Pancreatic cancer prediction method and prediction system based on three-dimensional visualization

Abstract

The embodiment of the invention relates to the technical field of medical imaging, and discloses a pancreatic cancer prediction method and prediction system based on three-dimensional visualization, and the method comprises the steps: carrying out the three-dimensional stacking of a two-dimensional pancreatic image sequence according to the depth sequence in an original abdomen CT image, so as to form a plurality of three-dimensional rectangular cubes, and drawing a three-dimensional visualization pancreas model according to the three-dimensional rectangular cube, and substituting the CT image feature data and the radiomics feature data obtained in the three-dimensional visualization pancreas model into the radiomics prediction model to calculate the prediction probability of the pancreatic cancer. By implementing the embodiment of the invention, the accuracy and reliability of pancreatic cancer disease prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a pancreatic cancer prediction method and prediction system based on three-dimensional visualization. Background Art

[0002] In the field of medical imaging, especially for the early diagnosis and prediction of pancreatic cancer, traditional two-dimensional CT image analysis has limitations. Although these two-dimensional images can provide cross-sectional information of the pancreas and its surrounding tissues, it is difficult to fully and intuitively display the three-dimensional morphology of the pancreas and its spatial relationship with the surrounding environment. In addition, when interpreting two-dimensional CT images, doctors often need to rely on personal experience and subjective judgment, which not only increases the complexity and time-consuming nature of diagnosis, but may also lead to misdiagnosis or missed diagnosis due to human factors. Summary of the invention

[0003] The embodiment of the present invention discloses a pancreatic cancer prediction method and prediction system based on three-dimensional visualization, which can improve the accuracy and reliability of pancreatic cancer disease prediction.

[0004] A first aspect of an embodiment of the present invention discloses a pancreatic cancer prediction method based on three-dimensional visualization, the method comprising:

[0005] The two-dimensional pancreatic image sequence is three-dimensionally stacked according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes;

[0006] According to the three-dimensional rectangular cube, a three-dimensional visualized pancreas model is drawn;

[0007] Substituting the CT image feature data and the radiomics feature data obtained in the three-dimensional visualized pancreatic model into the radiomics prediction model to calculate the prediction probability of pancreatic cancer;

[0008] Among them, the imaging omics prediction model includes imaging omics feature scores, intercept feature scores, cystic change feature scores, feature scores of pancreatic duct / pancreas diameter ratio ≥ 0.34, pancreatic duct perforation sign feature scores, pancreatic portal hypertension feature scores and arterial phase CT value feature scores. The imaging omics feature scores include pancreatic region shape flatness scores, pancreatic region abnormally low pixel intensity scores, pancreatic region abnormally high pixel intensity scores, pancreatic region average pixel intensity scores, pancreatic region average pixel intensity scores on a coarse scale and pancreatic region texture scores. The imaging omics feature score is the sum of the scores of each item in the imaging omics feature score, and the predicted probability is the sum of the scores of each item in the imaging omics prediction model.

[0009] A second aspect of an embodiment of the present invention discloses a prediction system, the prediction system comprising:

[0010] A three-dimensional stacking unit, used for three-dimensionally stacking the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes;

[0011] A drawing unit, used for drawing a three-dimensional visualized pancreas model according to the three-dimensional rectangular cube;

[0012] A first calculation unit, used for substituting the CT image feature data and the radiomics feature data obtained in the three-dimensional visualized pancreatic model into the radiomics prediction model to calculate the prediction probability of pancreatic cancer;

[0013] Among them, the imaging omics prediction model includes imaging omics feature scores, intercept feature scores, cystic change feature scores, feature scores of pancreatic duct / pancreas diameter ratio ≥ 0.34, pancreatic duct perforation sign feature scores, pancreatic portal hypertension feature scores and arterial phase CT value feature scores. The imaging omics feature scores include pancreatic region shape flatness scores, pancreatic region abnormally low pixel intensity scores, pancreatic region abnormally high pixel intensity scores, pancreatic region average pixel intensity scores, pancreatic region average pixel intensity scores on a coarse scale and pancreatic region texture scores. The imaging omics feature score is the sum of the scores of each item in the imaging omics feature score, and the predicted probability is the sum of the scores of each item in the imaging omics prediction model.

[0014] A third aspect of an embodiment of the present invention discloses a prediction system, the prediction system comprising:

[0015] A memory storing executable program code;

[0016] a processor coupled to the memory;

[0017] The processor calls the executable program code stored in the memory to execute a pancreatic cancer prediction method based on three-dimensional visualization disclosed in the first aspect of the embodiment of the present invention.

[0018] A fourth aspect of an embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute a pancreatic cancer prediction method based on three-dimensional visualization disclosed in the first aspect of an embodiment of the present invention.

[0019] A fifth aspect of an embodiment of the present invention discloses a computer program product. When the computer program product is run on a computer, the computer is enabled to execute part or all of the steps of any one of the three-dimensional visualization-based pancreatic cancer prediction methods of the first aspect.

[0020] A sixth aspect of an embodiment of the present invention discloses an application publishing platform, which is used to publish a computer program product, wherein when the computer program product is run on a computer, the computer executes part or all of the steps of any one of the pancreatic cancer prediction methods based on three-dimensional visualization in the first aspect.

[0021] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0022] In an embodiment of the present invention, a two-dimensional pancreatic image sequence is three-dimensionally stacked according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes; a three-dimensional visualized pancreatic model is drawn according to the three-dimensional rectangular cubes; the CT image feature data and the imaging genomics feature data obtained in the three-dimensional visualized pancreatic model are substituted into the imaging genomics prediction model to calculate the predicted probability of pancreatic cancer; wherein the imaging genomics prediction model includes an imaging genomics feature score, an intercept feature score, a cystic change feature score, a feature score of a pancreatic duct / pancreatic diameter ratio ≥ 0.34, a pancreatic duct perforation sign feature score, a pancreatic portal hypertension feature score and an arterial phase CT value feature score, the imaging genomics feature score includes a pancreatic region shape flatness score, a pancreatic region abnormally low pixel intensity score, a pancreatic region abnormally high pixel intensity score, a pancreatic region average pixel intensity score, a pancreatic region average pixel intensity score on a coarse scale and a pancreatic region texture score, the imaging genomics feature score is the sum of the scores of each item in the imaging genomics feature score, and the predicted probability is the sum of the scores of each item in the imaging genomics prediction model. It can be seen that the embodiments of the present invention can improve the accuracy and reliability of pancreatic cancer disease prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only 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.

[0024] Figure 1 It is a flow chart of a pancreatic cancer prediction method based on three-dimensional visualization disclosed in an embodiment of the present invention;

[0025] Figure 2 It is a flow chart of another pancreatic cancer prediction method based on three-dimensional visualization disclosed in an embodiment of the present invention;

[0026] Figure 3 is a structural schematic diagram of a prediction system disclosed in an embodiment of the present invention;

[0027] Figure 4is a schematic diagram of the structure of another prediction system disclosed in an embodiment of the present invention;

[0028] Figure 5 It is a schematic diagram of the structure of another prediction system disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0030] It should be noted that the terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present invention and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] The embodiment of the present invention discloses a pancreatic cancer prediction method and prediction system based on three-dimensional visualization, which can improve the accuracy and reliability of pancreatic cancer disease prediction.

[0032] The following is a detailed description with reference to the accompanying drawings.

[0033] Embodiment 1

[0034] See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a pancreatic cancer prediction method based on three-dimensional visualization disclosed in an embodiment of the present invention. Figure 1 As shown, the pancreatic cancer prediction method based on three-dimensional visualization may include the following steps.

[0035] The prediction system stacks the two-dimensional pancreatic image sequence in three-dimensional order according to the depth order in the original abdominal CT image to form multiple three-dimensional rectangular cubes.

[0036] As an optional implementation, in an embodiment of the present invention, after the system sorts the two-dimensional image sequence in depth order, it can normalize all the two-dimensional images by cropping or interpolation. Then, based on the width, height and depth (i.e., the number of images) of the image sequence, the system can create a three-dimensional array. Finally, the system can copy the sorted two-dimensional image sequence to the corresponding depth position of the three-dimensional array to obtain a three-dimensional rectangular cube.

[0037] As an optional implementation, in an embodiment of the present invention, first, the present application can use the original original abdominal CT images as input data, and these CT images contain multiple slices of the abdominal area, and each slice is a two-dimensional image. Subsequently, the system can use image segmentation network models (such as U-Net, 3D U-Net, ResNet, etc.) to segment the input CT images, and these models are trained to identify and segment the pancreatic area. After image segmentation, the image segmentation network model can output a series of two-dimensional images containing the pancreatic area, which are arranged in the order of the original CT images to form a two-dimensional pancreatic image sequence.

[0038] As an optional implementation, in an embodiment of the present invention, the present application can arrange the obtained two-dimensional pancreatic image sequence according to the depth order in the original CT image, which can ensure that the order of the images corresponds to their actual positions in the human body when stacked in three dimensions. Subsequently, the system can use image processing software or programming tools (such as MATLAB, Python, etc.) to stack the arranged two-dimensional pancreatic image sequence in the depth direction, and during the stacking process, each two-dimensional image corresponds to a slice layer in a three-dimensional rectangular cube. After stacking, the system can obtain a three-dimensional rectangular cube containing the pancreatic region, which can be further used for tasks such as three-dimensional reconstruction, visualization analysis, and surgical planning.

[0039] 102. The prediction system draws a three-dimensional visual pancreas model based on the three-dimensional rectangular cube.

[0040] 103. The prediction system substitutes the CT image feature data and the imaging genomics feature data obtained in the three-dimensional visualized pancreatic model into the imaging genomics prediction model to calculate the prediction probability of pancreatic cancer.

[0041] In an embodiment of the present invention, the imaging omics prediction model includes an imaging omics feature score, an intercept feature score, a cystic change feature score, a feature score of a pancreatic duct / pancreas diameter ratio ≥ 0.34, a pancreatic duct perforation sign feature score, a pancreatic portal hypertension feature score, and an arterial phase CT value feature score. The imaging omics feature score includes a pancreatic region shape flatness score, a pancreatic region abnormally low pixel intensity score, a pancreatic region abnormally high pixel intensity score, a pancreatic region average pixel intensity score, a pancreatic region average pixel intensity score at a coarse scale, and a pancreatic region texture score. The imaging omics feature score is the sum of the scores in the imaging omics feature score, and the prediction probability is the sum of the scores in the imaging omics prediction model.

[0042] As an optional implementation, in an embodiment of the present invention, the CT image feature data in this application may include grayscale value statistics (such as mean, standard deviation, histogram distribution, etc.), texture features (such as grayscale co-occurrence matrix, local binary pattern, etc.) and shape features of the image. In addition to the above-mentioned grayscale, texture and shape features, the imaging genomics feature data may also include more complex features, such as features based on wavelet transform, Laplace pyramid transform, fractional Brownian motion, etc., which are not limited in this application.

[0043] As an optional implementation, in the embodiment of the present invention, since the number of extracted features may be very large (hundreds or thousands), directly using all features for the prediction model may lead to overfitting or low computational efficiency. Therefore, the present application may perform feature selection or dimensionality reduction through statistical methods (such as chi-square test, mutual information), machine learning techniques (such as random forest, recursive feature elimination) or deep learning algorithms (such as autoencoders) to screen out the feature subset with the most predictive ability for the prediction target (i.e., pancreatic cancer).

[0044] As an optional implementation, in the embodiment of the present invention, compared with the traditional pancreatic analysis which is mostly based on two-dimensional CT images, the present application can more accurately reflect the shape and position of the pancreas in three-dimensional space by stacking the two-dimensional pancreatic image sequence in three dimensions and constructing a three-dimensional volume model. Compared with the traditional two-dimensional image analysis, the transition from two-dimensional to three-dimensional in this three-dimensional reconstruction method not only improves the accuracy of image analysis, but also provides more comprehensive information, which helps doctors to more accurately judge the abnormality of the pancreas, thereby improving the diagnostic accuracy of diseases such as pancreatic cancer.

[0045] As an optional implementation, in the embodiment of the present invention, the present application uses a triangulation algorithm and normal vector calculation to draw a three-dimensional visualized pancreas model with a high sense of reality and stereoscopic sense. This model can not only help doctors to intuitively observe the morphology and structure of the pancreas, but also can be examined from different angles through operations such as rotation and scaling, so as to facilitate the discovery of potential problem areas.

[0046] As an optional implementation, in the embodiment of the present invention, the CT image feature data and radiomics feature data extracted by the system from the three-dimensional visualized pancreas model provide a rich data source for radiomics analysis. Radiomics is an emerging technology that uses a large number of image features for disease prediction and diagnosis. The addition of these feature data can significantly improve the accuracy and reliability of the prediction model.

[0047] As an optional implementation, in the embodiment of the present invention, the automated 3D reconstruction and feature extraction process reduces the time doctors spend on manually analyzing and processing images, thus improving work efficiency. At the same time, the radiomics-based prediction model can quickly calculate the predicted probability of pancreatic cancer, providing doctors with a timely reference for formulating treatment plans.

[0048] As an optional implementation, in an embodiment of the present invention, radiomics itself relies on a large amount of image feature data, while the three-dimensional visualization model provides a richer and more accurate source of feature data. The combination of the two can significantly improve the accuracy and reliability of disease prediction.

[0049] As an optional implementation, in an embodiment of the present invention, the entire process from image input to calculation of prediction probability is highly automated, which not only reduces human intervention and errors, but also improves work efficiency and data processing capabilities.

[0050] exist Figure 1 In the pancreatic cancer prediction method based on three-dimensional visualization, the prediction system is used as an example to describe it. It should be noted that Figure 1 The execution subject of the pancreatic cancer prediction method based on three-dimensional visualization may also be an independent device associated with the prediction system, which is not limited in the embodiment of the present invention.

[0051] It can be seen that implementation Figure 1 The described pancreatic cancer prediction method based on three-dimensional visualization can improve the accuracy and reliability of pancreatic cancer disease prediction.

[0052] In addition, implementation Figure 1 A pancreatic cancer prediction method based on three-dimensional visualization is described, which reduces human intervention and errors.

[0053] Embodiment 2

[0054] See also Figure 2 , Figure 2 FIG. 1 is a flow chart of another pancreatic cancer prediction method based on three-dimensional visualization disclosed in an embodiment of the present invention. Figure 2 , the pancreatic cancer prediction method based on three-dimensional visualization may include the following steps:

[0055] 201. The prediction system uses the lasso regression formula to analyze the imaging genomics feature data training set to determine the weights of each item in the imaging genomics feature score.

[0056] As an optional implementation, in an embodiment of the present invention, the Lasso regression (Least Absolute Shrinkage and Selection Operator) formula is used to analyze the radiomics feature data training set, which is an effective way to determine the weight ratio of each image scoring feature (also called image biomarker or feature) in the radiomics score. Lasso regression can not only help the model to predict, but also has the ability of feature selection, by introducing regularization terms to reduce model complexity, avoid overfitting, and automatically select features that have a greater impact on the target variable.

[0057] 202. The prediction system uses multivariate logistic regression analysis to analyze the CT image feature data training set to determine the scoring weights of each item in the imaging genomics prediction model.

[0058] As an optional implementation, in the embodiment of the present invention, multivariate logistic regression can process multiple independent variables at the same time, considering the impact of their interactions on the prediction results. At the same time, the weight ratio given by the model can be directly used to explain the importance of each feature to the prediction result.

[0059] 203. The prediction system inputs the original abdominal CT image into a convolutional neural network with residual connection and attention gating module to obtain a residual convolution block.

[0060] The prediction system uses the Sigmoid function to normalize the output residual convolution block and generates a pixel probability map to obtain the probability that each pixel in the original abdominal CT image belongs to the pancreas.

[0061] 205. The prediction system uses the Dice loss function to calculate the error of the pixel probability map to determine the two-dimensional pancreatic image sequence.

[0062] As an optional implementation, in an embodiment of the present invention, the system may first receive the original abdominal CT images as input, which usually contain rich anatomical information, but may also contain noise and unnecessary details. In a specific convolutional neural network, the present application introduces residual connections and attention gating modules to enhance the learning ability and efficiency of the network. These residual connections allow the network to directly pass part of the input to a deeper layer, which helps to alleviate the gradient vanishing or gradient explosion problems in deep networks and accelerate the training process. These attention gating modules can enhance the network's sensitivity to key features by learning which areas in the image are more important to the task (i.e., the attention area). They can help the network focus more on the pancreatic area and reduce interference from non-related areas. After multi-layer convolution, activation, and nonlinear transformation of the network, the system can generate residual convolution blocks. These blocks not only contain high-level features of the image, but also retain part of the original input information through residual connections, which helps to maintain the integrity of the information. Subsequently, the system can use the Sigmoid function to normalize the output of the residual convolution block. The Sigmoid function compresses any real value into the (0,1) interval, which can be interpreted as a probability value. In this context, it generates a pixel probability map, where the value of each pixel represents the probability that the point belongs to the pancreas. This map provides the basis for subsequent image segmentation and analysis. The high-probability areas in the map correspond to parts that are more likely to be the pancreas, while the low-probability areas correspond to other tissues or background. In order to evaluate the accuracy of the pixel probability map and optimize the network parameters accordingly, the system can use the Dice loss function, which is a loss function commonly used in image segmentation tasks. It can directly measure the overlap between the predicted segmentation and the true segmentation, and has good robustness to the imbalance problem of foreground (in this case, the pancreas) and background. By minimizing the Dice loss, the system continuously optimizes its prediction ability and is ultimately able to accurately segment the pancreatic region from the original abdominal CT image. Since CT images are usually three-dimensional, but here we are discussing a two-dimensional processing process (possibly layer-by-layer processing of three-dimensional images), the final result is a series of two-dimensional pancreatic images, which together constitute a three-dimensional representation of the pancreas. However, it should be noted that a three-dimensional convolutional neural network can actually be constructed to directly process three-dimensional CT images, thereby more accurately segmenting the three-dimensional pancreatic structure.

[0063] As an optional implementation, in an embodiment of the present invention, traditional convolutional neural networks may face problems of insufficient feature extraction or information loss when processing complex medical images. The prediction system combines residual connections with attention gating modules, which not only retains the advantages brought by the depth of the network, but also enhances the network's attention to key areas through the attention mechanism, and can more effectively extract features from the original abdominal CT images, especially for the pancreas, which may appear complex in the image and have low contrast with surrounding tissues. Organs. Residual connections help keep information flowing in deep networks and reduce information loss, while the attention gating module enhances the network's attention to the pancreatic area and reduces the interference of background noise. These have significantly improved the accuracy of pancreatic segmentation.

[0064] As an optional implementation, in the present invention, residual connections not only improve the performance of the model, but also accelerate the training process by allowing gradients to flow directly through these connections. This solves the common gradient vanishing problem in deep networks, allowing the model to converge to the optimal solution faster.

[0065] As an optional implementation, in the embodiment of the present invention, the Sigmoid function normalizes the output to a probability value, making the model prediction smoother and helping to deal with uncertainty and noise in the image. In addition, the Dice loss function has good robustness to the imbalance problem of foreground and background, further enhancing the performance of the model in complex scenes.

[0066] As an optional implementation, in the embodiment of the present invention, the entire prediction process is highly automated, from image input to output of pancreatic image sequence, without the need for human intervention, which greatly improves the efficiency and accuracy of medical diagnosis.

[0067] 206. The prediction system stacks the two-dimensional pancreatic image sequence in three-dimensional order according to the depth order in the original abdominal CT image to form multiple three-dimensional rectangular cubes.

[0068] 207. The prediction system determines two rectangular vertices connected to the edge of each three-dimensional rectangular cube equivalent point; wherein the three-dimensional rectangular cube equivalent point is a point on the three-dimensional rectangular cube whose grayscale value is equal to a specified threshold.

[0069] In an embodiment of the present invention, the system may first obtain the grayscale value of each rectangular vertex in the three-dimensional rectangular cube, and then determine the rectangular vertices corresponding to the grayscale values ​​of two adjacent rectangular vertices located on both sides of a specified threshold as target vertices.

[0070] As an optional implementation, in an embodiment of the present invention, the system may first read three-dimensional image data, which is stacked by a series of two-dimensional images (slices), and the pixel points on each slice correspond to the grayscale value of a cube in three-dimensional space. Subsequently, for each cube in the three-dimensional cube, the system needs to access and extract the grayscale values ​​of its eight vertices (or corner points). These vertices are located at the four corners of the cube and correspond to the intersection of the three-dimensional space grid. Next, the system applies a specified grayscale threshold to distinguish different tissues or structures. This threshold is determined based on prior knowledge or an automatic algorithm (such as the Otsu method) to distinguish between tissues of interest (such as the pancreas) and background or other tissues. Subsequently, the system may traverse the vertices of all three-dimensional cubes, and for each vertex, check whether its grayscale value is on both sides of the specified threshold. If the grayscale values ​​of two adjacent vertices of a cube (directly connected in three-dimensional space) are respectively on both sides of the threshold, the two vertices (and the cubes in which they are located) are determined as vertices of the target cube or part of the target cube. This step is usually used for edge detection or interface recognition, especially when it is necessary to distinguish the boundaries of different tissues or structures. Once the target cube vertices are determined, the prediction system can further process these vertices or the cubes to which they belong to perform more complex tasks, such as 3D reconstruction, volume calculation, shape analysis, etc. Finally, the system may output the processing results in a visual form, such as a 3D rendered image, a slice view, or an analysis report, to facilitate diagnosis or further analysis by medical experts.

[0071] As an optional implementation, in an embodiment of the present invention, the traditional threshold segmentation method usually only considers the gray value of the pixel (or cube) and ignores its spatial position information. However, this method actually combines local gray information and spatial position information by determining two adjacent cube vertices located on both sides of the specified threshold as the target cube vertices, making the segmentation result more accurate and reliable, and can more accurately identify the boundaries of different tissues or structures. This method takes into account the continuity between cubes and the local changes in gray values, which helps to reduce the over-segmentation or under-segmentation caused by single threshold processing.

[0072] As an optional implementation, in the embodiment of the present invention, two adjacent cube vertices located on both sides of the specified threshold are usually located at the junction of different tissues or structures, that is, the edge area. Determining these vertices as target cube vertices is actually performing edge detection. This helps to more accurately reconstruct the three-dimensional shape of the tissue in subsequent processing and improve the reliability of image analysis.

[0073] As an alternative implementation, in the embodiments of the present invention, the entire process is completely automatically completed by the prediction system without manual intervention, greatly improving the automation level of medical image processing. This can not only save human resources, but also reduce human errors and improve the accuracy and efficiency of diagnosis.

[0074] 208. After the prediction system calculates the isosurface point coordinates of each three-dimensional rectangular cube using the linear interpolation algorithm, it determines whether there are N isosurface point coordinates in each three-dimensional rectangular cube. If so, steps 209 to 212 are executed; if not, this process ends.

[0075] As an alternative implementation, in the embodiments of the present invention, in each three-dimensional cube, points with gray values equal to the specified threshold are searched for, that is, the isosurface points of the three-dimensional rectangular cube. These points usually represent the boundary between the pancreas and the surrounding tissues. Since the gray values in the actual data change continuously, the isosurface points often do not exactly fall on a certain vertex, but are located on a certain edge of the cube. When the isosurface point is located on a certain edge of the cube, we can perform linear interpolation based on the gray values and coordinates of the two vertices on this edge to calculate the exact coordinates of the isosurface point.

[0076] As an alternative implementation, in the embodiments of the present invention, assume that we find a three-dimensional rectangular cube, and the gray values of the two vertices on a certain edge are G1 and G2 (G1 < Gthreshold < G2, where Gthreshold is the specified threshold), and the coordinates of these two vertices are (x1, y1, z1) and (x2, y2, z2) respectively. First, the system can first use the gray value to calculate the interpolation ratio t such that G1 + t(G2 - G1) = Gthreshold, and then solve the equation t = (Gthreshold - G1) / (G2 - G1) to obtain the result. Subsequently, the system uses the obtained interpolation ratio t to calculate the coordinates (x, y, z) of the isosurface point, that is, solve the system of equations: x = x1 + t(x2 - x1), y = y1 + t(y2 - y1), z = z1 + t(z2 - z1).

[0077] As an alternative implementation, in the embodiments of the present invention, in the visualization processes such as isosurface extraction and three-dimensional reconstruction of the present application, using the exact isosurface points of the three-dimensional rectangular cube can generate a smoother and more continuous isosurface, reduce the jagged edges or holes caused by data simplification, and improve the visual effect.

[0078] As an optional implementation, in the embodiment of the present invention, in a three-dimensional rectangular cube, by determining the two rectangular vertices connected by the edge where the three-dimensional rectangular cube is located, and using a linear interpolation algorithm to calculate the coordinates of the three-dimensional rectangular cube isovalued point, this method is more accurate than simple threshold segmentation. It can more accurately locate the boundary between the pancreas and the surrounding tissues, providing a reliable basis for subsequent three-dimensional reconstruction and feature extraction.

[0079] 209. The prediction system will use a triangulation algorithm to connect the coordinates of all equal-value points in a three-dimensional rectangular cube to form equal-value point triangular faces; where N is a natural number not less than 3.

[0080] As an optional implementation, in the embodiment of the present invention, triangulation is the process of connecting a set of points (here, 3D rectangular cube equivalent points) into a series of non-overlapping triangles, which together cover the entire isosurface. In three-dimensional space, there are a variety of triangulation algorithms to choose from, such as Delaunay triangulation, MarchingCubes algorithm (although Marching Cubes directly generates isosurfaces within voxels instead of just points, but the concept is similar), etc.

[0081] As an optional implementation, in an embodiment of the present invention, first, the system may select a suitable triangulation algorithm according to specific requirements and data characteristics. Subsequently, the system may use the equivalent points of a three-dimensional rectangular cube as input, and apply the triangulation algorithm to generate triangular faces, each of which is defined by three equivalent points, thereby generating a triangular mesh representation of the equivalent surface. Finally, the system calculates the triangular face normal vector for the equivalent point triangular face. The triangular face normal vector is a vector perpendicular to the triangular face, which is very important for lighting calculation, surface direction analysis, etc., and can provide a basis for subsequent image rendering, analysis, and visualization.

[0082] 210. The prediction system calculates the cross product of two edges of each equal-value point triangle to obtain the triangle normal vector, and normalizes each triangle normal vector.

[0083] As an optional implementation, in the embodiment of the present invention, the present application combines the judgment of equivalent points with the triangulation algorithm to form a new three-dimensional data processing method. This method not only considers the existence of equivalent points, but also uses these points to construct triangular surfaces with actual geometric significance, providing richer information for subsequent analysis.

[0084] As an optional implementation, in an embodiment of the present invention, the present application ensures the accuracy and consistency of the normal vector of each triangle by calculating the cross product of two edges in the triangle and normalizing the normal vector. This precise calculation method helps to improve the quality of subsequent graphics rendering, lighting simulation and other tasks.

[0085] As an optional implementation, in the embodiment of the present invention, the process integrates multiple technologies such as equivalent point judgment, triangulation, cross product calculation and normal vector normalization to form an innovative three-dimensional data processing method. This technical integration not only improves the accuracy and efficiency of data processing, but also expands the application scope of three-dimensional data analysis.

[0086] The prediction system connects the adjacent equal-valued triangles corresponding to the normal vectors of the triangles with the same direction, and performs three-dimensional visualization processing to draw a three-dimensional visualized pancreatic model.

[0087] As an optional implementation, in an embodiment of the present invention, the prediction system first uses a spatial index (such as an octree, a kd tree, or an R tree, etc.) to quickly find the adjacent triangles in space for each equal-value point triangle. This method can significantly reduce unnecessary comparisons and improve search efficiency. For each triangle, the system can calculate its normal vector and compare it with the normal vectors of the adjacent triangles. If the normal vectors of two adjacent triangles have the same direction (or are nearly the same, considering a certain tolerance range), the two triangles are considered to have the "same direction" and are marked as connectable candidates. Based on the matching results of the previous step, the system can connect adjacent triangles with the same direction. This usually involves updating the vertex list and edge list of the triangle to ensure that the topological structure of the entire model is correct. After completing the connection of the triangles, the system can pass these triangle data to the three-dimensional visualization engine. The visualization engine uses this data to render the three-dimensional model, including applying effects such as lighting, materials, and textures to generate a realistic three-dimensional visualization pancreas model.

[0088] As an optional implementation, in an embodiment of the present invention, by ensuring that only adjacent triangles with the same orientation are connected, the system can generate a more accurate three-dimensional model. This reduces the distortion or deformation of the model caused by incorrect connections. Reasonable triangle connection can optimize the topological structure of the model, making it more suitable for subsequent processing and analysis. For example, in medical imaging, the optimized model can be better used for volume measurement, shape analysis or pathological diagnosis.

[0089] As an optional implementation, in an embodiment of the present invention, three-dimensional visualization processing enables the pancreas model to be presented to the user in an intuitive and realistic manner. This not only helps medical experts to make more accurate diagnoses, but can also be used for teaching and scientific research purposes. In a three-dimensional visualization environment, users can rotate, scale, and slice the model to observe and analyze the model from different angles. This interactivity enhances the user's understanding and cognition of the model.

[0090] As an optional implementation, in an embodiment of the present invention, the present application may first store the vertex coordinates of all triangles in a buffer so that the GPU can quickly access them. If the triangles share vertices, the system can use an index buffer to reduce the duplication of vertex data and improve rendering efficiency. Subsequently, the system may store the normal vector of each triangle in a buffer for lighting and shadow calculations. Finally, the system may also assign appropriate materials and textures to the pancreas model to enhance the visual effect, so as to complete the rendering preparation of the three-dimensional visualized pancreas model.

[0091] As an optional implementation, in the embodiment of the present invention, in order to make the three-dimensional visualized pancreas model more vivid and useful, the system can add interactivity and animation effects. That is, user interaction: allowing the user to rotate, zoom, and translate the model to observe from different angles. Animation: If the model is based on time series data (such as dynamic CT scans), animation can be added to show the changes of the pancreas over time.

[0092] The prediction system substitutes the CT image feature data and the imaging genomics feature data obtained from the three-dimensional visualized pancreatic model into the imaging genomics prediction model to calculate the predicted probability of pancreatic cancer.

[0093] In an embodiment of the present invention, the imaging omics prediction model includes an imaging omics feature score, an intercept feature score, a cystic change feature score, a feature score of a pancreatic duct / pancreas diameter ratio ≥ 0.34, a pancreatic duct perforation sign feature score, a pancreatic portal hypertension feature score, and an arterial phase CT value feature score. The imaging omics feature score includes a pancreatic region shape flatness score, a pancreatic region abnormally low pixel intensity score, a pancreatic region abnormally high pixel intensity score, a pancreatic region average pixel intensity score, a pancreatic region average pixel intensity score at a coarse scale, and a pancreatic region texture score. The imaging omics feature score is the sum of the scores in the imaging omics feature score, and the prediction probability is the sum of the scores in the imaging omics prediction model.

[0094] It can be seen that implementation Figure 2 Another pancreatic cancer prediction method based on three-dimensional visualization is described, which can improve the accuracy and reliability of pancreatic cancer disease prediction.

[0095] In addition, implementation Figure 2 Another pancreatic cancer prediction method based on three-dimensional visualization is described, which can improve the efficiency and accuracy of model construction.

[0096] Embodiment 3

[0097] See also Figure 3 , Figure 3 Schematic diagram of a prediction system disclosed in an embodiment of the present invention. Figure 3 , the prediction system 300 may include a three-dimensional stacking unit 301, a drawing unit 302 and a first calculation unit 303, wherein:

[0098] The three-dimensional stacking unit 301 is used to perform three-dimensional stacking on the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes.

[0099] The drawing unit 302 is used to draw a three-dimensional visualized pancreas model according to the three-dimensional rectangular cube.

[0100] The first calculation unit 303 is used to substitute the CT image feature data and the radiomics feature data obtained in the three-dimensional visualized pancreas model into the radiomics prediction model to calculate the prediction probability of pancreatic cancer.

[0101] In this embodiment, the imaging omics prediction model includes an imaging omics feature score, an intercept feature score, a cystic change feature score, a feature score of a pancreatic duct / pancreas diameter ratio ≥ 0.34, a pancreatic duct perforation sign feature score, a pancreatic portal hypertension feature score, and an arterial phase CT value feature score. The imaging omics feature score includes a pancreatic region shape flatness score, a pancreatic region abnormally low pixel intensity score, a pancreatic region abnormally high pixel intensity score, a pancreatic region average pixel intensity score, a pancreatic region average pixel intensity score at a coarse scale, and a pancreatic region texture score. The imaging omics feature score is the sum of the scores in the imaging omics feature score, and the prediction probability is the sum of the scores in the imaging omics prediction model.

[0102] As an optional implementation, in an embodiment of the present invention, after the three-dimensional stacking unit 301 sorts the two-dimensional image sequence in depth order, it can normalize all the two-dimensional images by cropping or interpolation. Then, based on the width, height and depth (i.e., the number of images) of the image sequence, the system can create a three-dimensional array. Finally, the three-dimensional stacking unit 301 can copy the sorted two-dimensional image sequence to the corresponding depth position of the three-dimensional array to obtain a three-dimensional rectangular cube.

[0103] As an optional implementation, in an embodiment of the present invention, first, the three-dimensional stacking unit 301 can use the original original abdominal CT images as input data, and these CT images contain multiple slices of the abdominal area, and each slice is a two-dimensional image. Subsequently, the three-dimensional stacking unit 301 can use an image segmentation network model (such as U-Net, 3DU-Net, ResNet, etc.) to segment the input CT image, and these models are trained to identify and segment the pancreatic region. After image segmentation, the image segmentation network model can output a series of two-dimensional images containing the pancreatic region, which are arranged in the order of the original CT images to form a two-dimensional pancreatic image sequence.

[0104] As an optional implementation, in an embodiment of the present invention, the three-dimensional stacking unit 301 can arrange the obtained two-dimensional pancreatic image sequence according to the depth order of the original CT image, which can ensure that the order of the images corresponds to their actual positions in the human body during three-dimensional stacking. Subsequently, the three-dimensional stacking unit 301 can use image processing software or programming tools (such as MATLAB, Python, etc.) to stack the arranged two-dimensional pancreatic image sequence in the depth direction, and during the stacking process, each two-dimensional image corresponds to a slice layer in a three-dimensional rectangular cube. After stacking, the system can obtain a three-dimensional rectangular cube containing the pancreatic region, and this three-dimensional rectangular cube can be further used for tasks such as three-dimensional reconstruction, visualization analysis, and surgical planning.

[0105] As an optional implementation, in an embodiment of the present invention, the CT image feature data in this application may include grayscale value statistics (such as mean, standard deviation, histogram distribution, etc.), texture features (such as grayscale co-occurrence matrix, local binary pattern, etc.) and shape features of the image. In addition to the above-mentioned grayscale, texture and shape features, the imaging genomics feature data may also include more complex features, such as features based on wavelet transform, Laplace pyramid transform, fractional Brownian motion, etc., which are not limited in this application.

[0106] As an optional implementation, in the embodiment of the present invention, since the number of extracted features may be very large (hundreds or thousands), directly using all features for the prediction model may lead to overfitting or low computational efficiency. Therefore, the present application may perform feature selection or dimensionality reduction through statistical methods (such as chi-square test, mutual information), machine learning techniques (such as random forest, recursive feature elimination) or deep learning algorithms (such as autoencoders) to screen out the feature subset with the most predictive ability for the prediction target (i.e., pancreatic cancer).

[0107] As an optional implementation, in the embodiment of the present invention, compared with the traditional pancreatic analysis which is mostly based on two-dimensional CT images, the rendering unit 302 can more accurately reflect the shape and position of the pancreas in three-dimensional space by stacking the two-dimensional pancreatic image sequence in three dimensions and constructing a three-dimensional volume model. Compared with the traditional two-dimensional image analysis, the transition from two-dimensional to three-dimensional in this three-dimensional reconstruction method not only improves the accuracy of image analysis, but also provides more comprehensive information, which helps doctors to more accurately judge the abnormality of the pancreas, thereby improving the diagnostic accuracy of diseases such as pancreatic cancer.

[0108] As an optional implementation, in the embodiment of the present invention, the present application uses a triangulation algorithm and normal vector calculation to draw a three-dimensional visualized pancreas model with a high sense of reality and stereoscopic sense. This model can not only help doctors to intuitively observe the morphology and structure of the pancreas, but also can be examined from different angles through operations such as rotation and scaling, so as to facilitate the discovery of potential problem areas.

[0109] As an optional implementation, in the embodiment of the present invention, the CT image feature data and radiomics feature data extracted by the first computing unit 303 from the three-dimensional visualized pancreas model provide a rich data source for radiomics analysis. Radiomics is an emerging technology that uses a large number of image features for disease prediction and diagnosis. The addition of these feature data can significantly improve the accuracy and reliability of the prediction model.

[0110] As an optional implementation, in the embodiment of the present invention, the automated 3D reconstruction and feature extraction process reduces the time doctors spend on manually analyzing and processing images, thus improving work efficiency. At the same time, the radiomics-based prediction model can quickly calculate the predicted probability of pancreatic cancer, providing doctors with a timely reference for formulating treatment plans.

[0111] As an optional implementation, in an embodiment of the present invention, radiomics itself relies on a large amount of image feature data, while the three-dimensional visualization model provides a richer and more accurate source of feature data. The combination of the two can significantly improve the accuracy and reliability of disease prediction.

[0112] As an optional implementation, in an embodiment of the present invention, the entire process from image input to calculation of prediction probability is highly automated, which not only reduces human intervention and errors, but also improves work efficiency and data processing capabilities.

[0113] It can be seen that implementation Figure 3 The described prediction system can improve the accuracy and reliability of pancreatic cancer disease prediction.

[0114] In addition, implementation Figure 3 The described prediction system reduces human intervention and errors.

[0115] Embodiment 4

[0116] See also Figure 4 , Figure 4 is a schematic diagram of the structure of another prediction system disclosed in an embodiment of the present invention. Figure 4 The prediction system is composed of Figure 3 The prediction system is optimized. Figure 3 Compared with the prediction system, Figure 4 The forecasting system also includes:

[0117] The analysis unit 304 is used for analyzing the imaging genomics feature data training set using the lasso regression formula to determine the weights of each item in the imaging genomics feature score before the three-dimensional stacking unit 301 performs three-dimensional stereo stacking on the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes.

[0118] As an optional implementation, in an embodiment of the present invention, the present application uses the Lasso regression (Least Absolute Shrinkage and Selection Operator) formula to analyze the radiomics feature data training set, which is an effective way to determine the weight ratio of each image scoring feature (also called image biomarker or feature) in the radiomics score. Lasso regression can not only help the model to predict, but also has the ability of feature selection, by introducing regularization terms to reduce model complexity, avoid overfitting, and automatically select features that have a greater impact on the target variable.

[0119] As an optional implementation, in an embodiment of the present invention, the analysis unit 304 is also used to analyze the CT image feature data training set using a multivariate logistic regression analysis method to determine the scoring weights of each item in the imaging omics prediction model.

[0120] As an optional implementation, in the embodiment of the present invention, multivariate logistic regression can process multiple independent variables at the same time, considering the impact of their interactions on the prediction results. At the same time, the weight ratio given by the model can be directly used to explain the importance of each feature to the prediction result.

[0121] and Figure 3 Compared with the prediction system, Figure 4 The forecasting system also includes:

[0122] The input unit 305 is used for inputting the original abdominal CT image into a convolutional neural network with residual connection and attention gating module to obtain a residual convolution block after the three-dimensional stacking unit 301 three-dimensionally stacks the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes, and after the analysis unit 304 uses a multivariate logistic regression analysis method to analyze the CT image feature data training set to determine the scoring weights of each item in the imaging omics prediction model.

[0123] The processing unit 306 is used to normalize the output residual convolution block using a Sigmoid function and generate a pixel probability map to obtain the probability that each pixel in the original abdominal CT image belongs to the pancreas.

[0124] The second calculation unit 307 is used to perform error calculation on the pixel probability map using a Dice loss function to determine a two-dimensional pancreatic image sequence.

[0125] As an optional implementation, in an embodiment of the present invention, the input unit 305 may first receive the original abdominal CT images as input, which usually contain rich anatomical information, but may also contain noise and unnecessary details. In a specific convolutional neural network, the present application introduces residual connections and attention gating modules to enhance the learning ability and efficiency of the network. These residual connections allow the network to directly pass part of the input to a deeper layer, which helps to alleviate the gradient vanishing or gradient explosion problems in the deep network and accelerate the training process. These attention gating modules can enhance the network's sensitivity to key features by learning which areas in the image are more important to the task (i.e., the attention area). They can help the network focus more on the pancreatic area and reduce interference from non-related areas. After multi-layer convolution, activation and nonlinear transformation of the network, the input unit 305 can generate residual convolution blocks. These blocks not only contain high-level features of the image, but also retain part of the original input information through residual connections, which helps to maintain the integrity of the information. Subsequently, the processing unit 306 can use the Sigmoid function to normalize the output of the residual convolution block. The Sigmoid function compresses any real value into the interval (0,1) and can be interpreted as a probability value. In this context, it generates a pixel probability map, in which the value of each pixel represents the probability that the point belongs to the pancreas. This map provides the basis for subsequent image segmentation and analysis. The high-probability area in the map corresponds to the part that is more likely to be the pancreas, while the low-probability area corresponds to other tissues or background. In order to evaluate the accuracy of the pixel probability map and optimize the network parameters accordingly, the second computing unit 307 can use the Dice loss function, which is a loss function commonly used in image segmentation tasks. It can directly measure the overlap between the predicted segmentation and the true segmentation, and has good robustness to the imbalance problem of foreground (in this case, the pancreas) and background. By minimizing the Dice loss, the second computing unit 307 continuously optimizes its prediction ability, and finally can accurately segment the pancreatic area from the original abdominal CT image. Since CT images are usually three-dimensional, but here we are discussing a two-dimensional processing process (which may be layer-by-layer processing of three-dimensional images), the final result is a series of two-dimensional pancreatic images, which together constitute a three-dimensional representation of the pancreas. However, it should be noted that it is actually possible to construct a 3D convolutional neural network to directly process 3D CT images, thereby more accurately segmenting the 3D pancreatic structure.

[0126] As an optional implementation, in an embodiment of the present invention, traditional convolutional neural networks may face problems of insufficient feature extraction or information loss when processing complex medical images. The prediction system combines residual connections with attention gating modules, which not only retains the advantages brought by the depth of the network, but also enhances the network's attention to key areas through the attention mechanism, and can more effectively extract features from the original abdominal CT images, especially for the pancreas, which may appear complex in the image and have low contrast with surrounding tissues. Organs. Residual connections help keep information flowing in deep networks and reduce information loss, while the attention gating module enhances the network's attention to the pancreatic area and reduces the interference of background noise. These have significantly improved the accuracy of pancreatic segmentation.

[0127] As an optional implementation, in the present invention, residual connections not only improve the performance of the model, but also accelerate the training process by allowing gradients to flow directly through these connections. This solves the common gradient vanishing problem in deep networks, allowing the model to converge to the optimal solution faster.

[0128] As an optional implementation, in the embodiment of the present invention, the Sigmoid function normalizes the output to a probability value, making the model prediction smoother and helping to deal with uncertainty and noise in the image. In addition, the Dice loss function has good robustness to the imbalance problem of foreground and background, further enhancing the performance of the model in complex scenes.

[0129] As an optional implementation, in the embodiment of the present invention, the entire prediction process is highly automated, from image input to output of pancreatic image sequence, without the need for human intervention, which greatly improves the efficiency and accuracy of medical diagnosis.

[0130] and Figure 3 Compared with the prediction system, Figure 4 The drawing unit 302 includes:

[0131] The determination subunit 3021 is used to determine the two rectangular vertices connected by the edge where each 3D rectangular cube equivalent point is located; wherein the 3D rectangular cube equivalent point is a point on the 3D rectangular cube whose grayscale value is equal to a specified threshold.

[0132] As an optional implementation, in an embodiment of the present invention, the system may first read three-dimensional image data, which is stacked by a series of two-dimensional images (slices), and the pixel points on each slice correspond to the grayscale value of a cube in three-dimensional space. Subsequently, for each cube in the three-dimensional cube, the system needs to access and extract the grayscale values ​​of its eight vertices (or corner points). These vertices are located at the four corners of the cube and correspond to the intersection of the three-dimensional space grid. Next, the system applies a specified grayscale threshold to distinguish different tissues or structures. This threshold is determined based on prior knowledge or an automatic algorithm (such as the Otsu method) to distinguish between tissues of interest (such as the pancreas) and background or other tissues. Subsequently, the system may traverse the vertices of all three-dimensional cubes, and for each vertex, check whether its grayscale value is on both sides of the specified threshold. If the grayscale values ​​of two adjacent vertices of a cube (directly connected in three-dimensional space) are respectively on both sides of the threshold, the two vertices (and the cubes in which they are located) are determined as vertices of the target cube or part of the target cube. This step is usually used for edge detection or interface recognition, especially when it is necessary to distinguish the boundaries of different tissues or structures. Once the target cube vertices are determined, the prediction system can further process these vertices or the cubes to which they belong to perform more complex tasks, such as 3D reconstruction, volume calculation, shape analysis, etc. Finally, the system may output the processing results in a visual form, such as a 3D rendered image, a slice view, or an analysis report, to facilitate diagnosis or further analysis by medical experts.

[0133] As an optional implementation, in an embodiment of the present invention, the traditional threshold segmentation method usually only considers the gray value of the pixel (or cube) and ignores its spatial position information. However, this method actually combines local gray information and spatial position information by determining two adjacent cube vertices located on both sides of the specified threshold as the target cube vertices, making the segmentation result more accurate and reliable, and can more accurately identify the boundaries of different tissues or structures. This method takes into account the continuity between cubes and the local changes in gray values, which helps to reduce the over-segmentation or under-segmentation caused by single threshold processing.

[0134] As an optional implementation, in the embodiment of the present invention, two adjacent cube vertices located on both sides of the specified threshold are usually located at the junction of different tissues or structures, that is, the edge area. Determining these vertices as target cube vertices is actually performing edge detection. This helps to more accurately reconstruct the three-dimensional shape of the tissue in subsequent processing and improve the reliability of image analysis.

[0135] As an alternative implementation, in the embodiments of the present invention, the entire process is completely automatically completed by the prediction system without manual intervention, greatly improving the automation level of medical image processing. This can not only save human resources, but also reduce human errors and improve the accuracy and efficiency of diagnosis.

[0136] A judgment subunit 3022, configured to, after calculating the isopoint coordinates of each three-dimensional rectangular cube by using a linear interpolation algorithm, judge whether there are N isopoint coordinates in each three-dimensional rectangular cube.

[0137] As an alternative implementation, in the embodiments of the present invention, in each three-dimensional cube, points with a gray value equal to a specified threshold are found, that is, the isopoints of the three-dimensional rectangular cube. These points usually represent the boundary between the pancreas and the surrounding tissues. Since the gray values in the actual data change continuously, the isopoints often do not exactly fall on a certain vertex, but are located on a certain edge of the cube. When the isopoint is located on a certain edge of the cube, we can perform linear interpolation through the gray values and coordinates of the two vertices on this edge to calculate the exact coordinates of the isopoint.

[0138] As an alternative implementation, in the embodiments of the present invention, assume that we have found a three-dimensional rectangular cube, and the gray values of the two vertices on a certain edge are G1 and G2 (G1 < Gthreshold < G2, where Gthreshold is the specified threshold), and the coordinates of these two vertices are (x1, y1, z1) and (x2, y2, z2) respectively. First, the system can first use the gray value to calculate the interpolation ratio t such that G1 + t(G2 - G1) = Gthreshold, and then solve the equation t = (Gthreshold - G1) / (G2 - G1) to obtain the result. Subsequently, the system uses the obtained interpolation ratio t to calculate the coordinates (x, y, z) of the isopoint, that is, solve the system of equations: x = x1 + t(x2 - x1), y = y1 + t(y2 - y1), z = z1 + t(z2 - z1).

[0139] As an alternative implementation, in the embodiments of the present invention, in the visualization processes such as isosurface extraction and three-dimensional reconstruction of the present application, using the exact isopoints of the three-dimensional rectangular cube can generate a smoother and more continuous isosurface, reduce the jagged edges or holes caused by data simplification, and improve the visual effect.

[0140] As an optional implementation, in the embodiment of the present invention, in the three-dimensional rectangular cube, the judgment subunit 3022 determines the two rectangular vertices connected by the edge where the three-dimensional rectangular cube is located, and calculates the coordinates of the three-dimensional rectangular cube isovalued point by using a linear interpolation algorithm. This method is more accurate than simple threshold segmentation. It can more accurately locate the boundary between the pancreas and the surrounding tissues, providing a reliable basis for subsequent three-dimensional reconstruction and feature extraction.

[0141] The connecting subunit 3023 is used to connect all the coordinates of the equal-value points in a three-dimensional rectangular cube by using a triangulation algorithm to form an equal-value point triangular face when the judging subunit 3022 determines that there are N equal-value point coordinates in each three-dimensional rectangular cube; wherein N is a natural number not less than 3.

[0142] As an optional implementation, in the embodiment of the present invention, triangulation is the process of connecting a set of points (here, 3D rectangular cube equivalent points) into a series of non-overlapping triangles, which together cover the entire isosurface. In three-dimensional space, there are a variety of triangulation algorithms to choose from, such as Delaunay triangulation, MarchingCubes algorithm (although Marching Cubes directly generates isosurfaces within voxels instead of just points, but the concept is similar), etc.

[0143] As an optional implementation, in the embodiment of the present invention, first, the connection subunit 3023 can select a suitable triangulation algorithm according to specific needs and data characteristics. Subsequently, the connection subunit 3023 can use the three-dimensional rectangular cube equivalent points as input, and apply the triangulation algorithm to generate triangular faces, each of which is defined by three equivalent points, thereby generating a triangular mesh representation of the equivalent surface. Finally, the connection subunit 3023 calculates the triangular face normal vector for the equivalent point triangular face. The triangular face normal vector is a vector perpendicular to the triangular face, which is very important for lighting calculation, surface direction analysis, etc., and can provide a basis for subsequent image rendering, analysis and visualization.

[0144] The calculation subunit 3024 is used to calculate the cross product of two edges of each equal-value point triangle to obtain the triangle normal vector, and normalize each triangle normal vector.

[0145] As an optional implementation, in the embodiment of the present invention, the present application combines the judgment of equivalent points with the triangulation algorithm to form a new three-dimensional data processing method. This method not only considers the existence of equivalent points, but also uses these points to construct triangular surfaces with actual geometric significance, providing richer information for subsequent analysis.

[0146] As an optional implementation, in the embodiment of the present invention, the calculation subunit 3024 calculates the cross product of two edges in the triangle and normalizes the normal vector to ensure the accuracy and consistency of the normal vector of each triangle. This precise calculation method helps to improve the quality of subsequent graphics rendering, lighting simulation and other tasks.

[0147] As an optional implementation, in the embodiment of the present invention, the process integrates multiple technologies such as equivalent point judgment, triangulation, cross product calculation and normal vector normalization to form an innovative three-dimensional data processing method. This technical integration not only improves the accuracy and efficiency of data processing, but also expands the application scope of three-dimensional data analysis.

[0148] As an optional implementation, in the embodiment of the present invention, the connection subunit 3023 is also used to connect adjacent equal-value point triangles corresponding to triangle normal vectors with the same direction, and perform three-dimensional visualization processing to draw a three-dimensional visualized pancreas model.

[0149] As an optional implementation, in an embodiment of the present invention, the prediction system first uses a spatial index (such as an octree, a kd tree, or an R tree, etc.) to quickly find the adjacent triangles in space for each equal-value point triangle. This method can significantly reduce unnecessary comparisons and improve search efficiency. For each triangle, the system can calculate its normal vector and compare it with the normal vectors of the adjacent triangles. If the normal vectors of two adjacent triangles have the same direction (or are nearly the same, considering a certain tolerance range), the two triangles are considered to have the "same direction" and are marked as connectable candidates. Based on the matching results of the previous step, the system can connect adjacent triangles with the same direction. This usually involves updating the vertex list and edge list of the triangle to ensure that the topological structure of the entire model is correct. After completing the connection of the triangles, the system can pass these triangle data to the three-dimensional visualization engine. The visualization engine uses this data to render the three-dimensional model, including applying effects such as lighting, materials, and textures to generate a realistic three-dimensional visualization pancreas model.

[0150] As an optional implementation, in an embodiment of the present invention, by ensuring that only adjacent triangles with the same orientation are connected, the system can generate a more accurate three-dimensional model. This reduces the distortion or deformation of the model caused by incorrect connections. Reasonable triangle connection can optimize the topological structure of the model, making it more suitable for subsequent processing and analysis. For example, in medical imaging, the optimized model can be better used for volume measurement, shape analysis or pathological diagnosis.

[0151] As an optional implementation, in an embodiment of the present invention, three-dimensional visualization processing enables the pancreas model to be presented to the user in an intuitive and realistic manner. This not only helps medical experts to make more accurate diagnoses, but can also be used for teaching and scientific research purposes. In a three-dimensional visualization environment, users can rotate, scale, and slice the model to observe and analyze the model from different angles. This interactivity enhances the user's understanding and cognition of the model.

[0152] As an optional implementation, in an embodiment of the present invention, the present application may first store the vertex coordinates of all triangles in a buffer so that the GPU can quickly access them. If the triangles share vertices, the system can use an index buffer to reduce the duplication of vertex data and improve rendering efficiency. Subsequently, the system may store the normal vector of each triangle in a buffer for lighting and shadow calculations. Finally, the system may also assign appropriate materials and textures to the pancreas model to enhance the visual effect, so as to complete the rendering preparation of the three-dimensional visualized pancreas model.

[0153] As an optional implementation, in the embodiment of the present invention, in order to make the three-dimensional visualized pancreas model more vivid and useful, the system can add interactivity and animation effects. That is, user interaction: allowing the user to rotate, zoom, and translate the model to observe from different angles. Animation: If the model is based on time series data (such as dynamic CT scans), animation can be added to show the changes of the pancreas over time.

[0154] and Figure 3 Compared with the prediction system, Figure 4 The determination subunit 3021 includes:

[0155] The acquisition module 30211 is used to obtain the grayscale value of each rectangular vertex in the three-dimensional rectangular cube.

[0156] The determination module 30212 is used to determine the rectangular vertices corresponding to the grayscale values ​​of two adjacent rectangular vertices located on both sides of the specified threshold as target vertices.

[0157] It can be seen that the implementation Figure 4 Another prediction system described is capable of improving the accuracy and reliability of pancreatic cancer disease prediction.

[0158] In addition, implementation Figure 4 Another prediction system is described that can improve the efficiency and accuracy of model building.

[0159] Embodiment 5

[0160] See also Figure 5 , Figure 5 FIG. 1 is a schematic diagram of another prediction system disclosed in an embodiment of the present invention. Figure 5, the prediction system may include:

[0161] A memory 501 storing executable program codes;

[0162] a processor 502 coupled to the memory 501;

[0163] The processor 502 calls the executable program code stored in the memory 501 to execute Figure 1-2 Any pancreatic cancer prediction method based on 3D visualization.

[0164] The embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute Figure 1-2 Any pancreatic cancer prediction method based on 3D visualization.

[0165] The embodiments of the present invention further disclose a computer program product, wherein when the computer program product is run on a computer, the computer is enabled to execute part or all of the steps of the methods in the above method embodiments.

[0166] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0167] The above is a detailed introduction to a pancreatic cancer prediction method and prediction system based on three-dimensional visualization disclosed in an embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A pancreatic cancer prediction method based on three-dimensional visualization, characterized in that: include: The two-dimensional pancreatic image sequence is three-dimensionally stacked according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes; According to the three-dimensional rectangular cube, a three-dimensional visualized pancreas model is drawn; Substituting the CT image feature data and the radiomics feature data obtained in the three-dimensional visualized pancreatic model into the radiomics prediction model to calculate the prediction probability of pancreatic cancer; Among them, the imaging omics prediction model includes imaging omics feature scores, intercept feature scores, cystic change feature scores, feature scores of pancreatic duct / pancreas diameter ratio ≥ 0.34, pancreatic duct perforation sign feature scores, pancreatic portal hypertension feature scores and arterial phase CT value feature scores. The imaging omics feature scores include pancreatic region shape flatness scores, pancreatic region abnormally low pixel intensity scores, pancreatic region abnormally high pixel intensity scores, pancreatic region average pixel intensity scores, pancreatic region average pixel intensity scores on a coarse scale and pancreatic region texture scores. The imaging omics feature score is the sum of the scores of each item in the imaging omics feature score, and the predicted probability is the sum of the scores of each item in the imaging omics prediction model.

2. The method according to claim 1, characterized in that Before stacking the two-dimensional pancreatic image sequence in three-dimensional order according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes, the method further includes: Using the lasso regression formula, the radiomics feature data training set is analyzed to determine the weights of each item in the radiomics feature score; The multivariate logistic regression analysis method was used to analyze the CT image feature data training set to determine the scoring weights of each item in the imaging omics prediction model.

3. The method according to claim 2, characterized in that Before stacking the two-dimensional pancreatic image sequence in three dimensions according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes, and after analyzing the CT image feature data training set using a multivariate logistic regression analysis method to determine the scoring weights of each item in the radiomics prediction model, the method further includes: Inputting the original abdominal CT image into a convolutional neural network with a residual connection and an attention gating module to obtain a residual convolution block; The output residual convolution block is normalized using a Sigmoid function, and a pixel probability map is generated to obtain the probability that each pixel in the original abdominal CT image belongs to the pancreas; The Dice loss function is used to perform error calculation on the pixel probability map to determine the two-dimensional pancreatic image sequence.

4. The method according to any one of claims 1 to 3, characterized in that: Drawing a three-dimensional visualized pancreas model according to the three-dimensional rectangular cube includes: Determine two rectangular vertices connected by the edge of each three-dimensional rectangular cube equivalent point; wherein the three-dimensional rectangular cube equivalent point is a point on the three-dimensional rectangular cube whose grayscale value is equal to a specified threshold; After calculating the coordinates of the equivalent points of each three-dimensional rectangular cube by using a linear interpolation algorithm, determine whether there are N equivalent point coordinates in each three-dimensional rectangular cube; if so, connect all the equivalent point coordinates in one three-dimensional rectangular cube by using a triangulation algorithm to form an equivalent point triangular face; wherein N is a natural number not less than 3; Calculate the cross product of two edges of each of the equal-value point triangles to obtain a normal vector of the triangle, and normalize each of the normal vectors of the triangle; The adjacent triangular faces of equal value points corresponding to the triangular face normal vectors with the same direction are connected, and three-dimensional visualization processing is performed to draw the three-dimensional visualized pancreas model.

5. The method according to claim 4, characterized in that The step of determining two rectangular vertices connected by the edge where each 3D rectangular cube equivalent point is located includes: Obtaining the grayscale value of each rectangular vertex in the three-dimensional rectangular cube; The rectangular vertices corresponding to the grayscale values ​​of two adjacent rectangular vertices located on both sides of the specified threshold are determined as target vertices.

6. A prediction system, characterized in that: The prediction system comprises: A three-dimensional stacking unit, used for three-dimensionally stacking the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes; A drawing unit, used for drawing a three-dimensional visualized pancreas model according to the three-dimensional rectangular cube; A first calculation unit, used for substituting the CT image feature data and the radiomics feature data obtained in the three-dimensional visualized pancreatic model into the radiomics prediction model to calculate the prediction probability of pancreatic cancer; Among them, the imaging omics prediction model includes imaging omics feature scores, intercept feature scores, cystic change feature scores, feature scores of pancreatic duct / pancreas diameter ratio ≥ 0.34, pancreatic duct perforation sign feature scores, pancreatic portal hypertension feature scores and arterial phase CT value feature scores. The imaging omics feature scores include pancreatic region shape flatness scores, pancreatic region abnormally low pixel intensity scores, pancreatic region abnormally high pixel intensity scores, pancreatic region average pixel intensity scores, pancreatic region average pixel intensity scores on a coarse scale and pancreatic region texture scores. The imaging omics feature score is the sum of the scores of each item in the imaging omics feature score, and the predicted probability is the sum of the scores of each item in the imaging omics prediction model.

7. The prediction system according to claim 6, characterized in that The prediction system also includes: an analysis unit, used for analyzing the radiomics feature data training set using a lasso regression formula to determine the weights of each item in the radiomics feature score before the three-dimensional stacking unit three-dimensionally stacks the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes; The analysis unit is also used to analyze the CT image feature data training set using a multivariate logistic regression analysis method to determine the scoring weights of each item in the imaging omics prediction model.

8. The prediction system according to claim 7, characterized in that The prediction system also includes: An input unit is used for inputting the original abdominal CT image into a convolutional neural network with a residual connection and an attention gating module to obtain a residual convolution block after the three-dimensional stacking unit three-dimensionally stacks the two-dimensional pancreatic image sequence according to the depth order in the original abdominal CT image to form a plurality of three-dimensional rectangular cubes, and after the analysis unit analyzes the CT image feature data training set using a multivariate logistic regression analysis method to determine the weights of each score in the imaging omics prediction model; A processing unit, configured to normalize the output residual convolution block using a Sigmoid function and generate a pixel probability map to obtain the probability that each pixel in the original abdominal CT image belongs to the pancreas; The second calculation unit is used to perform error calculation on the pixel probability map by using a Dice loss function to determine the two-dimensional pancreatic image sequence.

9. A prediction system, characterized in that: The prediction system comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the pancreatic cancer prediction method based on three-dimensional visualization according to any one of claims 1 to 5.

Images