A method and system for estimating relative blood flow based on a pancreatic plain CT image

Abstract

The present application relates to the technical field of medical image processing, in particular to a relative blood flow estimation method and system based on a pancreatic plain CT image, the method comprising: constructing a pancreatic enhanced CT estimation model, and obtaining a brightness estimation module based on pre-training of the pancreatic enhanced CT estimation model; obtaining a target plain CT image; obtaining predicted arterial phase CT image brightness encoding, portal vein phase CT image brightness encoding and venous phase CT image brightness encoding based on the target plain CT image and the brightness estimation module; constructing a target tetrahedron structure in a high-dimensional space based on CT image brightness encoding in different phases; obtaining a target relative blood flow based on the ratio of the volume of the target tetrahedron structure to the volume of a reference tetrahedron structure; determining whether the target relative blood flow exceeds a threshold value, and classifying the subject based on the determination result. The present application provides effective assistance for early screening of pancreatic cancer based on processing of the plain CT image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a relative blood flow estimation method and system based on pancreatic plain CT images. BACKGROUND

[0002] As a highly malignant tumor, pancreatic cancer often has no obvious early symptoms, resulting in most patients being in the advanced stage of the disease when diagnosed, which greatly increases the difficulty of treatment and the mortality rate of patients. Therefore, improving the early detection rate of pancreatic cancer is of great significance for improving the prognosis of patients. Among current medical diagnostic methods, CT contrast-enhanced imaging technology provides important anatomical and functional information for the diagnosis of pancreatic cancer. CT scanning after intravenous injection of contrast agent can clearly show the blood flow of the pancreas and its surrounding tissues, helping doctors more accurately determine the presence of tumors and their properties. This technology not only improves the possibility of early detection of pancreatic cancer, but also provides valuable reference for clinical decision-making.

[0003] However, although CT contrast-enhanced imaging technology has important application value in the diagnosis of pancreatic cancer, there are still some deficiencies and limitations. First, this technology usually requires a certain prediction of pancreatic cancer in order to make necessary preparations and planning before scanning. However, due to the lack of obvious early symptoms of pancreatic cancer, doctors and patients often have difficulty making accurate predictions in the early stages of the disease, limiting the widespread application of CT contrast-enhanced imaging technology. CT contrast-enhanced imaging technology also requires patients to prepare, such as fasting, injection of contrast agent, and then CT scanning. This process is relatively cumbersome, not only increasing the discomfort of patients, but also possibly causing some patients to be unable to scan without preparation, thus missing the opportunity to detect the disease early. CT contrast-enhanced imaging technology also requires a large amount of medical resources, including CT equipment, medical staff, etc. In the case of tight medical resources, the widespread application of this technology may have some impact on the diagnosis and treatment of other patients.

[0004] At the same time, there is currently no pancreatic relative blood flow estimation tool, and doctors can only rely on image recognition technology to indirectly determine the blood flow of the pancreas. This method is not only time-consuming and labor-intensive, but also relatively low in accuracy, making it difficult to meet clinical needs.

[0005] Therefore, developing a new technology that can estimate contrast intensity using plain CT results through machine learning and further provide estimated blood flow is of great significance for improving the early detection rate and treatment effect of pancreatic cancer. SUMMARY

[0006] The present application aims to solve at least one of the above technical problems by providing a relative blood flow estimation method and system based on pancreatic plain CT images.

[0007] The application achieves the above-mentioned purpose by the following technical solutions:

[0008] A relative blood flow estimation method based on a pancreatic plain CT image, comprising the following steps:

[0009] A pancreatic enhanced CT estimation model is constructed, and a brightness estimation module is obtained by pre-training based on the pancreatic enhanced CT estimation model;

[0010] A pancreatic plain CT image of a subject, i.e., a target plain CT image, is obtained;

[0011] Based on the target plain CT image and the brightness estimation module, a predicted arterial phase CT image brightness encoding, a predicted portal vein phase CT image brightness encoding, and a predicted venous phase CT image brightness encoding are obtained;

[0012] The target plain CT image brightness encoding is defined as a reference point, and a target tetrahedron structure is constructed in a high-dimensional space based on the CT image brightness encodings in different phases;

[0013] A target relative blood flow is obtained based on a ratio of a volume of the target tetrahedron structure to a volume of a reference tetrahedron structure; the volume of the reference tetrahedron structure is an average value of the volumes of the tetrahedron structures formed by the CT image brightness encodings of a healthy pancreas in different phases in the high-dimensional space;

[0014] It is determined whether the target relative blood flow exceeds a threshold value, and the subject is classified based on the determination result.

[0015] Further, the pancreatic enhanced CT estimation model comprises a first training model and a second training model;

[0016] The first training model comprises an image encoding module, a splicing module, and a first image decoding module; plain CT image data is input into the image encoding module, the input of the splicing module is the output of the image encoding module and a plain CT image brightness encoding, and the data obtained by processing the splicing module is input into the first image decoding module to finally obtain an enhanced CT image; wherein the plain CT image brightness encoding serves as a reference point;

[0017] The second training model comprises an alignment module, an image encoding module, a brightness estimation module, a splicing module, and a second image decoding module; plain CT image data is input into the alignment module and the brightness estimation module respectively, the output of the alignment module is processed by the image encoding module, the output of the image encoding module and the output of the brightness estimation module are jointly input into the splicing module, and the output of the splicing module is processed by the second image decoding module to obtain a corresponding enhanced CT image;

[0018] The image encoding module and the splicing module in the second training model are obtained by training the first training model and remain unchanged in the second training model; initial parameters of the second image decoding module adopt parameters obtained by final training of the first image decoding module; and parameters of the second image decoding module are updated by training of the second training model.

[0019] Further, the brightness estimation module comprises an arterial phase brightness estimation unit, a portal vein phase brightness estimation unit and a venous phase brightness estimation unit.

[0020] Each unit is respectively used to predict the brightness relationship corresponding to the plain CT image and the enhanced CT image of each phase.

[0021] Further, the threshold is set by the following method:

[0022] Based on the data set for training the pancreatic enhanced CT estimation model, the relative blood flow values of all subjects are obtained;

[0023] The relative blood flow values of all subjects are fitted by Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold = 1 + .

[0024] Further, based on the judgment result, the subject is classified, comprising:

[0025] When the target relative blood flow does not exceed the threshold, the subject is classified as a low-risk population of pancreatic cancer;

[0026] When the target relative blood flow exceeds the threshold, the subject is classified as a high-risk population of pancreatic cancer, and the high-risk population of pancreatic cancer is further examined.

[0027] A relative blood flow estimation system based on a pancreatic plain CT image, comprising:

[0028] A construction module is configured to construct a pancreatic enhanced CT estimation model and pre-train a brightness estimation module based on the pancreatic enhanced CT estimation model;

[0029] An acquisition module is configured to acquire a pancreatic plain CT image of a subject, i.e., a target plain CT image;

[0030] A prediction module is configured to acquire predicted arterial phase CT image brightness encoding, portal vein phase CT image brightness encoding and venous phase CT image brightness encoding based on the target plain CT image and the brightness estimation module.

[0031] The computing module is configured to define the target plain CT image intensity encoding as a reference point, construct a target tetrahedron structure in a high-dimensional space based on the CT image intensity encoding in different phases, and obtain a target relative blood flow based on a ratio of a target tetrahedron structure volume to a reference tetrahedron structure volume.

[0032] The computing module is further configured to obtain the target relative blood flow based on a ratio of the target tetrahedron structure volume to a reference tetrahedron structure volume; wherein the reference tetrahedron structure volume is an average value of a tetrahedron structure volume formed by the CT image intensity encoding in different phases of a healthy pancreas in the high-dimensional space.

[0033] The judging module is configured to judge whether the target relative blood flow exceeds a threshold value, and classify the subject based on a judgment result.

[0034] Further, in the constructing module, the pancreatic enhanced CT estimation model comprises a first training model and a second training model.

[0035] The first training model comprises an image encoding module, a splicing module and a first image decoding module; the plain CT image data is input into the image encoding module, the input of the splicing module is the output of the image encoding module and the plain CT image intensity encoding, and the data obtained by processing the splicing module is input into the first image decoding module to finally obtain the enhanced CT image; wherein the plain CT image intensity encoding serves as a reference point.

[0036] The second training model comprises an alignment module, an image encoding module, a brightness estimation module, a splicing module and a second image decoding module; the plain CT image data is input into the alignment module and the brightness estimation module respectively, the output of the alignment module is processed by the image encoding module, the output of the image encoding module and the output of the brightness estimation module are jointly input into the splicing module, and the output of the splicing module is processed by the second image decoding module to obtain the corresponding enhanced CT image.

[0037] The image encoding module and the splicing module in the second training model are obtained by training the first training model and remain unchanged in the second training model; the initial parameters of the second image decoding module adopt the parameters finally obtained by training the first image decoding module; the parameters of the second image decoding module are updated by training the second training model.

[0038] Further, in the constructing module, the brightness estimation module comprises an arterial phase brightness estimation unit, a portal venous phase brightness estimation unit and a venous phase brightness estimation unit.

[0039] Each unit is respectively configured to predict the brightness relationship between the plain CT image and the enhanced CT image in each phase.

[0040] Further, in the judging module, the threshold is set by the following method:

[0041] Based on the data set for training the pancreatic enhancement CT estimation model, the relative blood flow values of all subjects are obtained;

[0042] The relative blood flow values of all subjects are fitted by Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold = 1 + .

[0043] Further, in the judging module, the subjects are classified based on the judgment result, comprising:

[0044] When the target relative blood flow does not exceed the threshold, the subject is classified as a low risk group of pancreatic cancer;

[0045] When the target relative blood flow exceeds the threshold, the subject is classified as a high risk group of pancreatic cancer, and further pancreatic examination is performed on the high risk group of pancreatic cancer.

[0046] The beneficial effects of the present application are:

[0047] Firstly, for the early detection problem of pancreatic cancer, the present application proposes an innovative solution, which can realize the prediction of enhanced CT effect through plain CT image, and further evaluate the relative blood flow of patients, providing a new and convenient means for early detection of pancreatic cancer. The present application not only fills the gap of pancreatic cancer indication method, but also greatly improves the detection rate of pancreatic cancer.

[0048] Secondly, the implementation of the present application is synchronized with plain CT scanning, without additional examination procedures or patient preparation, greatly reducing the discomfort of patients and saving medical resources. Doctors can obtain the risk prompt of pancreatic cancer while the patient is receiving routine CT scanning (plain CT), so as to take further diagnostic or therapeutic measures in time.

[0049] Thirdly, the present application assists doctors to quickly and directly predict pancreatic cancer through pancreatic CT scan image, making the use of enhanced contrast more targeted. This not only improves the accuracy of diagnosis, but also avoids unnecessary use of contrast agent, reducing the risk of patients.

[0050] Finally, the model used in the present application is designed to be simple and efficient, and can be run on ordinary computers or mobile phone platforms without large-scale computing support. This feature makes the application of the present application more widely used, and is convenient for popularization and use in medical institutions at all levels, bringing blessings to more patients.

[0051] In summary, the application not only improves the early detection rate of pancreatic cancer, but also provides a convenient and accurate diagnostic tool for doctors, saves medical resources, reduces the burden on patients, and has significant social benefits and clinical value. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The pancreatic enhancement CT image brightness of a certain subject changes from low to high throughout the process image;

[0053] Figure 2 The flow chart of the relative blood flow estimation method based on the pancreatic plain CT image of an embodiment of the application is shown in the figure;

[0054] Figure 3 The schematic diagram of constructing a target tetrahedral structure in a high-dimensional space according to an embodiment of the application is shown in the figure;

[0055] Figure 4 The first training model structure schematic diagram according to an embodiment of the application is shown in the figure;

[0056] Figure 5 The second training model structure schematic diagram according to an embodiment of the application is shown in the figure;

[0057] Figure 6 The image encoding module workflow schematic diagram according to an embodiment of the application is shown in the figure;

[0058] Figure 7 The image decoding module workflow schematic diagram according to an embodiment of the application is shown in the figure;

[0059] Figure 8 The brightness estimation module workflow schematic diagram according to an embodiment of the application is shown in the figure;

[0060] Figure 9 The system structure diagram of the relative blood flow estimation system based on the pancreatic plain CT image according to an embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0061] The content of the application will now be discussed with reference to exemplary embodiments. It should be understood that the discussed embodiments are only to enable those of ordinary skill in the art to better understand and thus implement the content of the application, and are not intended to imply any limitation on the scope of the application.

[0062] As used herein, the term "comprising" and variations thereof are to be construed as meaning "including but not limited to". The term "based on" is to be construed as "based at least in part on". The terms "one embodiment" and "an embodiment" are to be construed as "at least one embodiment".

[0063] During the process of pancreatic enhanced CT image scanning, contrast agent needs to be injected. According to the metabolic stage of the contrast agent in the blood vessels, it can be divided into the plain scan phase, the arterial phase, the portal vein phase and the venous phase. The enhanced CT images of different phases have different characteristics in brightness. Figure 1 Fig. 1 is a flowchart of a relative blood flow estimation method based on a pancreatic plain scan CT image according to an embodiment of the present application. Figure 1 As shown in Fig. 1, the brightness of the enhanced CT image reflects the concentration of the contrast agent, and the concentration of the contrast agent is positively correlated with the local blood flow. For example, in the arterial phase, the brightness of the enhanced CT image mainly reflects the arterial blood supply, and the cancerous region may have lower brightness than normal due to abnormal blood vessels. In the portal vein phase / venous phase, the brightness of the enhanced CT image reflects the capillary permeability and tissue interstitial enhancement, and the high pressure of the cancerous interstitial may inhibit the enhancement. Taking the brightness of the CT image in the plain scan phase as the baseline, the influence of the individual tissue density difference is eliminated. Compared with the uniform enhancement of the healthy pancreas, the CT image of the pancreatic cancer patient has the characteristics of less blood supply, insufficient arterial phase enhancement and delayed portal vein phase enhancement.

[0064] Based on the above theory, the present application uses an innovative model construction method. First, the pancreatic plain scan CT image of the subject is used to predict the four-phase images of the enhanced CT. Then, based on the brightness coding information in different phases, a high-dimensional geometric model (specifically a tetrahedral structure) is constructed. By calculating the volume change of these models under different blood flow states, the present application can quantify the blood flow difference between the cancerous pancreatic tissue and the normal pancreatic tissue. Based on this quantitative result, we can effectively classify the subjects, identify those individuals with relatively high blood flow exceeding the preset threshold as a high-risk group, and then implement more detailed pancreatic cancer risk assessment and corresponding medical examination on them.

[0065] Embodiment one

[0066] Figure 2 Fig. 2 is a flowchart of a relative blood flow estimation method based on a pancreatic plain scan CT image according to an embodiment of the present application. Figure 3 Fig. 3 is a schematic diagram of constructing a target tetrahedral structure in a high-dimensional space according to an embodiment of the present application. Figures 2-3 As shown in Fig. 3, according to an embodiment of the present application, a relative blood flow estimation method based on a pancreatic plain scan CT image includes the following steps:

[0067] Step S102, constructing a pancreatic enhanced CT estimation model and obtaining a brightness estimation module based on the pre-training of the pancreatic enhanced CT estimation model;

[0068] Step S104, obtaining the pancreatic plain scan CT image of the subject, i.e., the target plain scan CT image;

[0069] Step S106, based on the target plain scan CT image and the brightness estimation module, the predicted arterial phase CT image brightness encoding, portal vein phase CT image brightness encoding, venous phase CT image brightness encoding are obtained;

[0070] Step S108, the target plain scan CT image brightness encoding is defined as the reference point, and in this embodiment, the reference point is "00000000". Based on the CT image brightness encoding in different phases, a target tetrahedral structure is constructed in a high-dimensional space.

[0071] Step S110, the target relative blood flow is obtained based on the ratio of the target tetrahedral structure volume to the reference tetrahedral structure volume; wherein the reference tetrahedral structure volume is the average value of the tetrahedral structure volume formed by the CT image brightness encoding of the healthy pancreas in different phases in the high-dimensional space;

[0072] Step S112, whether the target relative blood flow exceeds the threshold value is judged, and the subject is classified based on the judgment result.

[0073] In this embodiment, a relative blood flow estimation method based on pancreas plain scan CT image is proposed, and the detailed implementation steps are as follows: first, in step S102, a pancreas enhanced CT estimation model is constructed, and a brightness estimation module is obtained by pre-training based on the model. This module is the core of subsequent analysis, which can predict the brightness characteristics of enhanced CT in different phases (arterial phase t1, portal vein phase t2, venous phase t3) according to the pancreas plain scan CT image. During the training process of the pancreas enhanced CT estimation model, a large number of pancreas CT image data are collected in advance, including the pancreas plain scan CT image and the pancreas enhanced CT image set of the same subject, a pancreas CT image data set is constructed, and the size of the data set is set according to actual requirements, for example, 500 cases of CT images of subjects are used to construct the pancreas CT image data set. The pancreas enhanced CT estimation model is trained based on the pre-constructed pancreas CT image data set, and the brightness estimation module is obtained.

[0074] Then, in step S104, the pancreas plain scan CT image of the subject, i.e. the target plain scan CT image, is obtained. This is the starting point of the whole analysis process, which provides the basic image information of the pancreas in the unenhanced state.

[0075] Then, in step S106, the pre-trained brightness estimation module is used to predict the CT image brightness encoding in arterial phase, portal vein phase and venous phase based on the target plain scan CT image. This brightness estimation module can capture the influence of enhanced CT contrast agent on image brightness at different time stages in the human body, so as to simulate the brightness characteristics of enhanced CT, and provide basic data for subsequent construction of tetrahedral structure.

[0076] In step S108, the target plain scan CT image intensity encoding t0 is defined as the reference point "00000000", and combined with the predicted CT image intensity encoding of different phases, a target tetrahedron structure is constructed in high-dimensional space. The tetrahedron structure takes the intensity encodings of the plain scan phase and different enhanced phases as the vertices, reflecting the dynamic process of the pancreatic blood flow over time. Each vertex of the tetrahedron represents the pancreatic intensity characteristics at a specific time point, and the volume of the tetrahedron can be regarded as a comprehensive measure of these intensity characteristics in high-dimensional space.

[0077] For example, the CT image intensity encoding corresponding to each phase is mapped into N-dimensional space to obtain non-coplanar coordinate points P0, P1, P2, P3, and an N-dimensional space tetrahedron is constructed based on the coordinates of the CT image intensity encodings in N-dimensional space. N is the number of bits of the encoder. The volume calculation process of the tetrahedron is as follows:

[0078] Step one, taking point P0(t0, "00000000") as the reference point, a vector starting from P0 is constructed;

[0079]

[0080]

[0081]

[0082] Step two, construct a matrix;

[0083] The vector obtained in step one is used as a column vector (or a row vector) to form a 3x3 square matrix A;

[0084]

[0085] Step three, calculate the volume V of the tetrahedron in N-dimensional space;

[0086]

[0087] Where det represents the determinant.

[0088] Subsequently, in step S110, the target relative blood flow is obtained by comparing the volume ratio of the target tetrahedron structure and the reference tetrahedron structure. The volume of the reference tetrahedron is the average volume of the tetrahedron structure formed by the CT image intensity encodings of the healthy pancreas in different phases in high-dimensional space, which represents a normal blood flow range. By comparing the volume ratio of the target tetrahedron and the reference tetrahedron, we can indirectly reflect the relative change of blood flow, thereby evaluating the blood flow condition of the pancreas.

[0089] Further, in step S112, it is determined whether the target relative blood flow exceeds a preset threshold. The threshold is based on a large amount of pancreatic data statistics, and is used to distinguish the normal and abnormal blood flow range. If the target relative blood flow exceeds the threshold, it indicates that the pancreatic blood flow of the subject is abnormal, and the subject may have a risk of pancreatic cancer. At this time, the system prompts the doctor of the risk and suggests further examination for diagnosis.

[0090] The four images (plain scan period, arterial period, portal vein period, and venous period) concerned in the embodiment are similar in shape, but their brightness differs due to the slight activity of the human body during CT scanning and the different running time of the enhanced CT contrast agent in the human body. The four brightnesses constitute a time sequence scanning cycle, which completely describes the blood flow process: t0 plain scan without enhancement → t1 arterial period → t2 portal vein period → t3 venous period → t0 plain scan without enhancement.

[0091] According to the time sequence scanning cycle, the algorithm extracts two time sequence evolution sequences in the image feature space: one is the image feature sequence , which reflects the change of the pancreatic shape; and the other is the brightness feature sequence , which reflects the change of the blood flow. Through further analysis of the brightness feature sequence, the relative blood flow can be accurately estimated.

[0092] The application realizes a blood flow estimation method based on non-enhanced pancreatic CT scan images by combining the plain scan CT image and the brightness estimation module and using the method of tetrahedral structure volume ratio. The method not only provides an auxiliary means for early screening of pancreatic cancer, but also has the advantages of simple operation, non-invasiveness, etc., and has a broad clinical application prospect.

[0093] Figure 4 Fig. 1 is a first training model structure schematic diagram of an embodiment of the application; Figure 5 Fig. 2 is a second training model structure schematic diagram of an embodiment of the application; Figure 6 Fig. 3 is an image encoding module workflow schematic diagram of an embodiment of the application; Figure 7 Fig. 4 is an image decoding module workflow schematic diagram of an embodiment of the application; Figure 8 Fig. 5 is a brightness estimation module workflow schematic diagram of an embodiment of the application. As shown in Fig. 5, according to an embodiment of the application, in step S102, the pancreatic enhanced CT estimation model includes a first training model and a second training model; Figures 4-8 The first training model includes an image encoding module (Encoder1), a splicing module (Concatenate), and a first image decoding module (Decoder1);

[0094]

[0095] ​The plain scan CT image data is input into the image encoding module, the input of the splicing module is the output of the image encoding module and the plain scan CT image brightness encoding, the data obtained by processing the splicing module is input into the first image decoding module to finally obtain the enhanced CT image.

[0096] The plain scan CT image brightness encoding is defined as the reference point, and the reference point is "00000000" in the embodiment.

[0097] The second training model comprises an alignment module (Align), an image encoding module (Encoder1), a brightness estimation module (Encoder2), a splicing module (Concatenate) and a second image decoding module (Decoder2).

[0098] The plain scan CT image data is input into the alignment module and the brightness estimation module respectively, the output of the alignment module is processed by the image encoding module, the output of the image encoding module and the output of the brightness estimation module are jointly input into the splicing module, and the output of the splicing module is processed by the second image decoding module to obtain the corresponding enhanced CT image.

[0099] The image encoding module and the splicing module in the second training model are obtained by training the first training model and remain unchanged in the second training model.

[0100] The initial parameters of the second image decoding module adopt the parameters obtained by finally training the first image decoding module, and the parameters of the second image decoding module are updated through the training of the second training model.

[0101] In this embodiment, the pancreatic enhanced CT estimation model aims to predict and generate corresponding enhanced CT images from plain CT image data. The model is based on deep learning techniques, particularly the combination of convolutional neural networks (CNN) and residual networks (ResNet), to extract, refine, and map image features. By training a large number of pancreatic CT image data, the model learns the mapping relationship between plain CT images and enhanced CT images, thereby achieving the effect of obtaining enhanced CT images without actual enhancement scanning. This technique not only improves diagnostic efficiency, but also significantly reduces the risk of radiation exposure for the subject. The pancreatic enhanced CT estimation model includes a first training model and a second training model. During the training process of the pancreatic enhanced CT estimation model, a large amount of pancreatic CT image data is collected, including a set of pancreatic plain CT images and pancreatic enhanced CT images of the same subject, to construct a pancreatic CT image dataset. In the process of constructing the pancreatic CT image dataset, the selected subjects are randomly sampled in a natural population, i.e., the subjects in the pancreatic CT image dataset include both healthy pancreatic subjects and pancreatic cancer subjects, and the number ratio of the two types of subjects is close to that of the natural population. The dataset is divided into a training set and a test set for model training and evaluation. The training set is used to train the first training model and the second training model, and after 400 rounds of training, the test set evaluation can achieve the expected accuracy, and the training is stopped to obtain the applicable base model.

[0102] The first training model is the basic framework of the pancreatic enhanced CT estimation model, composed of an image encoding module, a splicing module, and a first image decoding module. The input of the image encoding module is the plain CT image data, which gradually reduces the size of the image while extracting and refining the effective features of the image through a series of downsampling units and residual units. The downsampling unit reduces the image resolution and reduces the computational load; the residual unit alleviates the gradient vanishing problem in deep neural networks and maintains the integrity and details of the features. The input of the splicing module is the output of the image encoding module and the plain CT image brightness encoding, which is defined as "00000000" in this embodiment. The splicing module splices the two to fuse the brightness information of the image and the extracted features, providing more information for subsequent image decoding and improving the accuracy of the reconstructed image. The input of the first image decoding module is the output of the splicing module, which gradually increases the size of the features while refining and enhancing them through a series of upsampling units and residual units; finally, the features are mapped back to the image space through a convolution unit to obtain the reconstructed enhanced CT image. The upsampling unit restores the image size, the residual unit further enhances the features, and the convolution unit converts the features into image pixels.

[0103] The second training model is extended and optimized based on the first training model and consists of an alignment module, an image encoding module (shared with the first training model), a brightness estimation module, a stitching module (similar to the first training model), and a second image decoding module. The input of the alignment module is the plain CT image data, which is aligned using the Voxelmorph registration technology to ensure that the input image is consistent with the image used during model training in terms of spatial position, thereby improving the prediction accuracy of the model. The input of the brightness estimation module is the plain CT image data, which is estimated through a series of convolution, activation, pooling, and full connection operations to capture the brightness information of the image and provide stronger support for subsequent image stitching and decoding. The input of the stitching module is the output of the image encoding module and the output of the brightness estimation module, which are stitched to fuse the feature information and brightness information of the image and provide a more comprehensive input for image decoding to improve the quality of the reconstructed image. The input of the second image decoding module is the output of the stitching module, which gradually increases the size of the features through a series of upsampling units and residual units, and refines and enhances the features; finally, the reconstructed enhanced CT image is obtained. The initial parameters of the second image decoding module are obtained from the final training of the first image decoding module, which are updated in the second training model to adapt to the new network structure and input data. The second training model is trained based on the first training model, which utilizes the feature extraction and mapping capabilities learned by the first training model to accelerate the convergence of the model. By sharing the image encoding module and the stitching module, the training efficiency and stability of the model are improved. The parameters of the second image decoding module are updated to adapt to the new network structure and task requirements. The second training model combines image alignment, brightness estimation, and feature extraction, and can more accurately predict and generate enhanced CT images. At the same time, by sharing and updating the network modules, the training efficiency and prediction accuracy of the model are improved.

[0104] The pancreatic enhanced CT estimation model of the present application successfully realizes the prediction and generation of corresponding enhanced CT images from plain CT image data by combining the first training model and the second training model, utilizing deep learning technology (especially the combination of convolutional neural networks and residual networks). The model learns the mapping relationship between plain and enhanced CT images by training a large number of pancreatic CT image data, which not only significantly improves the diagnostic efficiency, but also greatly reduces the risk of radiation exposure for the subject. The second training model is extended and optimized based on the first training model and combines image alignment, brightness estimation, and feature extraction, further improving the accuracy of predicting and generating enhanced CT images, while sharing and updating the network modules to improve the training efficiency and stability of the model.

[0105] According to an embodiment of the present application, the luminance estimation module obtained by training in step S102 comprises an arterial phase luminance estimation unit, a portal vein phase luminance estimation unit and a venous phase luminance estimation unit.

[0106] Each unit is respectively used for predicting the luminance relationship between the plain scan CT image and the enhanced CT image of each phase.

[0107] In the present embodiment, the luminance estimation module is finely divided into three independent units: an arterial phase luminance estimation unit, a portal vein phase luminance estimation unit and a venous phase luminance estimation unit. Each unit undertakes a specific task, i.e. is respectively responsible for predicting the luminance relationship between the plain scan CT image and the enhanced CT image of the corresponding phase (arterial phase, portal vein phase, venous phase). In the training process of the pancreatic enhanced CT estimation model, a separate training strategy is adopted for the three different luminance estimation units. By independently training each unit, the model can more accurately learn the complex luminance change rules between the plain scan CT image and the enhanced CT image of each phase, thereby improving the prediction accuracy and reliability of the entire pancreatic enhanced CT estimation model. Specifically, the arterial phase luminance estimation unit is trained using the luminance relationship between the plain scan CT image and the arterial phase CT image, the portal vein phase luminance estimation unit is trained using the luminance relationship between the plain scan CT image and the portal vein phase CT image, and the venous phase luminance estimation unit is trained using the luminance relationship between the plain scan CT image and the venous phase luminance estimation unit.

[0108] The present application finely divides the luminance estimation module into three independent units of arterial phase, portal vein phase and venous phase, and adopts a separate training strategy, so that the model can more accurately learn the luminance change rules between the plain scan CT and the enhanced CT of each phase, thereby improving the prediction accuracy and reliability of the pancreatic enhanced CT estimation model, and providing more accurate and personalized auxiliary information for clinical use.

[0109] According to an embodiment of the present application, in step S112, the threshold is set by the following method:

[0110] Based on the data set used for training the pancreatic enhanced CT estimation model, the relative blood flow values of all subjects are obtained;

[0111] All the relative blood flow values are fitted by Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold = 1 + .

[0112] In this embodiment, the threshold setting method is defined clearly and reasonably: first, based on the data set for training the pancreatic enhanced CT estimation model, the relative blood flow values of all subjects are extracted. During the training process of the pancreatic enhanced CT estimation model, a pancreatic CT image data set is constructed in advance, which collects a large amount of pancreatic image data of subjects, including the pancreatic plain CT image and the pancreatic enhanced CT image set of the same subject. In order to make the final result closer to the actual situation of the natural population, during the construction of the pancreatic CT image data set, the selected subjects are randomly sampled in a natural population, that is, the subjects in the pancreatic CT image data set include both healthy subjects and subjects with pancreatic cancer, and the number ratio of the two types of subjects is close to that of the natural population. As provided in step S108, when calculating the relative blood flow value of each subject, the brightness encoding of the plain CT image of the subject is defined as "00000000", the first method can predict the arterial phase CT image brightness encoding, the portal vein phase CT image brightness encoding and the venous phase CT image brightness encoding of the subject by the method of step S106; the second method can be to obtain the arterial phase CT image brightness encoding, the portal vein phase CT image brightness encoding and the venous phase CT image brightness encoding of the subject by processing the pancreatic enhanced CT image set of the subject. Here, in order to make the relative blood flow value closer to the real data of the natural population, the second method is adopted.

[0113] The CT image brightness encoding of the subject judged as a healthy pancreas in the pancreatic CT image data set is averaged to obtain a reference tetrahedral structure volume, and the relative blood flow value of each subject is obtained by comparing the tetrahedral structure volume formed in the high-dimensional space of all subjects with the reference tetrahedral structure volume.

[0114] Then, the relative blood flow values are fitted by using the Poisson distribution, and the standard deviation σ of the relative blood flow Poisson distribution is obtained by calculation. On this basis, the threshold is set to 1 plus the standard deviation, that is, threshold = 1 + σ. The relative blood flow estimation value shows the characteristics of the Poisson distribution, and by calculating the relative blood flow data, we can fit the parameters lambda of the Poisson distribution, and then obtain the standard deviation When the relative blood flow estimation value of a subject deviates from the normal range and reaches or exceeds the threshold, it is considered to be suspected of blood flow abnormality.

[0115] The present application can more scientifically define the standard of blood flow abnormality through the fitting of the Poisson distribution, and improve the accuracy and reliability of the model for detecting blood flow abnormality.

[0116] According to an embodiment of the present application, in step S112, the subject is classified based on the judgment result, comprising:

[0117] When the target relative blood flow does not exceed the threshold value, the subject is classified as a low-risk group of pancreatic cancer;

[0118] When the target relative blood flow exceeds the threshold value, the subject is classified as a high-risk group of pancreatic cancer, and the high-risk group of pancreatic cancer is further examined.

[0119] In this embodiment, by comparing the target relative blood flow of the subject with the preset threshold value, the accurate classification of the subject is realized. Specifically, when the target relative blood flow does not exceed the threshold value, the subject is classified as a low-risk group of pancreatic cancer, and does not need to be further examined; when the target relative blood flow exceeds the threshold value, the subject is classified as a high-risk group of pancreatic cancer, and is recommended to receive more detailed pancreatic examination.

[0120] Based on the quantitative evaluation of the relative blood flow, the present application not only effectively utilizes the medical resources and avoids unnecessary examination, but also provides a scientific diagnosis reference for doctors, which helps to early detect the risk of pancreatic cancer and improve the accuracy and efficiency of diagnosis.

[0121] Example Two

[0122] As shown in Figures 2-8 According to an embodiment of the present application, a relative blood flow estimation method based on a pancreatic plain CT image, comprising the following steps:

[0123] Step S202, constructing a pancreatic enhanced CT estimation model, and obtaining a brightness estimation module based on the pre-trained pancreatic enhanced CT estimation model;

[0124] The pancreatic enhanced CT estimation model comprises a first training model and a second training model;

[0125] The first training model comprises an image encoding module, a splicing module and a first image decoding module;

[0126] The plain CT image data is input into the image encoding module, the input of the splicing module is the output of the image encoding module and the brightness encoding of the plain CT image, and the data obtained by processing the splicing module is input into the first image decoding module to finally obtain the enhanced CT image.

[0127] The brightness encoding of the plain CT image is "00000000";

[0128] The second training model comprises an alignment module, an image encoding module, a brightness estimation module, a splicing module and a second image decoding module;

[0129] The plain scan CT image data is respectively input into an alignment module and a brightness estimation module, the output of the alignment module is processed through an image encoding module, the output of the image encoding module and the output of the brightness estimation module are jointly input into a splicing module, and the output of the splicing module is processed through a second image decoding module to obtain a corresponding enhanced CT image;

[0130] The image encoding module and the splicing module in the second training model are obtained through the training of the first training model and remain unchanged in the second training model.

[0131] The initial parameters of the second image decoding module are the parameters obtained through the final training of the first image decoding module, and the parameters of the second image decoding module are updated through the training of the second training model.

[0132] In the training process of the pancreatic enhanced CT estimation model, a large amount of pancreatic CT image data is collected, including a set of pancreatic plain scan CT images and pancreatic enhanced CT images of the same subject, and a pancreatic CT image data set is constructed. In order to make the final result closer to the actual situation of the natural population, in the process of constructing the pancreatic CT image data set, the selected subjects are randomly sampled in a natural population, that is, the subjects in the pancreatic CT image data set include both healthy subjects and subjects with pancreatic cancer, and the number ratio of the two types of subjects is close to that of the natural population. The data set is divided into a training set and a test set for model training and evaluation. The training set is used to train the first training model and the second training model, and after 400 rounds of training, the test set can achieve the expected accuracy rate, and the training is stopped to obtain an applicable basic model.

[0133] In the training, t1, t2, and t3 are selected according to the time sequence (fixed time interval) for training, a decoding module needs to be constructed to enable the features to be decoded to obtain the MSEloss of the predicted and real highlight pictures. However, the decoding module is not used, but only used for training the encoding module. Specifically, the decoding module reads the splicing of the two groups of features and decodes to restore the enhanced image. In actual use, a new CT enhanced image is collected, input into the model, and the position information of the pancreatic cancer is obtained. The model can be deployed on a general computing device.

[0134] Step S204, acquiring a pancreatic plain scan CT image of the subject, i.e., a target plain scan CT image;

[0135] Step S206, obtaining the predicted arterial phase CT image brightness encoding, portal vein phase CT image brightness encoding, and venous phase CT image brightness encoding based on the target plain scan CT image and the brightness estimation module;

[0136] The brightness estimation module includes an arterial phase brightness estimation unit, a portal vein phase brightness estimation unit, and a venous phase brightness estimation unit.

[0137] Each unit is used to predict the brightness relationship corresponding to the plain CT image and each phase of the enhanced CT image.

[0138] The predicted arterial phase CT image brightness code is obtained based on the target plain CT image and the arterial phase brightness estimation unit, the predicted portal venous phase CT image brightness code is obtained based on the target plain CT image and the portal venous phase brightness estimation unit, and the predicted venous phase CT image brightness code is obtained based on the target plain CT image and the venous phase brightness estimation unit.

[0139] In step S208, the target plain CT image brightness code is defined as "00000000", and the target tetrahedron structure is constructed in the high-dimensional space based on the CT image brightness codes under different phases.

[0140] The non-enhanced CT (i.e. plain CT) brightness code is "00000000". For the t1 arterial phase, t2 portal venous phase, and t3 venous phase three enhanced images, the brightness codes are obtained through the brightness estimation module. After obtaining the brightness codes, the brightness codes form a tetrahedron in the high-dimensional space, and the volume of the tetrahedron describes the blood flow, which is called the brightness tetrahedron.

[0141] In step S210, the target relative blood flow is obtained based on the ratio of the target tetrahedron structure volume to the reference tetrahedron structure volume; wherein the reference tetrahedron structure volume is the average value of the tetrahedron structure volume formed by the CT image brightness codes of the healthy pancreas under different phases in the high-dimensional space.

[0142] For pancreatic cancer negative subjects, the brightness tetrahedron conforms to the general rule, and the corresponding tetrahedron is the negative general tetrahedron V - The average value of the tetrahedron structure volume corresponding to the pancreatic cancer negative subjects (i.e. healthy pancreas) is obtained (the data used in this process is based on the data of the pancreatic CT image data set determined as pancreatic cancer negative subjects), and the reference tetrahedron structure volume S - is obtained. For pancreatic cancer positive subjects, the brightness tetrahedron will be distorted, and the corresponding positive tetrahedron V + will also change in volume. Therefore, when the target tetrahedron structure volume S + of the subject is obtained, the target relative blood flow is obtained by dividing the target tetrahedron structure volume by the reference tetrahedron structure volume.

[0143] The target relative blood flow is used to determine whether the subject is suspected to be a pancreatic cancer positive subject.

[0144] In step S212, it is determined whether the target relative blood flow exceeds a threshold value, and the subject is classified based on the determination result.

[0145] wherein the threshold value is set by:

[0146] Based on the data set for training the pancreatic enhancement CT estimation model, the relative blood flow values of all subjects are obtained; specifically, the average value of the tetrahedral structure volume formed by the CT image brightness coding of the subjects judged as healthy pancreas in the pancreatic CT image data set at different phases in the high-dimensional space is obtained to obtain a reference tetrahedral structure volume, and the relative blood flow values of all subjects are obtained by comparing the tetrahedral structure volume formed by each subject in the high-dimensional space with the reference tetrahedral structure volume.

[0147] The relative blood flow values are fitted by a Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold value = 1 + .

[0148] Based on the judgment result, the subjects are classified, including:

[0149] When the target relative blood flow does not exceed the threshold value, the subject is classified as a low-risk population of pancreatic cancer;

[0150] When the target relative blood flow exceeds the threshold value, the subject is classified as a high-risk population of pancreatic cancer, and the high-risk population of pancreatic cancer is further subjected to pancreatic examination.

[0151] The present application not only limits to simple positive or negative judgment of the pancreatic CT scan image, but also greatly improves the detection accuracy through estimation of the relative blood flow, so that the doctor can more clearly understand the pancreatic condition of the patient, and provides strong support for subsequent treatment.

[0152] The present application realizes prediction of the enhanced CT image from the plain CT image, and evaluates the relative blood flow condition of the patient according to the prediction, which provides more rich diagnostic information for the doctor, and helps the doctor to more comprehensively understand the pancreatic blood flow condition of the patient, and then makes more accurate diagnosis.

[0153] The present application also shows significant advantages in the enhancement of pancreatic imaging, compared with the prior art, it can provide more perfect and clear pancreatic imaging images, so that the doctor can more clearly observe the detailed structure of the pancreas, thereby improving the accuracy of diagnosis.

[0154] The present application does not need to increase the additional burden on the patient and medical resources. It only needs to analyze and predict the existing CT scan image, which saves time and reduces cost. At the same time, the scheme can also assist the doctor to quickly and directly predict pancreatic cancer through the pancreatic CT scan image, so that the subsequent enhanced imaging examination is more targeted, and the diagnosis efficiency is further improved.

[0155] The judgment model used in the application has high flexibility, which can run on a high-performance computer platform, and can also be adapted to a general computer or a mobile phone platform, so that doctors can use the model for diagnosis at any time and anywhere, greatly improving the availability and popularity of the model.

[0156] Embodiment three

[0157] Figure 9 The structure diagram of a relative blood flow estimation system based on a pancreatic plain scan CT image according to an embodiment of the application is shown in FIG. 1. As shown in FIG. 1, a relative blood flow estimation system based on a pancreatic plain scan CT image comprises: Figure 9

[0158] The construction module 10 is configured to construct a pancreatic enhanced CT estimation model, and obtain a brightness estimation module based on the pancreatic enhanced CT estimation model.

[0159] The acquisition module 20 is configured to acquire a pancreatic plain scan CT image of a subject, i.e., a target plain scan CT image.

[0160] The prediction module 30 is configured to obtain a predicted arterial phase CT image brightness encoding, a predicted portal vein phase CT image brightness encoding, and a predicted venous phase CT image brightness encoding based on the target plain scan CT image and the brightness estimation module.

[0161] The calculation module 40 is configured to define a target plain scan CT image brightness encoding as a reference point “00000000”, and construct a target tetrahedron structure in a high-dimensional space based on CT image brightness encodings in different phases.

[0162] The calculation module 40 is further configured to obtain a target relative blood flow based on a ratio of a target tetrahedron structure volume to a reference tetrahedron structure volume; wherein the reference tetrahedron structure volume is an average value of tetrahedron structure volumes formed by CT image brightness encodings of a healthy pancreas in different phases in a high-dimensional space.

[0163] The judgment module 50 is configured to judge whether the target relative blood flow exceeds a threshold value, and classify the subject based on a judgment result.

[0164] According to an embodiment of the application, in the construction module 10, the pancreatic enhanced CT estimation model comprises a first training model and a second training model.

[0165] The first training model comprises an image encoding module, a splicing module, and a first image decoding module; plain scan CT image data is input into the image encoding module, the input of the splicing module is an output of the image encoding module and a plain scan CT image brightness encoding, and data obtained by processing of the splicing module is input into the first image decoding module to finally obtain an enhanced CT image.

[0166] ​The brightness of the plain CT image is encoded as a reference point '00000000';

[0167] The second training model comprises an alignment module, an image encoding module, a brightness estimation module, a splicing module and a second image decoding module; the plain CT image data are input into the alignment module and the brightness estimation module respectively, the output of the alignment module is processed by the image encoding module, the output of the image encoding module and the output of the brightness estimation module are jointly input into the splicing module, and the output of the splicing module is processed by the second image decoding module to obtain the corresponding enhanced CT image.

[0168] The image encoding module and the splicing module in the second training model are obtained by training the first training model and remain unchanged in the second training model; the initial parameters of the second image decoding module are the parameters obtained by the final training of the first image decoding module; and the parameters of the second image decoding module are updated through the training of the second training model.

[0169] According to an embodiment of the present application, in the construction module 10, the brightness estimation module comprises an arterial phase brightness estimation unit, a portal vein phase brightness estimation unit and a venous phase brightness estimation unit.

[0170] Each unit is used for predicting the brightness relationship corresponding to the plain CT image and the enhanced CT image of each phase.

[0171] According to an embodiment of the present application, in the judgment module 50, the threshold is set by the following method:

[0172] Based on the data set for training the pancreatic enhanced CT estimation model, the relative blood flow values of all subjects are obtained;

[0173] The relative blood flow values of all subjects are fitted by a Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold = 1 + .

[0174] According to an embodiment of the present application, in the judgment module 50, the subjects are classified based on the judgment result, comprising:

[0175] When the target relative blood flow does not exceed the threshold, the subject is classified as a low-risk population of pancreatic cancer;

[0176] When the target relative blood flow exceeds the threshold, the subject is classified as a high-risk population of pancreatic cancer, and the high-risk population of pancreatic cancer is further examined.

[0177] The application is based on quantitative evaluation of relative blood flow, which not only effectively utilizes medical resources and avoids unnecessary examination, but also provides a scientific diagnosis reference for doctors, helps to early discover the risk of pancreatic cancer, and improves the accuracy and efficiency of diagnosis.

[0178] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0179] The above description is only the preferred embodiment of the present application and the explanation of the technical principles applied. Those skilled in the art should understand that the scope of the application involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the technical solutions formed by mutually replacing the above features and the technical features disclosed in the present application (but not limited to) with similar functions.

[0180] It should be understood that the size of the serial number of the steps in the summary and the embodiments of the present application does not absolutely mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

Claims

1. A method of estimating relative blood flow based on a pancreatic non-contrast CT image, characterized by, The method comprises the following steps: constructing a pancreatic enhanced CT estimation model and obtaining a brightness estimation module based on pre-training of the pancreatic enhanced CT estimation model; obtaining a pancreatic plain CT image of a subject, i.e., a target plain CT image; based on the target plain CT image and the brightness estimation module, obtaining predicted arterial phase CT image brightness encoding, portal venous phase CT image brightness encoding and venous phase CT image brightness encoding; defining target plain CT image brightness encoding as a reference point, and constructing a target tetrahedron structure in a high-dimensional space based on CT image brightness encoding in different phases; obtaining a target relative blood flow based on a ratio of a volume of the target tetrahedron structure to a volume of a reference tetrahedron structure, wherein the volume of the reference tetrahedron structure is an average value of volumes of tetrahedron structures formed by CT image brightness encoding of a healthy pancreas in different phases in a high-dimensional space; judging whether the target relative blood flow exceeds a threshold value, and classifying the subject based on the judgment result; the pancreatic enhanced CT estimation model comprises a first training model and a second training model; the first training model comprises an image encoding module, a splicing module and a first image decoding module; plain CT image data is input into the image encoding module, the input of the splicing module is the output of the image encoding module and plain CT image brightness encoding, and the data obtained by processing of the splicing module is input into the first image decoding module to finally obtain an enhanced CT image; wherein the plain CT image brightness encoding serves as a reference point; the second training model comprises an alignment module, an image encoding module, a brightness estimation module, a splicing module and a second image decoding module; plain CT image data is input into the alignment module and the brightness estimation module respectively, the output of the alignment module is processed by the image encoding module, the output of the image encoding module and the output of the brightness estimation module are jointly input into the splicing module, and the output of the splicing module is processed by the second image decoding module to obtain a corresponding enhanced CT image; wherein the image encoding module and the splicing module in the second training model are obtained by training of the first training model and remain unchanged in the second training model; the initial parameters of the second image decoding module adopt the parameters finally obtained by training of the first image decoding module; the parameters of the second image decoding module are updated through training of the second training model.

2. The method of claim 1, wherein the method of estimating relative blood flow based on a non-contrast CT image of a pancreas is characterized by, the brightness estimation module comprises an arterial phase brightness estimation unit, a portal venous phase brightness estimation unit and a venous phase brightness estimation unit; each unit is respectively used to predict a brightness relationship corresponding to a plain CT image and an enhanced CT image in each phase.

3. The method of claim 1, wherein the method is based on a relative blood flow estimation method using a non-contrast CT image of a pancreas. the threshold value is set by the following method: based on a data set used for training the pancreatic enhanced CT estimation model, obtaining relative blood flow values of all subjects; The relative blood flow values of all subjects were fitted by a Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold = 1 .

4. The method of claim 1, wherein, based on the judgment result, classifying the subject, comprising: when the target relative blood flow does not exceed the threshold value, classifying the subject as a pancreatic cancer low-risk population; When the target relative blood flow exceeds the threshold value, the subject is classified as a high-risk group of pancreatic cancer, and further pancreatic examination is performed on the high-risk group of pancreatic cancer.

5. A system for estimating relative blood flow based on a pancreatic non-contrast CT image, characterized by, The method comprises the following steps: A construction module is configured to construct a pancreatic enhanced CT estimation model, and pre-train a brightness estimation module based on the pancreatic enhanced CT estimation model. An acquisition module is configured to acquire a pancreatic non-enhanced CT image of a subject, i.e., a target non-enhanced CT image. A prediction module is configured to predict, based on the target non-enhanced CT image and the brightness estimation module, an arterial phase CT image brightness encoding, a portal venous phase CT image brightness encoding, and a venous phase CT image brightness encoding. A calculation module is configured to define a target non-enhanced CT image brightness encoding as a reference point, and construct a target tetrahedron structure in a high-dimensional space based on CT image brightness encodings in different phases. The calculation module is further configured to obtain a target relative blood flow based on a ratio of a target tetrahedron structure volume to a reference tetrahedron structure volume, wherein the reference tetrahedron structure volume is an average value of tetrahedron structure volumes formed by CT image brightness encodings in different phases of a healthy pancreas in a high-dimensional space. A judgment module is configured to judge whether the target relative blood flow exceeds a threshold value, and classify the subject based on a judgment result. In the construction module, the pancreatic enhanced CT estimation model comprises a first training model and a second training model. The first training model comprises an image encoding module, a splicing module, and a first image decoding module; non-enhanced CT image data is input into the image encoding module, the input of the splicing module is the output of the image encoding module and non-enhanced CT image brightness encoding, and the data obtained by processing the splicing module is input into the first image decoding module to finally obtain an enhanced CT image; wherein the non-enhanced CT image brightness encoding serves as a reference point. The second training model comprises an alignment module, an image encoding module, a brightness estimation module, a splicing module, and a second image decoding module; non-enhanced CT image data is input into the alignment module and the brightness estimation module, respectively; the output of the alignment module is processed by the image encoding module; the output of the image encoding module and the output of the brightness estimation module are jointly input into the splicing module; and the output of the splicing module is processed by the second image decoding module to obtain a corresponding enhanced CT image. In the second training model, the image encoding module and the splicing module are trained by the first training model and remain unchanged in the second training model; the initial parameters of the second image decoding module adopt the parameters finally trained by the first image decoding module; and the parameters of the second image decoding module are updated through the training of the second training model.

6. The relative blood flow estimation system based on a non-contrast CT image of a pancreas according to claim 5, wherein, In the construction module, the brightness estimation module comprises an arterial phase brightness estimation unit, a portal venous phase brightness estimation unit, and a venous phase brightness estimation unit. Each unit is respectively configured to predict a brightness relationship between a non-enhanced CT image and an enhanced CT image in each phase.

7. The relative blood flow estimation system based on a non-contrast CT image of a pancreas according to claim 5, wherein, In the judgment module, the threshold value is set by the following method: Based on the data set for training the pancreatic enhancement CT estimation model, a relative blood flow value of all subjects is obtained; The relative blood flow values of all subjects were fitted by a Poisson distribution to obtain the standard deviation of the relative blood flow Poisson distribution , threshold = 1 . 8.The relative blood flow estimation system based on a pancreatic non-contrast CT image according to claim 5, wherein, In the judgment module, the subjects are classified based on the judgment result, comprising: When the target relative blood flow does not exceed the threshold value, the subjects are classified as a low-risk group of pancreatic cancer; When the target relative blood flow exceeds the threshold value, the subjects are classified as a high-risk group of pancreatic cancer, and the high-risk group of pancreatic cancer is further examined.

Images