Cervical cancer metastasis prediction model establishment method, system, equipment and medium

Abstract

The invention discloses a cervical cancer metastasis prediction model establishing method, system, equipment and medium, and relates to the technical field of cervical cancer metastasis prediction.The cervical cancer metastasis prediction model establishing method comprises the following steps that MRI images and SCCA detection data of a cervical cancer patient are collected and classified, and first MRI image related data are obtained; carrying out transfer feature extraction processing, and constructing a transfer feature vector and second MRI image related data; constructing a cervical cancer metastasis risk prediction model based on the second MRI image related data and a logistic regression model; obtaining a metastasis feature vector of the cervical cancer patient, predicting a metastasis risk by using the cervical cancer metastasis risk prediction model, and calculating an overall metastasis risk of the cervical cancer patient; the method is used for solving the problem that when an existing cervical cancer metastasis prediction technology is used for predicting the cervical cancer metastasis risk, only one type of characteristic indexes are relied on, and the metastasis risk cannot be predicted through the change condition of the same characteristic and the dynamic association between different characteristic changes.

Description

Technical Field

[0001] The present invention relates to the technical field of cervical cancer metastasis prediction, and in particular to a method, system, device and medium for establishing a cervical cancer metastasis prediction model. Background Art

[0002] Cervical cancer metastasis prediction technology refers to a series of methods and means used to evaluate the possibility and risk of tumor metastasis in cervical cancer patients. By combining multiple technologies, the biological behavior of cervical cancer is analyzed from different angles, aiming to determine early and accurately whether the tumor has a tendency to metastasize, provide a basis for clinical treatment decisions, and improve the patient's prognosis and quality of life.

[0003] Existing cervical cancer metastasis prediction technologies often rely on a single indicator, such as tumor markers or tumor size, when predicting the risk of cervical cancer metastasis. However, cancer metastasis is a complex process affected by multiple factors, and a single indicator cannot cover all relevant information. For example, if only tumor marker antigens are relied upon, factors that also have an important impact on metastasis, such as tumor size, pathological type, and patient immune status, will be ignored. If only tumor morphological characteristics are considered, the biological behavior of the tumor, the expression of molecular markers, etc., cannot be understood. Such information is crucial for accurately judging the risk of cervical cancer metastasis. For example, in the patent application with publication number CN110516759A, a machine-based prediction method is disclosed. A soft tissue sarcoma metastasis risk prediction system based on machine learning only relies on a single image feature when predicting the metastasis risk, which is easily interfered by external factors and causes fluctuations in the prediction results, affecting the prediction accuracy; moreover, the existing cervical cancer metastasis prediction technology often only analyzes static features when predicting the metastasis risk of cervical cancer, without considering the development and changes between features, and cannot use the dynamic correlation between different features to predict the metastasis risk; therefore, the existing cervical cancer metastasis prediction technology only relies on one type of feature indicator when predicting the metastasis risk of cervical cancer, and cannot use the changes in the same feature and the dynamic correlation between different feature changes to predict the metastasis risk. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent, by collecting MRI images and SCCA test data of cervical cancer patients and classifying them to obtain first MRI image related data; performing metastasis feature extraction processing, and constructing a metastasis feature vector and second MRI image related data; constructing a cervical cancer metastasis risk prediction model; obtaining the metastasis feature vector of cervical cancer patients, predicting the metastasis risk using the cervical cancer metastasis risk prediction model, and calculating the overall metastasis risk of cervical cancer patients; so as to solve the problem that the existing cervical cancer metastasis prediction technology only relies on one type of feature indicator when predicting the metastasis risk of cervical cancer, and cannot use the changes in the same features and the dynamic correlation between different feature changes to predict the metastasis risk.

[0005] To achieve the above objectives, in a first aspect, the present application provides a method for establishing a cervical cancer metastasis prediction model, comprising the following steps: Collect MRI images and SCCA test data of cervical cancer patients, classify them, and obtain first MRI image related data; Performing transfer feature extraction processing based on the first MRI image-related data, and constructing a transfer feature vector and the second MRI image-related data; A cervical cancer metastasis risk prediction model was constructed based on the second MRI image-related data and the logistic regression model; The metastasis feature vectors of cervical cancer patients were obtained, the metastasis risk was predicted using the cervical cancer metastasis risk prediction model, and the overall metastasis risk of cervical cancer patients was calculated.

[0006] Furthermore, collecting MRI images and SCCA test data of cervical cancer patients and classifying them to obtain first MRI image-related data includes the following sub-steps: Collect MRI images of the lesion area of ​​cervical cancer patients before cervical cancer metastasis, classify them according to the corresponding cervical cancer patients, and sort them in time order from far to near according to the shooting time of the MRI images, denoted as j=1, 2, ..., m, where j represents the number of times the MRI image is shot, and is marked as MRI image data; The SCCA detection data of cervical cancer patients in tumor marker examinations before cervical cancer metastasis are collected, classified according to the corresponding cervical cancer patients, and sorted in chronological order from recent to distant according to the SCCA detection time. According to the shooting time of the MRI image, for any MRI image, the SCCA detection data closest in time scale is recorded as the SCCA feature of the MRI image, and the SCCA features of all MRI images are repeatedly obtained. After completion, the SCCA feature sequence is obtained; The MRI image data and the SCCA detection data are divided into a non-metastatic group and a metastatic group according to whether the corresponding cervical cancer patient has cervical cancer metastasis, and the MRI image data and the SCCA detection data are merged and stored according to the corresponding cervical cancer patient, and marked as the first MRI image-related data.

[0007] Furthermore, performing transfer feature extraction processing based on the first MRI image-related data and constructing a transfer feature vector and the second MRI image-related data includes the following sub-steps: grayscale processing is performed on all MRI images in the first MRI image related data to obtain grayscale image data; for any grayscale image in the grayscale image data, pixel points in the lesion area of ​​the grayscale image are marked as cancer pixel points and pixel points at the edge of the lesion area are marked as cancer edge pixel points, and area feature acquisition, first shape feature acquisition and second shape feature acquisition are performed; The area feature acquisition includes: counting the number of cancer pixels in the lesion area, obtaining the actual area represented by each pixel in the corresponding grayscale image, calculating the area size of the lesion area, marking it as the area feature of the lesion area, recorded as Ba; The first shape feature acquisition includes: obtaining the coordinates of the cancer edge pixel points in the corresponding grayscale image, and randomly selecting the cancer edge pixel points as the starting pixel points, and obtaining the coordinate sequence of the cancer edge pixel points in a clockwise direction, which is recorded as G i =(x i ,y i ), i=0, 1, 2, ..., n; for any cancer edge pixel, the edge curvature is calculated by the first edge formula, which is as follows: , where k i represents the edge curvature of the i-th cancer edge pixel, Δx i =x i+1 -x i , Δy i =y i+1 -y i ; The edge curvature of all cancer edge pixels is obtained, recorded as an edge curvature sequence, and the first shape feature of the lesion area is calculated by the first shape feature formula. The first shape feature formula is as follows: , where Bb represents the first shape feature, k0 represents the average value of the edge curvature sequence; The second shape feature acquisition includes: based on the edge curvature sequence, for any edge curvature k i , to determine whether k is satisfied i >k i-1 And k i >k i+1, if it is satisfied, then mark the edge curvature as the peak curvature, count the number of peak curvatures, record it as c1, and calculate the average value of all peak curvatures, record it as c2, and calculate the second shape feature by the second shape feature formula. The second shape feature formula is as follows: , where Bc represents the second shape feature; The area feature, the first shape feature and the second shape feature are marked as basic features, and the basic features of all grayscale images are repeatedly obtained to obtain a basic feature sequence.

[0008] Furthermore, performing transfer feature extraction processing based on the first MRI image-related data and constructing a transfer feature vector and the second MRI image-related data also includes the following sub-steps: Based on the basic feature sequence and the SCCA feature sequence, area change feature acquisition, first shape change feature acquisition, second shape change feature acquisition and SCCA change feature acquisition are performed respectively; Acquiring the area change feature, acquiring the first shape change feature, acquiring the second shape change feature and acquiring the SCCA change feature includes: when j=1, setting the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature to 0; When j=2, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are calculated respectively using the first change feature formula. The first change feature formula is as follows: , where Y j (a, b, c, d) represent the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d), respectively. j (a, b, c, d) represent the area feature, the first shape feature, the second shape change feature and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d), T0 is the set standard time interval, T j is the time corresponding to the jth MRI image; When j>2, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are calculated respectively using the second change feature formula, and the second change feature formula is as follows: , Where q1 and q2 are the set weights, with a value range of (0, 1), and q1+q2=1; The area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are marked as basic change features, and the basic change features of all grayscale images are repeatedly obtained to obtain a basic change feature sequence.

[0009] Furthermore, performing transfer feature extraction processing based on the first MRI image-related data and constructing a transfer feature vector and the second MRI image-related data also includes the following sub-steps: The basic feature sequence and the basic change feature sequence are normalized according to the feature type, and all numerical values ​​are scaled to [0, 1]; the normalized basic feature sequence and basic change feature are obtained respectively; For any MRI image, a feature vector is set, marked as a transfer feature vector, and recorded as R={Ba, Bb, Bc, Bd, Ya, Yb, Yc, Yd}, where R represents the transfer feature vector of the corresponding MRI image; Ba, Bb, Bc, Bd, Ya, Yb, Yc and Yd represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature of the corresponding MRI image in order; The transfer feature vectors of all MRI images in the first MRI image-related data are obtained and marked as the second MRI image-related data.

[0010] Furthermore, constructing a cervical cancer metastasis risk prediction model based on the second MRI image-related data and the logistic regression model includes the following sub-steps: The first metastasis risk prediction model was constructed using the logistic regression model. The first metastasis risk prediction model is as follows: , where P(1) represents the risk probability of metastasis; Z1, Z2, Z3, Z4, Z5, Z6, Z7 and Z8 represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature in order; β0, β1, β2, β3, β4, β5, β6, β7 and β8 are the parameters of the first metastasis risk prediction model; The learning rate of the model training is set to v1, the training round is set to v2, the batch size is set to v3, and the first metastasis risk prediction model is trained using the second MRI image-related data to obtain the cervical cancer metastasis risk prediction model.

[0011] Furthermore, obtaining the metastasis feature vector of the cervical cancer patient, predicting the metastasis risk using the cervical cancer metastasis risk prediction model, and calculating the overall metastasis risk of the cervical cancer patient includes the following sub-steps: For patients with cervical cancer to be predicted, the metastasis feature vectors of all MRI images taken are obtained, recorded as R1, R2, ..., RF, where F is the total number of shots, and the corresponding shooting time t1, t2, ..., tF is obtained, and the time intervals between two adjacent times e1, e2, ..., e (F-1) are calculated; R1, R2, ..., RF are respectively input into the cervical cancer metastasis risk prediction model to obtain the corresponding metastasis risk probabilities P1, P2, ..., PF; For any transfer feature vector in P1, P2, ..., PF, it is recorded as Pf; the weighted intermediate variable of Pf is calculated using the intermediate variable formula, and the intermediate variable formula is as follows: , where Q k f represents the weighted intermediate variable of Pf, eu represents the time interval of the uth time, and W represents the set attenuation coefficient; repeatedly calculate the weighted intermediate variables of P1, P2, ..., PF to obtain the corresponding weighted intermediate variable Q k 1. Q k 2, ..., Q k F; Then use the weight calculation formula to calculate the weight of Rf. The weight calculation formula is as follows: , where Qf represents the weight of Rf, and the weights of P1, P2, ..., PF are calculated repeatedly to obtain the corresponding weights Q1, Q2, ..., QF; The metastasis risk formula is then used to calculate the overall metastasis risk of cervical cancer patients. The metastasis risk formula is as follows: , where P0 represents the overall metastasis risk of cervical cancer patients.

[0012] In a second aspect, the present application provides a system for establishing a cervical cancer metastasis prediction model, including a sample collection module, a feature acquisition module, a model establishment module, and a metastasis prediction module; The sample collection module is used to collect MRI images of the lesion area of ​​cervical cancer patients and SCCA detection data, and classify them to obtain first MRI image related data; The feature extraction module includes a feature extraction unit and a feature vector unit, the feature extraction unit performs transfer feature extraction processing based on the first MRI image related data, and the feature vector unit is used to construct a transfer feature vector and the second MRI image related data; The model building module builds a cervical cancer metastasis risk prediction model based on the second MRI image-related data and a logistic regression model; The metastasis prediction module includes a risk prediction unit and a risk calculation unit; the risk prediction unit is used to obtain the metastasis feature vector of cervical cancer patients and predict the metastasis risk using a cervical cancer metastasis risk prediction model; the risk calculation unit is used to calculate the overall metastasis risk of cervical cancer patients.

[0013] In a third aspect, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the above method are performed.

[0014] In a fourth aspect, the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the above method are performed.

[0015] Beneficial effects of the present invention: The present invention obtains first MRI image-related data by collecting MRI images and SCCA test data of cervical cancer patients and classifying them; performs metastasis feature extraction and processing based on the first MRI image-related data, and constructs a metastasis feature vector and second MRI image-related data; constructs a cervical cancer metastasis risk prediction model based on the second MRI image-related data and a logistic regression model; obtains the metastasis feature vector of cervical cancer patients, predicts the metastasis risk using the cervical cancer metastasis risk prediction model, and calculates the overall metastasis risk of cervical cancer patients; while using a variety of feature indicators, uses the changes in the same features and the dynamic association between different feature changes to predict the metastasis risk, thereby improving the accuracy of risk prediction. The present invention obtains multiple characteristic indicators of cervical cancer patients and obtains the change characteristics of multiple characteristics. The advantages are that it not only pays attention to each characteristic itself, but also pays attention to their changes, and learns the intrinsic connection between these characteristic changes and the relationship with metastasis risk in model training; it can dig out more in-depth disease information and improve the accuracy of prediction; the overall metastasis risk is calculated by a coefficient attenuation weighted method, and the advantage is that the weight of each examination in the assessment of the current metastasis risk prediction is reasonably allocated, so that the examination results closer to the current time and with a shorter time interval occupy a more important position in the risk assessment; improve its applicability and accuracy in various situations; by integrating morphological characteristics and marker changes to construct a metastasis feature vector, it can more comprehensively characterize the development status of cancer. These features complement each other, so that the model can more accurately capture potential information related to cervical cancer metastasis, thereby improving the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a functional block diagram of the system of the present invention; Figure 2 is a flow chart of the steps of the method of the present invention; Figure 3 A schematic diagram for obtaining a first shape feature of the present invention; Figure 4 is a flow chart of the transfer prediction module processing of the present invention; Figure 5 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0017] 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.

[0018] Example 1, please refer to Figure 1 As shown, the present application provides a system for establishing a cervical cancer metastasis prediction model, including a sample collection module, a feature acquisition module, a model establishment module and a metastasis prediction module; The sample collection module is used to collect MRI images of the lesion area of ​​cervical cancer patients and SCCA detection data, and classify them to obtain the first MRI image related data; The sample collection module is configured with a sample collection strategy, which includes: collecting MRI images of the lesion area of ​​cervical cancer patients before cervical cancer metastasis, that is, nuclear magnetic resonance images; classifying according to the corresponding cervical cancer patients, that is, classifying according to the cervical cancer patients to which the MRI images belong, and sorting in time order from far to near according to the shooting time of the MRI images, recorded as j=1, 2,..., m, where j represents the number of times the MRI image is shot, marked as MRI image data; for example, j=1 represents the first shot MRI image; Collect the SCCA detection data of cervical cancer patients in the tumor marker examination before cervical cancer metastasis, classify them according to the corresponding cervical cancer patients, and sort them in time order from near to far according to the SCCA detection time. According to the shooting time of the MRI image, for any MRI image, the SCCA detection data closest in time scale is recorded as the SCCA feature of the MRI image, and the SCCA features of all MRI images are repeatedly obtained. After completion, the SCCA feature sequence is obtained; for example, in the same year, an SCCA test was performed on April 2, an MRI image was taken on May 2, and another SCCA test was performed on May 10. The MRI image on May 2 is closer to the SCCA test on May 10, so the SCCA test result on May 10 is recorded as the SCCA feature of the MRI image on May 2; because the time of taking the MRI image of cervical cancer and the time of detecting SCCA may not be the same, in order to facilitate subsequent processing, the two need to be aligned in time scale; The MRI image data and the SCCA detection data are divided into a non-metastasis group and a metastasis group according to whether the corresponding cervical cancer patient has cervical cancer metastasis, and the MRI image data and the SCCA detection data are merged and stored according to the corresponding cervical cancer patient, and marked as the first MRI image-related data; In the specific implementation process, SCCA is the abbreviation of Squamous Cell Carcinoma Antigen. SCCA is a glycoprotein, which is a tumor-associated antigen isolated from cervical squamous cell carcinoma tissue. Under normal physiological conditions, the content of SCCA in human blood is extremely low. However, when related malignant tumors such as cervical squamous cell carcinoma occur, cancer cells will produce and release a large amount of SCCA, resulting in increased levels of SCCA in the blood. By detecting the concentration of SCCA in the blood, it can be used as an auxiliary means to help diagnose cervical cancer and other related diseases. The higher the SCCA level of cervical cancer patients, the more malignant the tumor cells may be and the greater the risk of metastasis. Moreover, if the SCCA level continues to rise, it often means that the tumor is progressing, and the cancer cells may be in an active state of proliferation and invasion, and the risk of metastasis also increases accordingly.

[0019] The feature extraction module includes a feature extraction unit and a feature vector unit. The feature extraction unit performs transfer feature extraction processing based on the first MRI image related data. The feature vector unit is used to construct a transfer feature vector and the second MRI image related data. The feature extraction unit is configured with a feature extraction strategy, which includes: graying all MRI images in the first MRI image-related data to obtain gray image data; for any gray image in the gray image data, marking the pixel points in the lesion area in the gray image as cancer pixel points and marking the pixel points at the edge of the lesion area as cancer edge pixel points, and performing area feature acquisition, first shape feature acquisition, and second shape feature acquisition; The area feature acquisition includes: counting the number of cancer pixels in the lesion area, obtaining the actual area represented by each pixel in the corresponding grayscale image, calculating the area size of the lesion area, marking it as the area feature of the lesion area, denoted as Ba; the formula for calculating the area size of the lesion area is as follows: area size = number of pixels * actual area represented by the pixels. For example, a lesion area has 80 pixels, and the actual area represented by each pixel is 0.2mm*0.2mm, then the area size of the lesion area = 80*0.2*0.2=3.2mm 2 ; The first shape feature acquisition includes: Figure 3As shown, the coordinates of the cancer edge pixel point in the corresponding grayscale image are obtained. In this embodiment, the image coordinate system is followed. The upper left corner of the first pixel in the grayscale image starts from , where (0, 0) represents the pixel in the upper left corner, and x represents the column coordinate, and y represents the row coordinate; and the cancer edge pixel point is randomly selected as the starting pixel point, and the coordinate sequence of the cancer edge pixel point is obtained in a clockwise direction, which is recorded as G. i =(x i ,y i ), i=0, 1, 2, ..., n; for any cancer edge pixel, the edge curvature is calculated by the first edge formula, which is as follows: , where k i represents the edge curvature of the i-th cancer edge pixel, Δx i =x i+1 -x i , Δy i =y i+1 -y i ; The edge curvature of all cancer edge pixels is obtained, recorded as an edge curvature sequence, and the first shape feature of the lesion area is calculated by the first shape feature formula. The first shape feature formula is as follows: , where Bb represents the first shape feature, k0 represents the average value of the edge curvature sequence; The first shape feature measures the uniformity of the distribution of cervical cancer lesions on the edge; the larger the first shape feature, the more dramatic the change in curvature at different positions on the edge; if the edge of the cervical cancer lesion area has a regular shape, such as a circle or an ellipse, the curvature is relatively evenly distributed on the edge, and the first shape feature is smaller; when there are multiple protrusions, depressions or other complex shapes on the edge, the first shape feature will increase; the larger the first shape feature, the more dramatic the change in the edge of the cervical cancer lesion area, which means that the cancer cells may have higher heterogeneity; this heterogeneity makes cancer cells more diverse in biological behavior, and some cancer cells may have stronger invasion and migration capabilities, and the risk of metastasis is greater; The second shape feature acquisition includes: based on the edge curvature sequence, for any edge curvature k i , to determine whether k is satisfied i >k i-1 And k i >k i+1, if it is satisfied, the edge curvature is marked as the peak curvature, and the number of peak curvatures is counted, denoted as c1, which represents the number of areas on the edge where the local curvature changes drastically; more peak curvatures mean that there are multiple places where the local curvature changes significantly, that is, there are more irregular shapes such as protrusions or depressions; and the average value of all peak curvatures is calculated, denoted as c2, which reflects the average level of the degree of local curvature change. The larger average peak value indicates that the areas on the edge where the curvature changes drastically are more curved, which means that the edge shape is more irregular; the second shape feature is calculated by the second shape feature formula, and the second shape feature formula is as follows: , where Bc represents the second shape feature; The second shape feature comprehensively considers the number and degree of local irregular areas; it can well quantify this local complex irregular shape; a large number of peak curvatures indicates that there are many points with drastic curvature changes at the edge of the cervical cancer lesion area, that is, there are a large number of irregular shapes such as protrusions and depressions at the edge, which means that cancer cells have stronger local infiltration ability and are more likely to break through the restrictions of surrounding tissues and grow outward; the average peak size indicates that the degree of these protrusions or depressions is more obvious, and the cancer cells are more invasive; the product of the two, that is, the larger the second shape feature, can better reflect the stronger invasive ability of cancer cells, making it more likely for cancer cells to break through the basement membrane and invade blood vessels or lymphatic vessels, thereby increasing the risk of metastasis of cervical cancer; Marking the area feature, the first shape feature and the second shape feature as basic features, repeatedly obtaining the basic features of all grayscale images, and obtaining a basic feature sequence; Based on the basic feature sequence and the SCCA feature sequence, area change feature acquisition, first shape change feature acquisition, second shape change feature acquisition and SCCA change feature acquisition are performed respectively; The area change feature acquisition, the first shape change feature acquisition, the second shape change feature acquisition and the SCCA change feature acquisition include: when j=1, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are set to 0; that is, the cervical cancer patient only took an MRI image once, and the corresponding change feature cannot be obtained, so it is set to 0; When j=2, that is, the cervical cancer patient took a second MRI image, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature were calculated using the first change feature formula. The first change feature formula is as follows: , where Y j (a, b, c, d) represent the area change feature, the first shape change feature, the second shape change feature, and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d). For example, Y jb represents the first shape change feature of the j-th MRI image; B j (a, b, c, d) represent the area feature, the first shape feature, the second shape change feature, and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d). For example, B j b represents the first shape feature of the MRI images taken j times; T0 is a set standard time interval. In the embodiment, T0 is 1 month. Because the time intervals of MRI images taken by cervical cancer patients are not completely consistent, in order to calculate the change rate based on a unified time scale, a standard time interval needs to be set; T j T is the time corresponding to the jth MRI image; j-1 The time corresponding to the j-1th MRI image, that is, the time corresponding to the last image When j>2, that is, the cervical cancer patient has taken more than two MRI images, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are calculated using the second change feature formula. The second change feature formula is as follows: , Wherein q1 and q2 are set weights, the value range is (0, 1), and q1+q2=1; in this embodiment, q1=0.6, q2=0.4; Mark the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature as basic change features, repeatedly obtain the basic change features of all grayscale images, and obtain a basic change feature sequence; The feature vector unit is configured with a feature vector strategy, and the feature extraction vector includes: normalizing the basic feature sequence and the basic change feature sequence according to the feature type, and scaling all the numerical values ​​to [0, 1]; that is, scaling the numerical values ​​of the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature to [0, 1] respectively; and obtaining the normalized basic feature sequence and the basic change feature respectively; For any MRI image, a feature vector is set, marked as a transfer feature vector, and recorded as R={Ba, Bb, Bc, Bd, Ya, Yb, Yc, Yd}, where R represents the transfer feature vector of the corresponding MRI image; Ba, Bb, Bc, Bd, Ya, Yb, Yc and Yd represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature of the corresponding MRI image in order; each MRI image has a corresponding transfer feature vector; Acquire transfer feature vectors of all MRI images in the first MRI image related data, and mark them as second MRI image related data; In the specific implementation process, the area change feature can intuitively reflect the development trend of the size of the cancer and reflect the proliferation of cells in the cervical cancer lesion area; the first shape change feature and the second shape change feature quantify the complexity and irregularity of the edge of the cervical cancer lesion area from the perspective of the edge morphology of the cervical cancer lesion area, which helps to understand the aggressiveness of cervical cancer; SCCA changes, as a dynamic indicator of tumor markers, can reflect the biological activity and cell metabolism of cervical cancer; calculating the changes in these features can dynamically observe the development of cervical cancer; because the metastasis risk of cervical cancer is not fixed, but changes with time, treatment and other factors; through the combination of these features, the metastasis risk of cervical cancer is comprehensively portrayed from different aspects, avoiding the one-sidedness of a single feature.

[0020] The model building module builds a cervical cancer metastasis risk prediction model based on the second MRI image-related data and the logistic regression model; The model building module is configured with a model building strategy, which includes: using a logistic regression model to build a first metastasis risk prediction model, the first metastasis risk prediction model is as follows: , where P(1) represents the risk probability of transfer; Z1, Z2, Z3, Z4, Z5, Z6, Z7 and Z8 represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature in order; β0, β1, β2, β3, β4, β5, β6, β7 and β8 are the parameters of the first transfer risk prediction model. These parameters need to be initialized to smaller values, for example, 0.003, during model training, because if the initial parameters are too large, it will affect the training of subsequent models; The learning rate of the model training is set to v1, the training round is set to v2, and the batch size is set to v3. In this embodiment, the learning rate v1=0.0005, the training round v2=180, and the batch size v3=16; the first metastasis risk prediction model is trained using the second MRI image-related data to obtain a cervical cancer metastasis risk prediction model; In the specific implementation process, if the performance of the trained model is not ideal, v1, v2, v3 and the parameters of the model can be adjusted to train the model again; the first metastasis risk prediction model is constructed based on the logistic regression model because the model has good interpretability; the coefficients of the model can intuitively reflect the direction and degree of influence of each feature on the risk of cervical cancer metastasis. For example, a positive coefficient indicates that an increase in the feature will lead to an increase in the risk of metastasis, while a negative coefficient indicates the opposite; it can help clinicians to interpret the prediction results in combination with clinical experience and professional knowledge; moreover, the calculation of the logistic regression model is relatively simple, and the output probability value is easy to understand. In clinical practice, it can be easily combined with other clinical information to provide a direct reference for formulating personalized treatment plans and follow-up plans.

[0021] The metastasis prediction module includes a risk prediction unit and a risk calculation unit; the risk prediction unit is used to obtain the metastasis feature vector of the cervical cancer patient and predict the metastasis risk using the cervical cancer metastasis risk prediction model; the risk calculation unit is used to calculate the overall metastasis risk of the cervical cancer patient; See also Figure 4 As shown, the risk prediction unit is configured with a risk prediction strategy, which includes: for the cervical cancer patient to be predicted, obtaining the transfer feature vectors of the MRI images taken all the time, recorded as R1, R2, ..., RF, where F is the total number of shots, and obtaining the corresponding shooting time t1, t2, ..., tF, calculating the time intervals e1, e2, ..., e (F-1) between two adjacent times; for example, e1=t2-t1; inputting R1, R2, ..., RF into the cervical cancer metastasis risk prediction model respectively to obtain the corresponding metastasis risk probabilities P1, P2, ..., PF; The risk calculation unit is configured with a risk calculation strategy, which includes: for any transfer feature vector among P1, P2, ..., PF, record it as Pf; calculate the weight intermediate variable of Pf using the intermediate variable formula, and the intermediate variable formula is as follows: , where Q k f represents the weighted intermediate variable of Pf, eu represents the u-th time interval, and W represents the set attenuation coefficient. In this embodiment, W=0.9; the attenuation coefficient represents a degree of "discount" or "weakening" of the risk factor; it reflects the characteristic that the influence of the risk factor on the final risk assessment result gradually weakens with the passage of time, distance change or the influence of other related factors; for example, in the prediction of cervical cancer metastasis risk, it may be believed that recent examination indicators or events have a greater impact on the current metastasis risk, and the longer the time, the more its influence should gradually decay. The attenuation coefficient here is used to quantify the degree of attenuation of this influence; repeatedly calculate the weighted intermediate variables of P1, P2, ..., PF to obtain the corresponding weighted intermediate variable Q k 1. Qk 2, ..., Q k F; Then use the weight calculation formula to calculate the weight of Rf. The weight calculation formula is as follows: , where Qf represents the weight of Rf, and the weights of P1, P2, ..., PF are calculated repeatedly to obtain the corresponding weights Q1, Q2, ..., QF; The metastasis risk formula is then used to calculate the overall metastasis risk of cervical cancer patients. The metastasis risk formula is as follows: , where P0 represents the overall metastasis risk of patients with cervical cancer; For example, a patient with cervical cancer has had a total of 5 MRI images taken, with the corresponding shooting times t1=1, t2=3, t3=5, t4=8, t5=11, in months, and in the same year; then the time intervals e1=2, e2=2, e1=3, e2=3; the corresponding metastasis risk probabilities for the 5 examinations are P1=0.21, P2=0.32, P3=0.33, P4=0.45, P5=0.58; Calculate the intermediate variables, W=0.9, Q k 1=0.9^(2+2+3+3)=0.35; repeat the calculation, Q k 2=0.43, Q k 3=0.53, Q k 4=0.73, Q k 5=1; Calculate the weight again, Q1=0.35 / (0.35+0.43+0.53+0.73+1)=0.12, repeat the calculation, Q2=0.14, Q3=0.17, Q4=0.24, Q5=0.33; Then calculate the overall transfer risk, overall transfer risk = 0.12*0.21+0.14*0.32+0.17*0.33+0.24*0.45+0.33*0.58=0.43; In the specific implementation process, the overall transfer risk is calculated by the exponential decay weighted method. The advantage is that the examinations closer to the current time account for a larger proportion in the risk combination calculation, making the risk assessment more in line with the dynamic development process of the disease. Moreover, when the examination time intervals are different, this method can reasonably adjust the weights according to the time intervals, emphasizing the importance of recent examinations, but not completely ignoring early examination information, and achieving an effective balance between historical and current examination information in the risk combination calculation. The attenuation coefficient can be adjusted according to the specific situation to improve its applicability and accuracy in various situations.

[0022] Example 2, please refer to Figure 2 As shown, the present application provides a method for establishing a cervical cancer metastasis prediction model, comprising the following steps: Step S1, collecting MRI images and SCCA test data of cervical cancer patients, and classifying them to obtain first MRI image related data; Step S1 includes the following sub-steps: Step S101, collecting MRI images of the lesion area of ​​cervical cancer patients before cervical cancer metastasis, classifying them according to the corresponding cervical cancer patients, and sorting them in time order from far to near according to the shooting time of the MRI images, denoted as j=1, 2, ..., m, where j represents the number of times the MRI images are shot, and marked as MRI image data; Step S102, collecting the detection data of SCCA in the tumor marker examination of cervical cancer patients before cervical cancer metastasis, classifying the cervical cancer patients according to the corresponding cervical cancer patients, and sorting them in chronological order from recent to distant according to the detection time of SCCA; Step S103, according to the shooting time of the MRI image, for any MRI image, the detection data of the SCCA closest to the image on the time scale is recorded as the SCCA feature of the MRI image, and the SCCA features of all MRI images are repeatedly obtained to obtain the SCCA feature sequence after completion; Step S104, dividing the MRI image data and the SCCA detection data into a non-metastatic group and a metastatic group according to whether the corresponding cervical cancer patient has cervical cancer metastasis, and merging and storing the MRI image data and the SCCA detection data according to the corresponding cervical cancer patient, and marking them as the first MRI image-related data.

[0023] Step S2, performing transfer feature extraction processing based on the first MRI image-related data, and constructing a transfer feature vector and the second MRI image-related data; Step S2 includes the following sub-steps: Step S201, grayscale processing is performed on all MRI images in the first MRI image related data to obtain grayscale image data; for any grayscale image in the grayscale image data, pixel points in the lesion area of ​​the grayscale image are marked as cancer pixel points and pixel points at the edge of the lesion area are marked as cancer edge pixel points, and area feature acquisition, first shape feature acquisition and second shape feature acquisition are performed; Step S202, area feature acquisition includes: counting the number of cancer pixels in the lesion area, and acquiring the actual area represented by each pixel in the corresponding grayscale image, calculating the area size of the lesion area, marking it as the area feature of the lesion area, denoted as Ba; Step S203, the first shape feature acquisition includes: acquiring the coordinates of the cancer edge pixel points in the corresponding grayscale image, and randomly selecting the cancer edge pixel points as the starting pixel points, and acquiring the coordinate sequence of the cancer edge pixel points in a clockwise direction, denoted as G i =(x i,y i ), i=0, 1, 2, ..., n; for any cancer edge pixel, the edge curvature is calculated by the first edge formula, which is as follows: , where k i represents the edge curvature of the i-th cancer edge pixel, Δx i =x i+1 -x i , Δy i =y i+1 -y i ; The edge curvature of all cancer edge pixels is obtained, recorded as an edge curvature sequence, and the first shape feature of the lesion area is calculated by the first shape feature formula. The first shape feature formula is as follows: , where Bb represents the first shape feature, k0 represents the average value of the edge curvature sequence; Step S203, the second shape feature acquisition includes: based on the edge curvature sequence, for any edge curvature k i , to determine whether k is satisfied i >k i-1 And k i >k i+1 , if it is satisfied, then mark the edge curvature as the peak curvature, count the number of peak curvatures, record it as c1, and calculate the average value of all peak curvatures, record it as c2, and calculate the second shape feature by the second shape feature formula. The second shape feature formula is as follows: , where Bc represents the second shape feature; Step S204, marking the area feature, the first shape feature and the second shape feature as basic features, repeatedly acquiring the basic features of all grayscale images, and obtaining a basic feature sequence; Step S205, performing area change feature acquisition, first shape change feature acquisition, second shape change feature acquisition, and SCCA change feature acquisition based on the basic feature sequence and the SCCA feature sequence; Step S206, area change feature acquisition, first shape change feature acquisition, second shape change feature acquisition and SCCA change feature acquisition include: when j=1, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are set to 0; Step S207, when j=2, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are calculated respectively using the first change feature formula, and the first change feature formula is as follows: , where Y j(a, b, c, d) represent the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d), respectively. j (a, b, c, d) represent the area feature, the first shape feature, the second shape change feature and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d), T0 is the set standard time interval, T j is the time corresponding to the jth MRI image; Step S208, when j>2, the area change characteristic, the first shape change characteristic, the second shape change characteristic and the SCCA change characteristic are calculated respectively using the second change characteristic formula, and the second change characteristic formula is as follows: , Where q1 and q2 are the set weights, with a value range of (0, 1), and q1+q2=1; Step S209, marking the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature as basic change features, repeatedly acquiring the basic change features of all grayscale images, and obtaining a basic change feature sequence; Step S210, normalizing the basic feature sequence and the basic change feature sequence according to the feature type, scaling all numerical values ​​to [0, 1], and obtaining the normalized basic feature sequence and basic change feature respectively; Step S211, setting a feature vector for any MRI image, marked as a transfer feature vector, denoted as R={Ba, Bb, Bc, Bd, Ya, Yb, Yc, Yd}, where R represents the transfer feature vector of the corresponding MRI image; Ba, Bb, Bc, Bd, Ya, Yb, Yc and Yd represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature of the corresponding MRI image in order, respectively; Step S212: acquiring transfer feature vectors of all MRI images in the first MRI image related data and marking them as second MRI image related data.

[0024] Step S3, constructing a cervical cancer metastasis risk prediction model based on the second MRI image-related data and the logistic regression model; Step S3 includes the following sub-steps: Step S301, constructing a first metastasis risk prediction model using a logistic regression model. The first metastasis risk prediction model is as follows: , where P(1) represents the risk probability of metastasis; Z1, Z2, Z3, Z4, Z5, Z6, Z7 and Z8 represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature in order; β0, β1, β2, β3, β4, β5, β6, β7 and β8 are the parameters of the first metastasis risk prediction model; Step S302, setting the learning rate of model training to v1, the training round to v2, and the batch size to v3; Step S303: Use the second MRI image related data to train the first metastasis risk prediction model to obtain a cervical cancer metastasis risk prediction model.

[0025] Step S4, obtaining the metastasis feature vector of the cervical cancer patient, predicting the metastasis risk using the cervical cancer metastasis risk prediction model, and calculating the overall metastasis risk of the cervical cancer patient; Step S4 includes the following sub-steps: Step S401, for the cervical cancer patient to be predicted, obtain the transfer feature vectors of the MRI images taken previously, recorded as R1, R2, ..., RF, where F is the total number of shots, and obtain the corresponding shooting times t1, t2, ..., tF; Step S402, calculating the time intervals e1, e2, ..., e(F-1) between two adjacent intervals; inputting R1, R2, ..., RF into the cervical cancer metastasis risk prediction model to obtain the corresponding metastasis risk probabilities P1, P2, ..., PF; Step S403: For any transfer feature vector among P1, P2, ..., PF, record it as Pf; calculate the weight intermediate variable of Pf using the intermediate variable formula, and the intermediate variable formula is as follows: , where Q k f represents the weighted intermediate variable of Pf, eu represents the time interval of the uth time, and W represents the set attenuation coefficient; repeatedly calculate the weighted intermediate variables of P1, P2, ..., PF to obtain the corresponding weighted intermediate variable Q k 1. Q k 2, ..., Q k F; Step S404, the weight of Rf is calculated using a weight calculation formula, and the weight calculation formula is as follows: , where Qf represents the weight of Rf, and the weights of P1, P2, ..., PF are calculated repeatedly to obtain the corresponding weights Q1, Q2, ..., QF; Step S405, the overall metastasis risk of cervical cancer patients is calculated using the metastasis risk formula, which is as follows: , where P0 represents the overall metastasis risk of cervical cancer patients.

[0026] Example 3, please refer to Figure 5 As shown, Figure 5 The structural schematic diagram of an electronic device is illustrated, and the electronic device may include: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The memory stores computer-readable instructions, and the processor may call the instructions in the memory. When the computer-readable instructions are executed by the processor, the steps in a method for establishing a cervical cancer metastasis prediction model are executed to achieve the following functions: collecting MRI images and SCCA detection data of cervical cancer patients, and classifying them to obtain first MRI image-related data; performing metastasis feature extraction processing based on the first MRI image-related data, and constructing a metastasis feature vector and second MRI image-related data; constructing a cervical cancer metastasis risk prediction model based on the second MRI image-related data and a logistic regression model; obtaining the metastasis feature vector of cervical cancer patients, predicting the metastasis risk using the cervical cancer metastasis risk prediction model, and calculating the overall metastasis risk of cervical cancer patients.

[0027] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0028] Embodiment 4, the present application also provides a computer-readable storage medium, the present application provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method for establishing a cervical cancer metastasis prediction model are executed to achieve the following functions: collect MRI images and SCCA detection data of cervical cancer patients, and classify them to obtain first MRI image-related data; perform metastasis feature extraction and processing based on the first MRI image-related data, and construct a metastasis feature vector and second MRI image-related data; construct a cervical cancer metastasis risk prediction model based on the second MRI image-related data and a logistic regression model; obtain the metastasis feature vector of cervical cancer patients, predict the metastasis risk using the cervical cancer metastasis risk prediction model, and calculate the overall metastasis risk of cervical cancer patients.

[0029] Through the description of the above implementation methods, the embodiments of the present invention can be provided as methods, systems or computer program products. Based on such an understanding, the above technical solutions can be essentially or partly contributed to the prior art in the form of software products, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and include several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0030] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or units can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of systems, modules and units can be electrical, mechanical or other forms.

[0031] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for establishing a cervical cancer metastasis prediction model, characterized in that: The steps include: Collect MRI images and SCCA test data of cervical cancer patients, classify them, and obtain first MRI image related data; Performing transfer feature extraction processing based on the first MRI image-related data, and constructing a transfer feature vector and the second MRI image-related data; A cervical cancer metastasis risk prediction model was constructed based on the second MRI image-related data and the logistic regression model; The metastasis feature vectors of cervical cancer patients were obtained, the metastasis risk was predicted using the cervical cancer metastasis risk prediction model, and the overall metastasis risk of cervical cancer patients was calculated.

2. The method for establishing a cervical cancer metastasis prediction model according to claim 1, characterized in that: Collecting MRI images and SCCA test data of cervical cancer patients and classifying them to obtain first MRI image-related data includes the following sub-steps: Collect MRI images of the lesion area of ​​cervical cancer patients before cervical cancer metastasis, classify them according to the corresponding cervical cancer patients, and sort them in time order from far to near according to the shooting time of the MRI images, denoted as j=1, 2, ..., m, where j represents the number of times the MRI image is shot, and is marked as MRI image data; The SCCA detection data of cervical cancer patients in tumor marker examinations before cervical cancer metastasis are collected, classified according to the corresponding cervical cancer patients, and sorted in chronological order from recent to distant according to the SCCA detection time. According to the shooting time of the MRI image, for any MRI image, the SCCA detection data closest in time scale is recorded as the SCCA feature of the MRI image, and the SCCA features of all MRI images are repeatedly obtained. After completion, the SCCA feature sequence is obtained; The MRI image data and the SCCA detection data are divided into a non-metastatic group and a metastatic group according to whether the corresponding cervical cancer patient has cervical cancer metastasis, and the MRI image data and the SCCA detection data are merged and stored according to the corresponding cervical cancer patient, and marked as the first MRI image-related data.

3. The method for establishing a cervical cancer metastasis prediction model according to claim 2, characterized in that: Performing transfer feature extraction based on the first MRI image-related data and constructing a transfer feature vector and the second MRI image-related data includes the following sub-steps: grayscale processing is performed on all MRI images in the first MRI image related data to obtain grayscale image data; for any grayscale image in the grayscale image data, pixel points in the lesion area of ​​the grayscale image are marked as cancer pixel points and pixel points at the edge of the lesion area are marked as cancer edge pixel points, and area feature acquisition, first shape feature acquisition and second shape feature acquisition are performed; The area feature acquisition includes: counting the number of cancer pixels in the lesion area, obtaining the actual area represented by each pixel in the corresponding grayscale image, calculating the area size of the lesion area, marking it as the area feature of the lesion area, recorded as Ba; The first shape feature acquisition includes: obtaining the coordinates of the cancer edge pixel points in the corresponding grayscale image, and randomly selecting the cancer edge pixel points as the starting pixel points, and obtaining the coordinate sequence of the cancer edge pixel points in a clockwise direction, which is recorded as G i =(x i ,y i ), i=0, 1, 2, ..., n; for any cancer edge pixel, the edge curvature is calculated by the first edge formula, which is as follows: , where k i represents the edge curvature of the i-th cancer edge pixel, Δx i =x i+1 -x i , Δy i =y i+1 -y i ; The edge curvature of all cancer edge pixels is obtained, recorded as an edge curvature sequence, and the first shape feature of the lesion area is calculated by the first shape feature formula. The first shape feature formula is as follows: , where Bb represents the first shape feature, k0 represents the average value of the edge curvature sequence; The second shape feature acquisition includes: based on the edge curvature sequence, for any edge curvature k i , to determine whether k is satisfied i >k i-1 And k i >k i+1 , if it is satisfied, then mark the edge curvature as the peak curvature, count the number of peak curvatures, record it as c1, and calculate the average value of all peak curvatures, record it as c2, and calculate the second shape feature by the second shape feature formula. The second shape feature formula is as follows: , where Bc represents the second shape feature; The area feature, the first shape feature and the second shape feature are marked as basic features, and the basic features of all grayscale images are repeatedly obtained to obtain a basic feature sequence.

4. The method for establishing a cervical cancer metastasis prediction model according to claim 3, characterized in that: Performing transfer feature extraction based on the first MRI image related data and constructing a transfer feature vector and the second MRI image related data also includes the following sub-steps: Based on the basic feature sequence and the SCCA feature sequence, area change feature acquisition, first shape change feature acquisition, second shape change feature acquisition and SCCA change feature acquisition are performed respectively; Acquiring the area change feature, acquiring the first shape change feature, acquiring the second shape change feature and acquiring the SCCA change feature includes: when j=1, setting the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature to 0; When j=2, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are calculated respectively using the first change feature formula. The first change feature formula is as follows: , where Y j (a, b, c, d) represent the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d), respectively. j (a, b, c, d) represent the area feature, the first shape feature, the second shape change feature and the SCCA change feature corresponding to the j-th MRI image in the order of (a, b, c, d), T0 is the set standard time interval, T j is the time corresponding to the jth MRI image; When j>2, the area change characteristic, the first shape change characteristic, the second shape change characteristic and the SCCA change characteristic are calculated respectively using the second change characteristic formula, and the second change characteristic formula is as follows: ， Where q1 and q2 are the set weights, with a value range of (0, 1), and q1+q2=1; The area change feature, the first shape change feature, the second shape change feature and the SCCA change feature are marked as basic change features, and the basic change features of all grayscale images are repeatedly obtained to obtain a basic change feature sequence.

5. The method for establishing a cervical cancer metastasis prediction model according to claim 4, characterized in that: Performing transfer feature extraction based on the first MRI image related data and constructing a transfer feature vector and the second MRI image related data also includes the following sub-steps: The basic feature sequence and the basic change feature sequence are normalized according to the feature type, and all the numerical values ​​are scaled to [0, 1]; the normalized basic feature sequence and basic change feature are obtained respectively; For any MRI image, a feature vector is set, marked as a transfer feature vector, and recorded as R={Ba, Bb, Bc, Bd, Ya, Yb, Yc, Yd}, where R represents the transfer feature vector of the corresponding MRI image; Ba, Bb, Bc, Bd, Ya, Yb, Yc and Yd represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature of the corresponding MRI image in order; The transfer feature vectors of all MRI images in the first MRI image-related data are obtained and marked as the second MRI image-related data.

6. The method for establishing a cervical cancer metastasis prediction model according to claim 5, characterized in that: The construction of a cervical cancer metastasis risk prediction model based on the second MRI image-related data and the logistic regression model includes the following sub-steps: The first metastasis risk prediction model was constructed using the logistic regression model. The first metastasis risk prediction model is as follows: , where P(1) represents the risk probability of transfer; Z1, Z2, Z3, Z4, Z5, Z6, Z7 and Z8 represent the area feature, the first shape feature, the second shape feature, the SCCA feature, the area change feature, the first shape change feature, the second shape change feature and the SCCA change feature in order; β 0, β1, β2, β3, β4, β5, β6, β7, and β8 are parameters of the first metastasis risk prediction model; The learning rate of the model training is set to v1, the training round is set to v2, and the batch size is set to v3. The first metastasis risk prediction model is trained using the second MRI image-related data to obtain a cervical cancer metastasis risk prediction model.

7. The method for establishing a cervical cancer metastasis prediction model according to claim 6, characterized in that: Obtaining the metastasis feature vector of cervical cancer patients, predicting the metastasis risk using the cervical cancer metastasis risk prediction model, and calculating the overall metastasis risk of cervical cancer patients includes the following sub-steps: For patients with cervical cancer to be predicted, the metastasis feature vectors of all MRI images taken are obtained, recorded as R1, R2, ..., RF, where F is the total number of shots, and the corresponding shooting time t1, t2, ..., tF is obtained, and the time intervals between two adjacent times e1, e2, ..., e (F-1) are calculated; R1, R2, ..., RF are respectively input into the cervical cancer metastasis risk prediction model to obtain the corresponding metastasis risk probabilities P1, P2, ..., PF; For any transfer feature vector in P1, P2, ..., PF, it is recorded as Pf; the weighted intermediate variable of Pf is calculated using the intermediate variable formula, and the intermediate variable formula is as follows: , where Q k f represents the weighted intermediate variable of Pf, eu represents the time interval of the uth time, and W represents the set attenuation coefficient; repeatedly calculate the weighted intermediate variables of P1, P2, ..., PF to obtain the corresponding weighted intermediate variable Q k 1. Q k 2, ..., Q k F; Then use the weight calculation formula to calculate the weight of Rf. The weight calculation formula is as follows: , where Qf represents the weight of Rf, and the weights of P1, P2, ..., PF are calculated repeatedly to obtain the corresponding weights Q1, Q2, ..., QF; The metastasis risk formula is then used to calculate the overall metastasis risk of cervical cancer patients. The metastasis risk formula is as follows: , where P0 represents the overall metastasis risk of cervical cancer patients.

8. A system for establishing a cervical cancer metastasis prediction model, applicable to a method for establishing a cervical cancer metastasis prediction model according to any one of claims 1 to 7, characterized in that: It includes a sample collection module, a feature acquisition module, a model building module and a transfer prediction module; The sample collection module is used to collect MRI images of the lesion area of ​​cervical cancer patients and SCCA detection data, and classify them to obtain first MRI image related data; The feature extraction module includes a feature extraction unit and a feature vector unit, the feature extraction unit performs transfer feature extraction processing based on the first MRI image related data, and the feature vector unit is used to construct a transfer feature vector and the second MRI image related data; The model building module builds a cervical cancer metastasis risk prediction model based on the second MRI image-related data and a logistic regression model; The metastasis prediction module includes a risk prediction unit and a risk calculation unit; the risk prediction unit is used to obtain the metastasis feature vector of cervical cancer patients and predict the metastasis risk using a cervical cancer metastasis risk prediction model; the risk calculation unit is used to calculate the overall metastasis risk of cervical cancer patients.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in the method according to any one of claims 1 to 7 are executed.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 7 are executed.

Images