An artificial intelligence-based ovarian cancer risk assessment method

Abstract

The application discloses an ovarian cancer risk assessment method based on artificial intelligence, relates to the field of intelligent medical treatment, and comprises the following steps: acquiring multi-modal dynamic image data and physiological characteristic parameters of the ovary and pelvic cavity of a user; performing space-time alignment and feature enhancement processing on the multi-modal dynamic image data, and performing abnormal value calibration and standardization conversion on the physiological characteristic parameters, so that a unified dimension fusion data set is constructed; and the application integrates multiple types of image and physiological data, accurately processes and analyzes features, can efficiently identify ovarian cancer suspected cases and abnormal areas, accurately predicts future incidence probability, supports early screening and risk early warning, provides reliable reference for clinical diagnosis and treatment, and improves evaluation efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare technology, specifically to an artificial intelligence-based method for assessing ovarian cancer risk. Background Technology

[0002] Ovarian cancer is a common malignant tumor of the female reproductive system. Early symptoms are often subtle and easily overlooked, while late-stage cancer is prone to metastasis and spread. Common symptoms include abdominal distension, abdominal pain, and abnormal vaginal bleeding. Its onset is related to factors such as genetics and hormones.

[0003] The invention patent application with application number 202510768590.0 discloses an artificial intelligence-based ovarian cancer risk assessment auxiliary system. This application aims to solve the problems of "traditional ovarian feature extraction methods having details lost, inaccurate contour extraction, incomplete feature extraction and insufficient data utilization; traditional ovarian cancer risk assessment model construction methods having a single kernel function, unreasonable penalty terms, incomplete objective functions and unclear optimal conditions; and traditional model performance improvement methods having unclear parameter performance evaluation, insufficient diversity maintenance, imperfect competition mechanism, unreasonable compensation step size and insufficient crossover variation".

[0004] However, most current medical technology research and development for ovarian cancer focuses on intelligent diagnosis, with few effective technologies for ovarian cancer risk assessment and prediction.

[0005] Therefore, we propose an artificial intelligence-based method for ovarian cancer risk assessment. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides an artificial intelligence-based method for ovarian cancer risk assessment, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions;

[0008] This invention discloses an artificial intelligence-based method for ovarian cancer risk assessment, comprising:

[0009] This process involves acquiring multimodal dynamic imaging data and physiological characteristic parameters of the user's ovaries and pelvis; performing spatiotemporal alignment and feature enhancement on the multimodal dynamic imaging data, and outlier calibration and standardization on the physiological characteristic parameters to construct a unified fusion dataset; mining the fusion dataset to extract dynamic morphological features of ovarian tissue, functional metabolic correlation features, and physiological parameter coupling features to form a multidimensional risk feature vector; inputting the multidimensional risk feature vector into a pre-trained diagnostic model, and outputting a suspected diagnosis of ovarian cancer and abnormal area identification through feature matching and probability calculation; based on the diagnostic results and the multidimensional risk feature vector, constructing a temporal correlation model of ovarian cancer progression risk according to the user's historical physiological parameters, and outputting the probability of ovarian cancer incidence in different future time windows through feature temporal evolution analysis; and integrating the diagnostic results and the predicted incidence probability to generate a risk assessment report.

[0010] The multimodal dynamic imaging data includes structural and functional images at different time points, and the physiological characteristic parameters include metabolic and immune indicators.

[0011] Furthermore, when performing spatiotemporal alignment on multimodal dynamic image data, the spatiotemporal alignment accuracy follows the following rules:

[0012] ;

[0013] In the formula: This refers to spatiotemporal alignment error; This represents the total number of image time nodes; The number of pixels in a single frame of image; The image grayscale value of the i-th pixel at the t-th time node; The image grayscale value of the i-th pixel at the t-th time node after alignment; Set a preset alignment error threshold; Represents the L2 norm;

[0014] The feature enhancement processing logic for multimodal dynamic image data is as follows:

[0015] ;

[0016] In the formula: To enhance the feature value of the i-th pixel at the t-th time node; This represents the feature value of the i-th pixel at the t-th time node. For enhancement coefficient; These are the maximum and minimum values ​​of the original eigenvalues; The organization saliency score is given for the i-th pixel at the t-th time node.

[0017] Furthermore, the outlier calibration, or standardization transformation, of the physiological characteristic parameters follows:

[0018] ;

[0019] In the formula: These are the calibrated physiological characteristic parameter values; These are the original physiological characteristic parameter values; For calibration coefficients; This is the sample mean of this physiological characteristic parameter; The standard deviation is the sample standard deviation. These are the standardized physiological characteristic parameter values; Let be the dynamic mean of the physiological characteristic parameters corresponding to the t-th time point; Let be the dynamic standard deviation at time point t; It serves as a time-series stability adjustment factor.

[0020] Furthermore, the dynamic morphological characteristics of the ovarian tissue include:

[0021] Contour complexity: , Indicates the perimeter of the ovarian tissue outline. This indicates the average volume of ovarian tissue. This represents the total number of time points. Let be the volume of the ovarian tissue at time point t; This represents the volume of the ovarian tissue at time point t-1.

[0022] The rate of volume change is the ratio of the volume difference between adjacent time points to the volume of the previous time point.

[0023] Shape irregularity is characterized by the ratio of contour fitting error to the total area of ​​the contour.

[0024] Furthermore, the functional metabolic correlation features include the uniformity of metabolic activity distribution, the correlation coefficient between functional imaging signal intensity and metabolic indicators, and the temporal synergistic change rate among different metabolic indicators.

[0025] Among them, the uniformity of metabolic activity distribution is characterized by the ratio of the standard deviation of metabolic activity value to the mean; the correlation coefficient between functional imaging signal intensity and metabolic indicators is characterized by the linear correlation between time series data of functional imaging signal intensity and time series data of corresponding metabolic indicators calculated by Pearson correlation analysis; the temporal synergistic change rate between different metabolic indicators is characterized by the ratio of the product of the changes in two metabolic indicators at adjacent time points to the sum of the absolute values ​​of the changes in the two indicators.

[0026] Furthermore, the pre-trained diagnostic model includes a feature fusion layer, an attention mechanism layer, and a probability output layer, as well as a fully connected layer between the attention mechanism layer and the probability output layer;

[0027] The feature fusion layer maps image modal features and physiological parameter modal features to a feature subspace of a unified dimension, respectively, to obtain feature vectors for each modality. , For modal indices, 1 ≤ m ≤ M; then, the modal attention weight matrix is ​​used. Calculate the global attention weights for each modality. The The features are obtained by normalizing the cosine similarity between each modal feature and the preset modal importance benchmark vector, and finally cross-modal feature fusion is performed: ;

[0028] In the formula: This is the fused high-dimensional feature vector; The total number of modes; The feature optimization matrix within the modality;

[0029] The attention mechanism layer calculates the attention weights for each dimension of the fused features as follows:

[0030] ;

[0031] In the formula: This represents the attention weight for the k-th feature dimension after fusion. This represents the significance score of the k-th feature dimension; This represents the total dimension of the fused features. The significance score for the j-th feature dimension;

[0032] The probability output layer inputs the attention-weighted fused features into the fully connected layer and outputs a logits vector containing both positive and negative ovarian cancer classes. , Let represent the logits values ​​for the positive and negative categories, respectively. Then, the logits vector is converted into a probability distribution for suspected diagnoses using the following formula:

[0033] ;

[0034] In the formula: The probability of a positive diagnosis for suspected ovarian cancer;

[0035] when When the threshold is exceeded, it is judged as a suspected positive result for ovarian cancer. If the threshold is not exceeded, it is judged as a suspected negative result for ovarian cancer.

[0036] Furthermore, the expression for the time-series association model of ovarian cancer progression risk is as follows:

[0037] ;

[0038] In the formula: This represents the probability of onset in the T-th time window. For model bias terms; The total number of risk characteristics; The weight coefficient for the k-th risk feature; Let be the predicted value of the k-th risk feature in the T-th time window; This represents the total number of historical time points. Let be the temporal influence coefficient at the t-th historical time point; Let be the coupled feature value of the physiological parameters at the t-th historical time node.

[0039] Furthermore, the aforementioned ;

[0040] In the formula: This represents the actual observed value of the k-th risk feature at the most recent historical time node n; This represents the change in the k-th risk feature at the s-th and s+1-th historical time points; Let be the time-series decay coefficient of the k-th risk feature; Let be the trend confidence coefficient of the k-th risk feature in the s-th time interval; The volatility adjustment factor for the k-th risk characteristic; Let be the historical volatility standard deviation of the k-th risk feature.

[0041] Furthermore, the aforementioned Characterized by the intensity of the interaction between metabolic and immune indicators:

[0042] ;

[0043] In the formula: The coupling strength is a physiological parameter. The number of metabolic indicators and the number of immune indicators; Let be the i-th standardized metabolic index value; Let j be the standardized immune indicator value; Let be the interaction correlation coefficient between the i-th metabolic indicator and the j-th immune indicator.

[0044] Furthermore, the abnormal area identifier is determined by linking image feature anomaly scoring with spatial positioning;

[0045] The abnormal score ;

[0046] In the formula: , , Preset weights for morphological differences, abnormal functional metabolism, and abnormal coupling of physiological parameters; The degree of morphological difference between the target region and normal ovarian tissue; This is a quantitative value for the degree of abnormality of functional metabolic indicators; This is a quantification of the degree of abnormality in the coupling characteristics of physiological parameters;

[0047] Will Areas exceeding a preset abnormal threshold are marked as suspected abnormal areas for ovarian cancer, and the three-dimensional spatial coordinates of the area are output simultaneously.

[0048] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:

[0049] This invention provides an artificial intelligence-based method for ovarian cancer risk assessment. During execution, this method integrates multimodal dynamic imaging and physiological characteristic parameters, and through precise processing, improves data consistency and effectiveness. It comprehensively captures abnormalities in ovarian tissue morphology, functional metabolism, and physiological parameters, enhancing the comprehensiveness and accuracy of risk identification. It rapidly outputs suspected diagnostic results and spatial localization of abnormal areas, reducing missed diagnoses and misdiagnoses. Furthermore, it combines historical data to construct a temporal correlation model, accurately predicting future incidence probabilities. This provides clinicians with a comprehensive assessment basis that combines immediate diagnosis and long-term risk warning, thereby assisting in early intervention and personalized treatment, and improving the effectiveness and predictability of ovarian cancer risk assessment. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating an artificial intelligence-based method for assessing ovarian cancer risk. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] The present invention will be further described below with reference to embodiments.

[0054] Example:

[0055] This embodiment presents an artificial intelligence-based method for ovarian cancer risk assessment, such as... Figure 1 As shown, it includes:

[0056] Acquire multimodal dynamic imaging data and physiological characteristic parameters of the user's ovaries and pelvis;

[0057] Spatiotemporal alignment and feature enhancement processing are performed on multimodal dynamic image data, and outlier calibration and standardization transformation are performed on physiological feature parameters to construct a unified fusion dataset;

[0058] When performing spatiotemporal alignment on multimodal dynamic image data, the spatiotemporal alignment accuracy follows the following rules:

[0059] ;

[0060] In the formula: This refers to spatiotemporal alignment error; This represents the total number of image time nodes; The number of pixels in a single frame of image; The image grayscale value of the i-th pixel at the t-th time node; The image grayscale value of the i-th pixel at the t-th time node after alignment; Set a preset alignment error threshold; Represents the L2 norm;

[0061] The above formula quantifies the accuracy of spatiotemporal alignment by calculating the ratio of the sum of the L2 norms of the grayscale values ​​of the original image and the aligned image. At the same time, it introduces a preset error threshold to achieve quantitative control of the alignment effect, realize overall error control of all time nodes and all pixels, and ensure the accuracy of subsequent feature extraction.

[0062] The feature enhancement processing logic for multimodal dynamic image data is as follows:

[0063] ;

[0064] In the formula: To enhance the feature value of the i-th pixel at the t-th time node; This represents the feature value of the i-th pixel at the t-th time node. For enhancement coefficient; These are the maximum and minimum values ​​of the original eigenvalues; The organization saliency score is given for the i-th pixel at the t-th time node;

[0065] The above formula combines the distribution differences of the original feature values ​​with the tissue saliency features. It achieves feature enhancement by multiplying the original feature values ​​with a dynamically adjusted enhancement factor. The enhancement coefficient is dynamically adapted according to the degree of difference between the original features. The tissue saliency score integrates texture similarity, grayscale contrast and spatial location weights to achieve targeted enhancement of multimodal image features.

[0066] in, >0 indicates that the more insignificant the difference between the original feature value and other features, the larger the value; conversely, the more significant the difference, the smaller the value. The determination is made by weighted sum of texture similarity, grayscale contrast and spatial location weights between the tissue to which the i-th pixel belongs at the t-th time node and the ovarian target tissue;

[0067] For structural images, such as CT and MRI, the feature value is the gray value of the image pixel; for functional images, such as PET and DWI, the feature value is the functional signal value of the image pixel.

[0068] The standardization transformation for calibrating outliers of physiological characteristic parameters follows the following rules:

[0069] ;

[0070] In the formula: These are the calibrated physiological characteristic parameter values; These are the original physiological characteristic parameter values; For calibration coefficients; This is the sample mean of this physiological characteristic parameter; The standard deviation is the sample standard deviation. These are the standardized physiological characteristic parameter values; Let be the dynamic mean of the physiological characteristic parameters corresponding to the t-th time point; Let be the dynamic standard deviation at time point t; It serves as a time-series stability adjustment factor;

[0071] The above formula addresses the potential abnormal deviations and temporal fluctuations in the original physiological characteristic parameters. First, it uses an exponential function that integrates the sample mean and standard deviation to perform flexible calibration of outliers. The calibration coefficient is dynamically adjusted according to the degree of deviation. Then, it combines the temporal dynamic mean and standard deviation for standardization transformation. The temporal stability adjustment factor adapts to the fluctuation amplitude at different time points, achieving the dual goals of outlier correction and unified scale across time series data, thereby improving the reliability and comparability of physiological parameters.

[0072] in, >0 indicates that the larger the deviation of the original physiological characteristic parameter from the sample mean, the larger the value; the smaller the deviation, the smaller the value. >0 indicates that the value of a physiological characteristic parameter is larger when the fluctuation range is greater at different time points, and smaller when the fluctuation range is smaller.

[0073] The fused dataset was mined to extract dynamic morphological features of ovarian tissue, functional metabolic correlation features, and physiological parameter coupling features in sequence, forming a multidimensional risk feature vector.

[0074] The dynamic morphological characteristics of ovarian tissue include:

[0075] Contour complexity: , Indicates the perimeter of the ovarian tissue outline. This indicates the average volume of ovarian tissue. This represents the total number of time points. Let be the volume of the ovarian tissue at time point t; This represents the volume of the ovarian tissue at time point t-1.

[0076] The rate of volume change is the ratio of the volume difference between adjacent time points to the volume of the previous time point.

[0077] The irregularity of shape is characterized by the ratio of the contour fitting error to the total area of ​​the contour.

[0078] Functional metabolic association features include the uniformity of metabolic activity distribution, the correlation coefficient between functional imaging signal intensity and metabolic indicators, and the temporal synergistic change rate among different metabolic indicators;

[0079] Among them, the uniformity of metabolic activity distribution is characterized by the ratio of the standard deviation of metabolic activity value to the mean; the correlation coefficient between functional imaging signal intensity and metabolic index is characterized by the linear correlation between the time series data of functional imaging signal intensity and the corresponding time series data of metabolic index calculated by Pearson correlation analysis; the temporal synergistic change rate between different metabolic indexes is characterized by the ratio of the product of the changes of two metabolic indexes at adjacent time points to the sum of the absolute values ​​of the changes of the two indexes.

[0080] The multidimensional risk feature vector is input into the pre-trained diagnostic model, and through feature matching and probability calculation, the suspected diagnosis result of ovarian cancer and the abnormal area identification are output.

[0081] The pre-trained diagnostic model includes a feature fusion layer, an attention mechanism layer, a probability output layer, and a fully connected layer between the attention mechanism layer and the probability output layer;

[0082] The feature fusion layer maps image modal features and physiological parameter modal features to a feature subspace of uniform dimension, thereby obtaining feature vectors for each modality. , For modal indices, 1 ≤ m ≤ M; then, the modal attention weight matrix is ​​used. Calculate the global attention weights for each modality. , The features are obtained by normalizing the cosine similarity between each modal feature and the preset modal importance benchmark vector, and finally cross-modal feature fusion is performed: ;

[0083] In the formula: This is the fused high-dimensional feature vector; The total number of modes; The feature optimization matrix within the modality;

[0084] The above formula first maps different modal features to a unified feature subspace. Then, it calculates global attention weights by combining modal attention weight matrices with cosine similarity to achieve adaptive allocation of the importance of each modality. Finally, it optimizes single modal features by using intramodal feature optimization matrices and performs weighted fusion. This not only solves the problem of inconsistent dimensions of multimodal features, but also dynamically adjusts the weights according to the matching degree between the modality and the reference vector, ensuring the relevance and effectiveness of the fused features.

[0085] The attention mechanism layer calculates the attention weights for each dimension of the fused features as follows:

[0086] ;

[0087] In the formula: This represents the attention weight for the k-th feature dimension after fusion. This represents the significance score of the k-th feature dimension; This represents the total dimension of the fused features. The significance score for the j-th feature dimension;

[0088] The above formula is based on the significance scores of each feature dimension after fusion. It is normalized by the softmax function to obtain the attention weight of each dimension. This enables the model to automatically focus on the feature dimensions that contribute more to the diagnostic results, suppress the interference of irrelevant features, realize the adaptive selection of feature dimensions, improve the accuracy of subsequent probability calculations, and enhance the model's ability to capture key risk features.

[0089] The probability output layer inputs the attention-weighted fused features into the fully connected layer, and outputs a logits vector containing both positive and negative ovarian cancer classes. , Let represent the logits values ​​for the positive and negative categories, respectively. Then, the logits vector is converted into a probability distribution for suspected diagnoses using the following formula:

[0090] ;

[0091] In the formula: The probability of a positive diagnosis for suspected ovarian cancer;

[0092] The above formula converts the logits vector output by the fully connected layer into a probability distribution through the softmax function, which intuitively quantifies the suspected probability of ovarian cancer. By calculating the exponential ratio of the logits values ​​of the positive and negative categories, it ensures that the probability value is within a reasonable range, providing a clear quantitative basis for the diagnostic results. At the same time, it is convenient to combine with preset thresholds to quickly determine the suspected results, thereby improving the efficiency of the diagnostic process.

[0093] when When the threshold is exceeded, it is judged as a suspected positive result for ovarian cancer. If the threshold is not exceeded, the result is considered a negative result for suspected ovarian cancer.

[0094] in, The intra-class scatter and inter-class separation based on each modal feature are determined through least squares iterative optimization. , It is obtained by weighting the matching degree and variance contribution of its feature dimensions with the core pathological indicators in the ovarian cancer pathological feature database;

[0095] Abnormal area identification is determined by linking image feature anomaly scoring with spatial positioning.

[0096] Abnormal scoring ;

[0097] In the formula: , , Preset weights for morphological differences, abnormal functional metabolism, and abnormal coupling of physiological parameters; The degree of morphological difference between the target region and normal ovarian tissue; This is a quantitative value for the degree of abnormality of functional metabolic indicators; This is a quantification of the degree of abnormality in the coupling characteristics of physiological parameters;

[0098] The above formula integrates three core indicators: the degree of morphological difference, the degree of abnormality in functional metabolism, and the degree of abnormality in the coupling of physiological parameters. The contribution of each indicator is dynamically adjusted by preset weight coefficients. The weight values ​​are adaptively optimized according to the correlation strength between different indicators and the suspected ovarian cancer diagnosis, so as to comprehensively quantify the multi-dimensional abnormal information and provide a reasonable score for the accurate identification of abnormal areas.

[0099] Will Areas exceeding a preset abnormal threshold are marked as suspected abnormal areas for ovarian cancer, and the three-dimensional spatial coordinates of the area are output simultaneously.

[0100] in, The values ​​are determined by comparing the contour parameters, volume parameters, and morphological irregularity parameters of the target region with those of normal ovarian tissue. The values ​​are determined by calculating the deviation of actual measured values ​​of functional metabolic indicators from the normal reference range, combined with the ratio of the differences after standardization. The value is determined by calculating the deviation between the results of the physiological parameter coupling strength calculation and the normal coupling benchmark value, combined with the abnormal deviation rate of the coupling correlation coefficient.

[0101] in, , , The value range is preset to [0.2, 0.8]. When the contribution of morphological difference to the suspected ovarian cancer determination is higher than that of functional metabolic abnormalities and abnormal coupling of physiological parameters, The larger the value, the weaker the diagnostic indicative power of morphological differences compared to the latter two. The smaller the value, the stronger the correlation between abnormal functional metabolic indicators and pathological features of ovarian cancer. The larger the value, the weaker the correlation is compared to abnormal coupling of morphological differences and physiological parameters. The smaller the value, the more significant the impact of abnormal interactions between metabolic and immune indicators on risk assessment. The larger the value, the greater the impact, even when it is lower than morphological differences and functional metabolic abnormalities. The smaller the value;

[0102] Based on the diagnostic results and multidimensional risk feature vectors, a time-series correlation model of ovarian cancer progression risk is constructed according to the user's historical physiological parameters. The probability of ovarian cancer incidence in different time windows in the future is output through feature time-series evolution analysis.

[0103] The expression for the time-series association model of ovarian cancer progression risk is:

[0104] ;

[0105] In the formula: This represents the probability of onset in the T-th time window. For model bias terms; The total number of risk characteristics; The weight coefficient for the k-th risk feature; Let be the predicted value of the k-th risk feature in the T-th time window; This represents the total number of historical time points. Let be the temporal influence coefficient at the t-th historical time point; The physiological parameter coupling feature value at the t-th historical time node;

[0106] The above formula integrates the future predicted value of risk features with the coupling features of historical physiological parameters. By weighted summing of model bias terms, risk feature weight coefficients and time series influence coefficients, and combining the sigmoid function, the result is mapped to the probability of disease occurrence. The risk feature weights are adjusted according to the strength of their association with ovarian cancer, and the time series influence coefficients change dynamically with the time interval between historical nodes and the prediction window, thereby achieving accurate time series prediction of the risk of disease occurrence in different future time windows.

[0107] in, The initial value range is set to [-0.8, 0.5]. The higher the degree of abnormality of the coupling characteristics of historical physiological parameters and the larger the sum of risk feature weights, the smaller the value. The lower the value is, the lower the sum of risk feature weights and the higher the value is, the better. The initial value range is set to [0.02, 0.35]. The stronger the statistical association between this risk characteristic and the incidence of ovarian cancer, and the higher the differentiation between benign and malignant lesions in clinical samples, the larger the value will be. The weaker the association with the incidence of ovarian cancer and the lower the differentiation, the smaller the value will be. The initial value range is set to [0.1, 1.5]. The smaller the time interval between the historical time node and the prediction time window T, the higher the matching degree between the physiological parameter coupling characteristics of that node and the pathological characteristics of ovarian cancer, and the larger the value; conversely, the larger the time interval, the smaller the value.

[0108] ;

[0109] In the formula: This represents the actual observed value of the k-th risk feature at the most recent historical time node n; This represents the change in the k-th risk feature at the s-th and s+1-th historical time points; Let be the time-series decay coefficient of the k-th risk feature; Let be the trend confidence coefficient of the k-th risk feature in the s-th time interval; The volatility adjustment factor for the k-th risk characteristic; Let be the historical volatility standard deviation of the k-th risk characteristic;

[0110] The above formula is based on the feature observations of the most recent historical nodes, superimposed with the feature changes of each historical time interval, and combined with the time-series decay coefficient, trend confidence coefficient and fluctuation adjustment factor to dynamically correct the feature change trend and fluctuation impact. The time-series decay coefficient adapts to feature stability, the trend confidence coefficient reflects the consistency of change, and the fluctuation adjustment factor is associated with clinical relevance, ultimately achieving accurate prediction of future changes in risk features.

[0111] in, The preset value range is (0.5, 10]. The larger the value is when the temporal stability of the feature itself is stronger, the smaller the value is when the temporal stability is weaker. The value range is preset to (0,1]. The higher the consistency between the feature change in the s-th time interval and the change in the adjacent time interval, the larger the value; the lower the consistency, the smaller the value. The preset value range is (0.1, 2.0]. The higher the clinical relevance of the feature to the diagnosis of ovarian cancer, the larger the value; the lower the clinical relevance, the smaller the value.

[0112] Characterized by the intensity of the interaction between metabolic and immune indicators:

[0113] ;

[0114] In the formula: The coupling strength is a physiological parameter. The number of metabolic indicators and the number of immune indicators; Let be the i-th standardized metabolic index value; Let j be the standardized immune indicator value; Let be the interaction correlation coefficient between the i-th metabolic indicator and the j-th immune indicator;

[0115] The above formula quantifies the interaction strength between metabolic and immune indicators by calculating the weighted sum of the cross-products of the two indicators and dividing it by the square root of the sum of squares of the two types of indicators. The interaction correlation coefficient is dynamically adjusted according to the synergistic effect of the indicators in the pathological process. It takes into account both the standardized values ​​of individual indicators and highlights the interaction correlation between indicators, providing a quantitative basis for the abnormal assessment of the coupling characteristics of physiological parameters.

[0116] in, The initial value range is preset to [-1.0, 1.0]. The higher the synergistic effect of the i-th metabolic indicator and the j-th immune indicator in the pathological process of ovarian cancer, the larger its absolute value. When there is no significant correlation between the two, the value approaches 0.

[0117] Integrate diagnostic results with predicted incidence rates to generate a risk assessment report;

[0118] The multimodal dynamic imaging data includes structural and functional images at different time points, and the physiological characteristic parameters include metabolic and immune indicators.

[0119] The method described in the above embodiments integrates multiple types of imaging and physiological data. Through precise processing and feature analysis, it can efficiently identify suspected cases and abnormal areas of ovarian cancer, accurately predict the future incidence probability, support early screening and risk warning, provide reliable reference for clinical diagnosis and treatment, and improve assessment efficiency and accuracy.

[0120] The following is an application example of the method described in the above embodiments:

[0121] A 35-year-old female user underwent an ovarian cancer risk assessment due to pelvic discomfort. Medical personnel first obtained multimodal dynamic imaging data of her ovaries and pelvis (including CT and MRI structural images and PET and DWI functional images at different time points), as well as physiological characteristic parameters such as metabolic and immune indicators.

[0122] Data preprocessing: Spatiotemporal alignment was performed on the multimodal dynamic image data. The calculated spatiotemporal alignment error was 0.03, which is less than the preset threshold of 0.05, meeting the accuracy requirements. Subsequently, feature enhancement processing was performed, combining the differences in original feature values ​​with tissue significance scores, to optimize the feature values ​​of each pixel, making them easier to distinguish. Outlier calibration was performed on physiological feature parameters to correct parameter values ​​that deviated from the sample mean. Then, standardization transformation was performed to adapt the data to subsequent analysis standards, ultimately constructing a unified fusion dataset.

[0123] Feature extraction: Multidimensional risk feature vectors were extracted from the fused dataset. In terms of ovarian tissue morphological dynamic features, the contour complexity was calculated to be 1.8, the average volume change rate was 0.07, and the morphological irregularity was 0.12. In terms of functional metabolic correlation features, the uniformity of metabolic activity distribution was 0.35, the correlation coefficient between functional image signal intensity and metabolic indicators was 0.72, and the average temporal co-change rate among different metabolic indicators was 0.41. At the same time, physiological parameter coupling features were extracted to complete the construction of multidimensional risk feature vectors.

[0124] Diagnosis and Anomaly Identification: A multidimensional risk feature vector is input into a pre-trained diagnostic model. After feature fusion and attention weight allocation, the model outputs a positive probability of 0.86 for a suspected ovarian cancer diagnosis, exceeding the preset threshold of 0.7, thus classifying it as a suspected ovarian cancer case. Anomaly region scoring is calculated, with morphological difference weights, functional metabolic abnormality weights, and physiological parameter coupling abnormality weights set to 0.3, 0.4, and 0.3 respectively, resulting in an anomaly score of 0.82, higher than the preset anomaly threshold of 0.6. The three-dimensional spatial coordinates of the abnormal region (x=125.3, y=89.6, z=92.1) are simultaneously output.

[0125] Ovarian cancer incidence prediction and report generation: A time-series correlation model for ovarian cancer progression risk is constructed by combining the user's historical physiological parameters. Through feature time-series evolution analysis, the predicted incidence rate of ovarian cancer in the next year is 0.78, and the incidence rate in the next three years is 0.89. Finally, the suspected diagnosis results, abnormal area information, and incidence rate prediction values ​​are integrated to generate an ovarian cancer risk assessment report that includes the assessment process, core indicators, risk level, and clinical recommendations, providing a reference for subsequent diagnosis and treatment.

[0126] In summary, the methods described in the above embodiments integrate multimodal dynamic imaging and physiological characteristic parameters during execution, improve data consistency and effectiveness through precise processing, comprehensively capture abnormalities in ovarian tissue morphology, functional metabolism, and physiological parameters, enhance the comprehensiveness and accuracy of risk identification, quickly output suspected diagnostic results and spatial location of abnormal areas, reduce missed diagnoses and misdiagnoses, and construct a time-series correlation model based on historical data to accurately predict future incidence probabilities. This provides clinicians with a comprehensive assessment basis that combines immediate diagnosis and long-term risk warning, thereby assisting in early intervention and personalized treatment, and improving the effectiveness and predictability of ovarian cancer risk assessment.

[0127] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based method for ovarian cancer risk assessment, characterized in that, include: Acquire multimodal dynamic imaging data and physiological characteristic parameters of the user's ovaries and pelvis; Spatiotemporal alignment and feature enhancement processing are performed on multimodal dynamic image data, and outlier calibration and standardization transformation are performed on physiological feature parameters to construct a unified fusion dataset; When performing spatiotemporal alignment on multimodal dynamic image data, the spatiotemporal alignment accuracy follows the following rules: ； In the formula: This refers to spatiotemporal alignment error; This represents the total number of image time nodes; The number of pixels in a single frame of image; The image grayscale value of the i-th pixel at the t-th time node; The image grayscale value of the i-th pixel at the t-th time node after alignment; Set a preset alignment error threshold; Represents the L2 norm; The feature enhancement processing logic for multimodal dynamic image data is as follows: ； In the formula: To enhance the feature value of the i-th pixel at the t-th time node; This represents the feature value of the i-th pixel at the t-th time node. For enhancement coefficient; These are the maximum and minimum values ​​of the original eigenvalues; The organization saliency score is given for the i-th pixel at the t-th time node; The outlier calibration and standardization transformation of the physiological characteristic parameters follows the following: ； In the formula: These are the calibrated physiological characteristic parameter values; These are the original physiological characteristic parameter values; For calibration coefficients; This is the sample mean of this physiological characteristic parameter; The standard deviation is the sample standard deviation. These are the standardized physiological characteristic parameter values; Let be the dynamic mean of the physiological characteristic parameters corresponding to the t-th time point; Let be the dynamic standard deviation at time point t; It serves as a time-series stability adjustment factor; The fused dataset was mined to extract dynamic morphological features of ovarian tissue, functional metabolic correlation features, and physiological parameter coupling features in sequence, forming a multidimensional risk feature vector. The multidimensional risk feature vector is input into the pre-trained diagnostic model, and through feature matching and probability calculation, the suspected diagnosis result of ovarian cancer and the abnormal area identification are output. Based on the diagnostic results and multidimensional risk feature vectors, a time-series correlation model of ovarian cancer progression risk is constructed according to the user's historical physiological parameters. The probability of ovarian cancer incidence in different time windows in the future is output through feature time-series evolution analysis. Integrate diagnostic results with predicted incidence rates to generate a risk assessment report; The multimodal dynamic imaging data includes structural and functional images at different time points, and the physiological characteristic parameters include metabolic and immune indicators.

2. The artificial intelligence-based ovarian cancer risk assessment method according to claim 1, characterized in that, The dynamic morphological characteristics of the ovarian tissue include: Contour complexity: , Indicates the perimeter of the ovarian tissue outline. This indicates the average volume of ovarian tissue. This represents the total number of time points. Let be the volume of the ovarian tissue at time point t; This represents the volume of the ovarian tissue at time point t-1. The rate of volume change is the ratio of the volume difference between adjacent time points to the volume of the previous time point. Shape irregularity is characterized by the ratio of contour fitting error to the total area of ​​the contour.

3. The artificial intelligence-based ovarian cancer risk assessment method according to claim 1, characterized in that, The functional metabolic correlation features include the uniformity of metabolic activity distribution, the correlation coefficient between functional imaging signal intensity and metabolic indicators, and the temporal synergistic change rate among different metabolic indicators. Among them, the uniformity of metabolic activity distribution is characterized by the ratio of the standard deviation of metabolic activity value to the mean; the correlation coefficient between functional imaging signal intensity and metabolic indicators is characterized by the linear correlation between time series data of functional imaging signal intensity and time series data of corresponding metabolic indicators calculated by Pearson correlation analysis; the temporal synergistic change rate between different metabolic indicators is characterized by the ratio of the product of the changes in two metabolic indicators at adjacent time points to the sum of the absolute values ​​of the changes in the two indicators.

4. The artificial intelligence-based ovarian cancer risk assessment method according to claim 1, characterized in that, The pre-trained diagnostic model includes a feature fusion layer, an attention mechanism layer, and a probability output layer, as well as a fully connected layer between the attention mechanism layer and the probability output layer. The feature fusion layer maps image modal features and physiological parameter modal features to a feature subspace of a unified dimension, respectively, to obtain feature vectors for each modality. , For modal indexes, 1 ≤ m ≤ M; Then through the modal attention weight matrix Calculate the global attention weights for each modality. The The features are obtained by normalizing the cosine similarity between each modal feature and the preset modal importance benchmark vector, and finally cross-modal feature fusion is performed: ; In the formula: This is the fused high-dimensional feature vector; The total number of modes; The feature optimization matrix within the modality; The attention mechanism layer calculates the attention weights for each dimension of the fused features as follows: ； In the formula: This represents the attention weight for the k-th feature dimension after fusion. This represents the significance score of the k-th feature dimension; This represents the total dimension of the fused features. The significance score for the j-th feature dimension; The probability output layer inputs the attention-weighted fused features into the fully connected layer and outputs a logits vector containing both positive and negative ovarian cancer classes. , Let represent the logits values ​​for the positive and negative categories, respectively. Then, the logits vector is converted into a probability distribution for suspected diagnoses using the following formula: ； In the formula: The probability of a positive diagnosis for suspected ovarian cancer; when When the threshold is exceeded, it is judged as a suspected positive result for ovarian cancer. If the threshold is not exceeded, it is judged as a suspected negative result for ovarian cancer.

5. The artificial intelligence-based ovarian cancer risk assessment method according to claim 1, characterized in that, The expression for the time-series association model of ovarian cancer progression risk is: ； In the formula: This represents the probability of onset in the T-th time window. For model bias terms; The total number of risk characteristics; The weight coefficient for the k-th risk feature; Let be the predicted value of the k-th risk feature in the T-th time window; This represents the total number of historical time points. Let be the temporal influence coefficient at the t-th historical time point; Let be the coupled feature value of the physiological parameters at the t-th historical time node.

6. The artificial intelligence-based ovarian cancer risk assessment method according to claim 5, characterized in that, The ; In the formula: This represents the actual observed value of the k-th risk feature at the most recent historical time node n; This represents the change in the k-th risk feature at the s-th and s+1-th historical time points; Let be the time-series decay coefficient of the k-th risk feature; Let be the trend confidence coefficient of the k-th risk feature in the s-th time interval; The volatility adjustment factor for the k-th risk characteristic; Let be the historical volatility standard deviation of the k-th risk feature.

7. The artificial intelligence-based ovarian cancer risk assessment method according to claim 5, characterized in that, The Characterized by the intensity of the interaction between metabolic and immune indicators: ； In the formula: The coupling strength is a physiological parameter. The number of metabolic indicators and the number of immune indicators; Let be the i-th standardized metabolic index value; Let j be the standardized immune indicator value; Let be the interaction correlation coefficient between the i-th metabolic indicator and the j-th immune indicator.

8. The artificial intelligence-based ovarian cancer risk assessment method according to claim 1, characterized in that, The abnormal area identifier is determined by linking image feature anomaly scoring with spatial positioning. The abnormal score ; In the formula: , , Preset weights for morphological differences, abnormal functional metabolism, and abnormal coupling of physiological parameters; The degree of morphological difference between the target region and normal ovarian tissue; This is a quantitative value for the degree of abnormality of functional metabolic indicators; This is a quantification of the degree of abnormality in the coupling characteristics of physiological parameters; Will Areas exceeding a preset abnormal threshold are marked as suspected abnormal areas for ovarian cancer, and the three-dimensional spatial coordinates of the area are output simultaneously.

Images