinates in horizontal, vertical and depth directions; Indicates the time point of image acquisition and is used to analyze the dynamic changes of the image. These data provide basic input for subsequent processing.

[0133] 2. Data preprocessing module: In order to improve the quality and consistency of the image, the system processes the collected image data Preprocessing is performed, including noise removal, image enhancement, and standardization. The standardization formula is as follows:

[0134] ,

[0135] in: Represents the standardized image pixel value; is the number of pixels in the local window.

[0136] is the mean value of pixels in the local window;

[0137] is the standard deviation of pixels within the local window; through standardization, the image data is converted into a zero-mean, unit-variance form, thereby eliminating the impact of brightness and contrast differences on subsequent analysis.

[0138] 3. Multimodal image fusion module: For multimodal images (such as CT and MRI images), the system uses a fusion algorithm to align them spatially and generate comprehensive image data The goal of fusion is to optimize the following objective function

[0139] ,

[0140] in: Represents the fusion error, the smaller it is, the more accurate the registration is; : pixel value of CT image after preprocessing; : pixel values ​​of the preprocessed MRI image, adjusted by the offset of the registration; Represents the spatial offset between CT and MRI images, which are aligned in the horizontal, vertical and depth directions respectively; : The number of pixels in the local window. Through iterative optimization , the system can generate aligned fused images , thereby providing high-quality input for subsequent analysis.

[0141] 4. Image segmentation module: after fusion of image data The system uses a convolutional neural network (CNN) to extract the lesion area. The segmentation process is described by the following formula:

[0142] ,

[0143] in: represents the prediction result of the lesion area; is the activation function (such as ReLU or sigmoid); is the weight matrix of the convolution kernel; is the bias term. * is the convolution operation. Segmentation result It is a binary image, where the area with a value of 1 represents the diseased tissue and the area with a value of 0 represents the background.

[0144] 5. Dynamic Lesion Tracking Module: Segmentation Results is the input of the dynamic lesion tracking module. The system calculates the lesion volume by the following formula :

[0145] ,

[0146] in: Indicates the lesion area; is the volume of a single voxel, which depends on the resolution of the image. In order to monitor the development trend of the lesion, the system further calculates the volume growth rate:

[0147] ,

[0148] in: and are the lesion volumes at the current and previous moments respectively; GrowthRate represents the dynamic change of the volume.

[0149] 6. 3D modeling module: using the lesion area provided by the dynamic lesion tracking module , generate a three-dimensional mesh model of the lesion area, and the interpolation formula is: ; in: : The voxel value of the interpolation point; : Isosurface threshold, used to define the boundary of the lesion area; and are the voxel gray values ​​at both ends of the interpolation point; : Interpolation weight, which indicates the position ratio of the interpolation point between the two end voxels; 7. The 3D modeling module combines preoperative planning to generate the optimal surgical path :

[0150] ,

[0151] in:

[0152] : Optimal path, which indicates the optimal navigation path in surgical planning;

[0153] : Path set, representing all possible paths;

[0154] : The number of nodes in the path;

[0155] :node The weight of a path to measure its importance in path optimization;

[0156] :Path Node The loss value of risks or costs;

[0157] 8. The dynamic lesion tracking module combines the triangular patch data output by the 3D modeling module to calculate the surface area of ​​the lesion area :

[0158] ,

[0159] in:

[0160] :time Surface area of ​​the lesion;

[0161] : The number of triangular patches, indicating how many triangles the surface of the lesion area consists of;

[0162] : No. The area of ​​a triangular patch is calculated from the mesh data output by the 3D modeling module;

[0163] ;

[0164] ;

[0165] : The cross product of two edge vectors;

[0166] : The modulus of the cross product, which means the area of ​​the triangle is twice.

[0167] The system combines the calculation results of volume and surface area to analyze the shape change rate ShapeRate of the lesion:

[0168] ,

[0169] in: and are the lesion surface areas at the current and previous moments respectively; ShapeRate represents the dynamic change of the surface area.

[0170] Finally, the system generates a diagnosis report, including the volume and surface area of ​​the lesion, as well as treatment recommendations. The system supports remote diagnosis, and the imaging diagnosis report can be transmitted to other medical institutions via the network for remote consultation.

[0171] The system also includes a federated learning module to achieve multi-center data sharing through the following formula:

[0172] ,

[0173] in:

[0174] : Global model parameters, used for parameter optimization in the image segmentation module, in the After iterations, the weighted average of the local model parameters is obtained, including and ;

[0175] : No. The local model parameters of each data center are obtained from the local training results of the image segmentation module;

[0176] : No. The number of samples in each data center;

[0177] : The number of pixels in the local window.

[0178] The following is a detailed description with reference to several embodiments.

[0179] Example 1: Analysis of lesion of kidney tumor;

[0180] 1. Data collection and preprocessing;

[0181] Data collection: The patient's kidney images were collected by CT and MRI, with resolutions of and ,The CT slice spacing is 1.5 mm, and the MRI slice spacing is 2.0 mm.

[0182] Standardization processing: the collected image data For standardization, the formula is as follows:

[0183] ,

[0184] Among them, the mean of the local window , standard deviation , pixel value ,but:

[0185] ,

[0186] 2. Multimodal image fusion;

[0187] Fusion of CT and MRI images through registration formula:

[0188] ,

[0189] Weight , assuming the spatial offset , after optimization, the fusion error 0.01, generate comprehensive images .

[0190] 3. Image segmentation;

[0191] The lesion area is extracted by convolutional neural network (CNN), and the segmentation formula is:

[0192] ,

[0193] Convolution kernel weights , Bias And the activation function After optimization, the lesion area is obtained , which contains the number of voxels .

[0194] 4. Calculation of lesion volume;

[0195] The volume of the lesion area is calculated using the dynamic lesion tracking module:

[0196] ,

[0197] The voxel size is mm, single pixel volume:

[0198] ,

[0199] Lesion volume:

[0200] ,

[0201] 5. Surface area calculation;

[0202] Generate a triangular mesh through the 3D modeling module, assuming the number of triangular faces generated , each area Calculated using the following formula:

[0203] ,

[0204] Cross Product , mold length , the area of ​​a single triangle:

[0205] ,

[0206] Surface Area:

[0207] ,

[0208] 6. Rate of change of lesion shape;

[0209] Assume that the surface area at the previous moment is , then the shape change rate:

[0210] .

[0211] Example 2: Tumor lesion monitoring;

[0212] 1. Initial volume and surface area calculation: After the acquired bladder images are segmented, the number of voxels in the lesion area is , the voxel size is (1.5,1.5,2.0)mm.

[0213] Single voxel volume:

[0214] ,

[0215] Lesion volume:

[0216] ,

[0217] Assume that 300 triangular patches are generated, each with an area of , surface area:

[0218] ,

[0219] 2. Changes after one month After segmentation, the number of voxels in the lesion area increased to , the number of triangle patches generated is .

[0220] New volume:

[0221] ,

[0222] New Surface Area:

[0223] ,

[0224] Volume growth rate:

[0225] ,

[0226] Surface area change rate:

[0227] ShapeRate ,

[0228] Example 3: Application of the federated learning module in kidney tumor image segmentation;

[0229] In medical scenarios, data from different hospitals is usually difficult to share centrally. The federated learning module can achieve distributed collaborative training, protecting data privacy while improving model performance. The following is an example of the application of the federated learning module to kidney tumor image segmentation.

[0230] 1. Multi-center data distribution: Assume that there are two hospitals (Hospital A and Hospital B) participating in federated learning:

[0231] Hospital A has 500 kidney tumor images, and Hospital B has 800 images; each hospital trains its local model independently and , using the loss function of the image segmentation module:

[0232] ,

[0233] in: : The true label of pixel i; : Pixels The predicted probability of .

[0234] 3. Federated learning global model update: The federated learning module calculates the global model parameters by weighted average :

[0235] ,

[0236] in: : No. The number of images in each hospital;

[0237] ,

[0238] Specific calculation:

[0239] ,

[0240] Element-wise calculation:

[0241] ,

[0242] Global model parameters:

[0243] .

[0244] 4. Global model verification:

[0245] Global Model After the update, push it to Hospital A and Hospital B for verification. Assumption: The segmentation accuracy of the verification set of Hospital A is improved to 85.2%; the segmentation accuracy of the verification set of Hospital B is improved to 87.8%.

[0246] 5. Effects after module optimization: Improved segmentation performance: The global model of federated learning is better than the local models independently trained by each hospital, which improves the overall performance of image segmentation. Privacy protection: The original image data does not need to be shared, only the model parameters are shared , ensuring data privacy. Collaborative efficiency: The global model is quickly updated through weighted averaging, avoiding the high cost of large-scale centralized training.

[0247] Through Example 1, Example 2 and Example 3, the application of the intelligent digital image processing system of the present invention in urology has been fully verified. The specific conclusions are as follows:

[0248] Complete processing flow: The system starts with image acquisition, and goes through preprocessing, multimodal fusion, image segmentation and other steps, ultimately achieving accurate analysis of lesion volume, surface area and its dynamic changes.

[0249] The functions of each module are demonstrated through formula calculations and data examples, reflecting the scientific nature and operability of the system.

[0250] Accuracy of calculation: The calculations of lesion volume, surface area, volume growth rate and shape change rate in the embodiments are all based on actual imaging data (such as CT, MRI), and meet the actual needs of medical imaging processing, ensuring the accuracy and rationality of the results.

[0251] Module synergy: The modules in the system work closely together, from data acquisition to intelligent diagnosis. Each module provides reliable data support for subsequent steps, ultimately generating high-quality imaging diagnostic reports.

[0252] Practical application prospects: Example 1 (kidney tumor) and Example 2 (bladder tumor) demonstrate the wide applicability of the system in urological tumor monitoring and diagnosis.

[0253] The system can not only analyze the static characteristics of the lesion area (such as volume and surface area), but also dynamically track the development trend of the lesion, providing comprehensive auxiliary support for clinical diagnosis and treatment. The application of the federated learning module in multi-center data collaboration solves the pain points of medical data privacy protection, and realizes distributed collaborative optimization through model parameter sharing, effectively improving the segmentation accuracy of the system.

[0254] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent digital image processing system for urology, characterized in that: The system includes: Data acquisition module: receiving multimodal imaging data from medical imaging equipment ,in: Indicates at time 3D image pixel value at the moment; are the spatial coordinates of the image, indicating the positions in the horizontal, vertical and depth directions respectively; is the time point, indicating the time of image acquisition; Data preprocessing module: input image data Perform noise removal, image enhancement and standardization to generate preprocessed images , calculated using the following formula: , in: : Standardized image pixel value; : The mean pixel value in the local window, calculated as ; : The standard deviation of pixels within the local window is calculated as ; : The number of pixels in the local window; : The first The intensity value of each pixel; Multimodal image fusion module: realizes the registration of multimodal images through the following objective function to generate aligned comprehensive image data : ， in: : Fusion error, the smaller the error, the more accurate the registration. : CT image pixel value after preprocessing; : pixel values ​​of the preprocessed MRI image, adjusted by the offset of the registration; : The spatial offset between CT and MRI images, aligned in horizontal, vertical and depth directions respectively; : The number of pixels in the local window; Image segmentation module: Extract lesion area through convolutional neural network (CNN) , calculated using the following formula: ， in: : Prediction results of lesion area; : Activation function, use ReLU or sigmoid activation; : Convolution kernel weight matrix, used to extract image features; *: convolution operation; : Comprehensive image data after multi-modal fusion alignment; : Bias term, used to adjust the linear offset of feature extraction results; Dynamic lesion tracking module: according to the lesion area output by the image segmentation module , calculate the volume of the lesion area And predict the future volume by the following formula : ， ， in: : Lesion volume at time t; :time The lesion area; : The volume of each voxel; : adjustment coefficient, used to correct the predicted volume change; : Lesion volume at the previous time point; 3D modeling module: using the lesion area provided by the dynamic lesion tracking module , generate a three-dimensional mesh model of the lesion area, and the interpolation formula is: , in: : voxel value of the interpolation point; : Isosurface threshold, used to define the boundary of the lesion area; and are the voxel gray values ​​at both ends of the interpolation point; : interpolation weight, indicating the position ratio of the interpolation point between the two end voxels; Intelligent diagnosis support module: Combines the results of the 3D modeling module to generate an imaging diagnosis report, including lesion area analysis, dynamic trend prediction and treatment recommendations.

2. An intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The image segmentation module optimizes the parameters of the model through the following loss function and : , in: : Loss function, used to measure the error between the segmentation result and the true label; : No. The true label of each pixel (0 or 1); : No. The predicted probability of a pixel is given by calculate; : The number of pixels in the local window.

3. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The dynamic lesion tracking module calculates the lesion volume growth rate by the following formula: , in: GrowthRate: The growth rate of the lesion volume, which indicates the ratio of the current volume to the volume at the previous moment; :time The volume of the lesion; ; :time The lesion area is defined by the output of the image segmentation module; ; : the volume of a single voxel; ; : The spatial offset between CT and MRI images; :time The lesion volume is calculated in the same way as .

4. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The multimodal image fusion module introduces a weighted optimization objective function: , in: : Fusion error, used to evaluate the registration accuracy between CT and MRI images; : Pixel weighting factor, indicating pixel weights in the registration process; : CT image pixel value after preprocessing; : Pixel values ​​of MRI images after preprocessing, adjusted by registration offset; : The spatial offset between CT and MRI images, used for registration; : The number of pixels in the local window.

5. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The 3D modeling module generates the optimal surgical path in combination with preoperative planning : , in: : Optimal path, indicating the optimal navigation path in surgical planning; : Path set, representing all possible paths; : The number of nodes in the path; :node The weight of a path to measure its importance in path optimization; : Path node The loss value of risk or cost.

6. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The imaging diagnostic report generated by the system includes: lesion size, location, and volume; Prediction of lesion development trend.

7. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The system also includes a federated learning module that implements multi-center data sharing through the following formula: , in: : Global model parameters, used for parameter optimization in the image segmentation module, in the After iterations, the weighted average of the local model parameters is obtained, including and ; : No. The local model parameters of each data center are obtained from the local training results of the image segmentation module; : No. The number of samples in each data center; : The number of pixels in the local window, .

8. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The data preprocessing module uses a filtering algorithm to remove image noise, and the calculation formula is: , in: : Pixel value of the image after filtering; : The original pixel value within the filter window; : The total number of pixels in the filter window; : The first The spatial coordinates of the pixels.

9. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The dynamic lesion tracking module combines the triangular patch data output by the 3D modeling module to calculate the surface area of ​​the lesion area. : , in: :time Surface area of ​​the lesion; : The number of triangular patches, indicating how many triangles the surface of the lesion area consists of; : No. The area of ​​a triangular patch is calculated from the mesh data output by the 3D modeling module; ； ； : The cross product of two edge vectors; : The modulus of the cross product, which means the area of ​​the triangle is twice; The lesion shape change rate is further calculated: , Among them: ShapeRate: lesion shape change rate, indicating the change ratio of the current surface area to the surface area at the previous moment; , The time and of the lesion surface area.

10. The intelligent digital image processing system for urology as claimed in claim 1, characterized in that: The system supports remote diagnosis function, and imaging diagnosis reports can be transmitted to other medical institutions via the network for remote consultation.

Images