Breast benign nodule follow-up visit data processing and nodule risk early warning method

Abstract

The invention relates to the field of medical image analysis and artificial intelligence, and discloses a mammary gland benign nodule follow-up visit data processing and nodule risk early warning method, which comprises the following steps of: obtaining gray-scale ultrasonic images and color Doppler blood flow images of a baseline and a plurality of follow-up visit time points; an improved U-Net model is adopted to automatically segment the nodule area, and morphological features and blood flow features are extracted; constructing a feature change rate matrix based on the percentage change rate of each time point feature relative to the baseline, quantifying a dynamic evolution trend in combination with a time decay weight matrix, and generating a dynamic malignant risk score by coupling patient age and breast cancer high risk factors; and further dynamically optimizing a risk judgment threshold value through a deep reinforcement learning model, dividing low, medium and high risk levels according to a comparison result of the score and the threshold value, and outputting differentiated clinical intervention suggestions. According to the invention, accurate and personalized early warning of the malignant risk of the breast benign nodule is realized, and intelligent support is provided for clinical follow-up decision.

Description

Technical Field

[0001] The present invention relates to the field of medical image analysis and artificial intelligence technology, and specifically to a method for follow-up data processing and nodule risk early warning of benign breast nodules. Background Art

[0002] Follow-up and malignancy risk assessment of benign breast nodules present complex challenges in clinical practice. Although most benign and malignant nodules can be differentiated based on ultrasound imaging morphology, echogenicity, and blood flow characteristics, some malignant lesions exhibit features similar to benign nodules (such as regular morphology, uniform echogenicity, or low-velocity blood flow) in a single examination. Relying solely on static features can easily lead to misdiagnosis, necessitating long-term follow-up of 6 to 24 months to observe their dynamic evolution and clarify their nature.

[0003] At the same time, some benign nodules may increase in size or change in morphology during follow-up, further interfering with diagnostic accuracy. Existing technologies are mostly based on feature extraction and fixed threshold judgment at a single time point. They are unable to effectively quantify the dynamic changes in parameters such as lesion area, long diameter, short diameter, and blood flow percentage over time. They also fail to incorporate the impact of individual patient differences (such as age and breast cancer risk factors) on risk assessment, resulting in insufficient accuracy in identifying "dynamic malignancy" or "pseudo-progressive benign lesions."

[0004] In addition, traditional methods rely on manual outlining of the lesion area and are susceptible to measurement cursor errors, image watermark interference, and multimodal image spatial registration deviations, making it difficult to ensure the consistency and repeatability of feature extraction.

[0005] How to use automation technology to achieve dynamic feature analysis of multi-time point and multi-modal images and establish an adaptive risk warning model based on clinical factors has become a key issue in optimizing the existing follow-up decision-making process. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention provides a follow-up data processing and nodule risk warning method for benign breast nodules, which solves the problem of insufficient accuracy of malignant risk warning of benign breast nodules in the existing technology due to reliance on single static feature evaluation, fixed risk threshold and lack of coupling with clinical factors.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for processing follow-up data of benign breast nodules and early warning of nodule risks, comprising the following steps: Grayscale ultrasound images and color Doppler blood flow images were obtained from patients at baseline and at least two follow-up time points; Automatically segmenting the nodule area in the grayscale ultrasound image and the color Doppler blood flow image, and extracting the morphological features and blood flow features at the baseline time point and each follow-up time point; Based on the characteristic values ​​at each follow-up time point, the characteristic change rate relative to the baseline time point was calculated, and combined with the patient's age and high-risk factors for breast cancer to generate a dynamic malignancy risk score; According to the comparison result of the dynamic malignancy risk score and the preset threshold, a nodule risk warning signal is output.

[0008] Preferably, the at least two follow-up time points include at least three time points among 6 months, 12 months, 18 months and 24 months, and the time interval is an integer multiple of 6 months.

[0009] Preferably, in the step of extracting the morphological characteristics and blood flow characteristics at the baseline time point and each follow-up time point: The morphological characteristics include: Area A: the total number of pixels of the nodule area in the grayscale ultrasound image, converted to the actual area according to the image scale; Long diameter L: the maximum length of the first principal component direction of the principal component analysis of the nodule area; Short diameter S: the maximum length of the second principal component direction of the nodule area; Length ratio R = L / S; The blood flow feature is the pixel ratio B of the color Doppler blood flow signal in the nodule area, which is calculated as follows: Where M is the binary mask of the nodule area, M(x,y)∈{0,1} indicates whether the pixel (x,y) belongs to the nodule; I c (x, y) is the blood flow signal intensity at pixel (x, y) in the color Doppler image; T is the preset blood flow signal intensity threshold; δ(·) is the indicator function, which takes the value 1 when the condition is met and 0 otherwise.

[0010] Preferably, in the step of automatically segmenting the nodule area in the grayscale ultrasound image and the color Doppler blood flow image: Automatic segmentation is achieved through an improved U-Net model, with grayscale ultrasound images and color Doppler blood flow images as input and binary masks as output; Before segmentation, a mutual information-based image registration algorithm was used to align grayscale ultrasound images and color Doppler blood flow images to ensure that the spatial position of the same nodule was consistent in both images.

[0011] Preferably, the step of calculating the characteristic change rate relative to the baseline time point based on the characteristic value at each follow-up time point includes: For each follow-up time point t i , calculate the percentage change rate of each feature relative to the baseline time point t0: X∈{A,L,S,R,B} Among them, X t0 is the characteristic value at the baseline time point t0; The follow-up time point t i A, L, S, R, and B are the area, major diameter, minor diameter, aspect ratio, and blood flow ratio, respectively.

[0012] Preferably, the step of generating a dynamic malignancy risk score based on the patient's age and breast cancer risk factors includes: Construct the feature change rate matrix C, whose elements are the feature change rates at each time point; Construct the time-attenuated weight matrix W, whose diagonal elements are: The off-diagonal elements are 0, where α>0 is the attenuation coefficient optimized by historical data; Calculate the weighted feature matrix C W =W·C, and extract its trace as a dynamic evolution trend indicator, the trace is defined as the matrix C W Sum of the main diagonal elements: Combined with the patient's age A g and breast cancer risk factor marker R f ∈{0,1}, generate the final score: Score=β1·Trace(C W )+β2·A g +β3·R f Among them, β1, β2, and β3 are predefined weight coefficients.

[0013] Preferably, the weight coefficients β1, β2, and β3 are determined by genetic algorithm optimization, and the objective function is to maximize the sum of the sensitivity and specificity of malignant nodule classification.

[0014] Preferably, the step of outputting a nodule risk warning signal based on the comparison result of the dynamic malignancy risk score with a preset threshold comprises: The nodule risk level is determined based on the comparison between the dynamic malignancy risk score Score and the preset threshold θ: Among them, Δ θ is the preset buffer threshold; Clinical intervention measures are recommended, including at least one of continued follow-up, needle biopsy, or surgical resection.

[0015] Preferably, the preset threshold θ is dynamically optimized through a deep reinforcement learning model, specifically including: State space definition: state variable s t Including patient age A g , high-risk factor identification R f And the historical feature change rate matrix C; action space definition: action variable a t is the threshold adjustment δ, and the update rule is: θ t+1 =θ t +δ Reward function design: Define the immediate reward r based on the pathological diagnosis results t : Strategy Optimization: Maximizing Cumulative Rewards with Deep Q-Networks Where γ is the discount factor.

[0016] The present invention also provides a follow-up data processing and nodule risk warning system for benign breast nodules, comprising: An image acquisition module, used to acquire grayscale ultrasound images and color Doppler blood flow images at baseline and multiple follow-up time points; Image processing module, used to automatically segment nodule areas and extract morphological and blood flow characteristics; A dynamic scoring module is used to calculate the rate of change of features and generate a malignancy risk score in combination with clinical factors; The early warning module is used to output a visual early warning report based on the comparison results between the score and the threshold.

[0017] The present invention provides a method for processing follow-up data of benign breast nodules and for early warning of nodule risks. It has the following beneficial effects: 1. This method continuously tracks follow-up data at multiple time points and uses a time-attenuated weight matrix to weight the rate of change of features, effectively quantifying the dynamic evolution of nodule morphology and blood flow characteristics. Compared to traditional single-shot detection or static assessment methods, it can more objectively reflect the biological behavior of nodules, avoid misjudgments caused by occasional measurement errors, and improve the temporal sensitivity and accuracy of malignancy risk assessment.

[0018] 2. This invention combines grayscale ultrasound morphological features and color Doppler blood flow signal distribution with clinical data such as patient age and high-risk factors to generate a dynamic score using a linear weighted model. This multidimensional information fusion mechanism overcomes the limitations of single imaging modalities or clinical indicators, significantly improving the sensitivity and specificity of malignancy risk classification and providing more comprehensive data support for clinical decision-making.

[0019] 3. This invention dynamically optimizes the risk assessment threshold through a deep reinforcement learning model, adapting to differences in the distribution of malignant risk across different patient groups (e.g., differences in malignancy rates between younger and older patients), thus avoiding overdiagnosis or missed diagnosis caused by fixed thresholds. This mechanism enables the early warning system to have online learning and iteration capabilities, continuously improving its warning accuracy as clinical data accumulates.

[0020] 4. This invention uses an improved U-Net model for automatic nodule segmentation and combines it with a mutual information registration algorithm to ensure spatial consistency of multimodal images, significantly reducing the cost of manual delineation and alignment. This fully automated process reduces subjective errors and improves data processing efficiency, making it suitable for long-term follow-up management of large clinical cohorts.

[0021] 5. This invention utilizes a three-level risk classification mechanism with a buffered threshold design, combined with differentiated clinical intervention recommendations (follow-up, biopsy, or surgery), to provide physicians with a step-by-step decision-making framework. This design avoids frequent level switching caused by minor score fluctuations, while also complying with the individualized management principles of clinical guidelines, effectively balancing diagnostic sensitivity with the need for rational resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Schematic diagram of the method flow of the present invention; Figure 2 Schematic diagram of the system architecture of the present invention.

[0023] Among them, 10, image acquisition module; 20, image processing module; 30, dynamic scoring module; 40, early warning module. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. 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 making creative efforts are within the scope of protection of the present invention.

[0025] Please see the attached Figure 1 The present invention provides a follow-up data processing and nodule risk warning method for benign breast nodules, which realizes personalized malignant risk warning by dynamically analyzing the evolution trend of ultrasound image features at multiple time points and combining the patient's clinical factors.

[0026] like Figure 1 As shown, the follow-up data processing and nodule risk warning method for benign breast nodules may include the following steps: S1. Obtain grayscale ultrasound images and color Doppler blood flow images of patients at baseline and follow-up time points; S2, automatically segmenting the nodule area in the grayscale ultrasound image and the color Doppler blood flow image, and extracting the morphological features and blood flow features; S3. Based on the characteristic values ​​at each follow-up time point, calculate the characteristic change rate relative to the baseline time point, and generate a dynamic malignancy risk score based on the patient's age and breast cancer risk factors; S4. Output a nodule risk warning signal based on the comparison result between the dynamic malignancy risk score and the preset threshold.

[0027] The following is a detailed description of each step in the method of the present invention, which comprehensively explains the specific implementation principles, technical details and processes of each step.

[0028] Regarding step S1, in this embodiment, step S1 aims to obtain multimodal ultrasound imaging data of the patient at baseline and multiple follow-up time points, and ensure image quality and spatial consistency through preprocessing technology, thereby providing reliable input for subsequent feature extraction and dynamic analysis.

[0029] In this embodiment, the patient's baseline time point is defined as the imaging examination time when the benign breast nodule is first diagnosed, denoted as t0. Follow-up time points must include at least two subsequent examination times (e.g., t1 and t2). The preferred follow-up period is an integer multiple of 6 months, such as 6 months, 12 months, 18 months, or 24 months. By continuously collecting data at multiple time points, the dynamic evolution of nodule morphology and blood flow characteristics can be captured.

[0030] In this embodiment, the grayscale ultrasound image (I g ) is used to display the morphological structure of the nodule. Preferably, its imaging parameters include: probe frequency range 5-12MHz, depth adjustment 4-6cm, gain setting 50-60dB. Color Doppler blood flow image (I c ) is used to reflect the blood flow signal distribution within and around the nodule. Preferably, the blood flow scale is set to 5-10 cm / s, and the filtering parameters are 50-100 Hz. Both images are stored in DICOM format and contain metadata such as pixel spacing, imaging timestamp, and device parameters.

[0031] In this embodiment, the grayscale ultrasound image I g Anisotropic diffusion filtering is performed to suppress speckle noise and preserve edge information. Specifically, the Perona-Malik model is used, and its partial differential equation is defined as: Among them, the diffusion coefficient The calculation formula is: Where, is the image gradient amplitude, k is the edge sensitivity coefficient, and the preferred value range is 30-70. Through iterative calculation (the preferred number of iterations is 5-15 times), uniform regional noise can be effectively smoothed while avoiding edge blur.

[0032] For color Doppler blood flow image I c In this embodiment, the red channel (corresponding to the main component of the blood flow signal) is extracted and converted into a single-channel grayscale image, and then a median filter (preferably a window size of 3×3 or 5×5) is used to remove isolated noise points.

[0033] In this embodiment, the grayscale ultrasound image I g Color Doppler blood flow image I c Spatial alignment is required to ensure that the morphological features of the same nodule are consistent with the location of blood flow feature extraction. To achieve this goal, a rigid registration algorithm based on mutual information (MI) is used. The specific steps include: 1. Region of Interest (ROI) extraction: In the grayscale image I g Manually or automatically select the nodule area as the reference ROI; 2. Transformation model definition: for color Doppler image I c Apply translation Δx, Δy, rotation θ, and scaling s to construct the parameter space T = (Δx, Δy, θ, s); 3. Similarity metric calculation: Mutual information is used as the registration metric, and the calculation formula is: Where p(a,b) is the joint probability distribution of the grayscale of the two images, p(a) and p(b) are the marginal probability distributions; In this embodiment, the pre-processed image data is structured and stored by time point and patient ID, with metadata tags (including acquisition time, device model, registration parameters, etc.) attached. Furthermore, a visualization tool is used to overlay the multimodal images before and after registration, allowing manual verification of spatial alignment accuracy. If significant deviations are found, a re-registration process is triggered.

[0034] Regarding step S2, in this embodiment, step S2 extracts the morphology and blood flow characteristics of the nodule from the registered grayscale ultrasound image and color Doppler blood flow image through multimodal image segmentation and feature quantification technology, providing a quantitative data basis for dynamic malignancy risk assessment.

[0035] In this embodiment, an improved U-Net model is used to achieve accurate segmentation of the nodule area.

[0036] The improved U-Net model introduces a residual connection module in the encoder part to alleviate the gradient vanishing problem in deep network training; A channel attention mechanism is embedded in the skip connection, and background noise interference is suppressed by recalibrating the feature channel weights.

[0037] The input of the model is the superposition of the registered grayscale ultrasound image and the color Doppler blood flow image, and the output is the binary mask M∈{0,1} of the nodule area. H×W , where H and W are the height and width of the image, respectively. Preferably, a weighted combination of the Dice loss function and the cross entropy loss function is used during training to balance segmentation accuracy and boundary smoothness.

[0038] In this embodiment, the following morphological features are calculated based on the segmentation mask M: Area (A): By counting the total number of non-zero pixels in the mask M and combining it with the pixel spacing parameter in the DICOM header file, the number of pixels is converted into the actual physical area (unit: square millimeters). The calculation formula is: A = N pixel ×(Δx×Δy) Among them, N pixel is the number of non-zero pixels in the mask, Δx and Δy are the actual sizes of the pixels in the horizontal and vertical directions, respectively.

[0039] Long diameter (L) and short diameter (S): Principal component analysis (PCA) is performed on the mask M to extract the maximum projection lengths in the first principal component direction and the second principal component direction, which are defined as the long diameter L and the short diameter S, respectively.

[0040] Specifically, the covariance matrix of the mask area is calculated and its eigenvalues ​​and eigenvectors are solved. The maximum length L in the direction of the first principal component and the maximum length S in the direction of the second principal component are determined by the following formulas: Where λ1 and λ2 are the first two eigenvalues ​​of the covariance matrix, and the coefficient 4 is used to convert the standard deviation into actual length (unit: mm).

[0041] Aspect ratio (R): The ratio of the major diameter to the minor diameter reflects the irregularity of the nodule shape. The calculation formula is: In this embodiment, the blood flow signal ratio (B) of the nodule area is calculated based on the registered color Doppler blood flow image. Specifically, the blood flow image is first threshold segmented to extract the pixel area with intensity exceeding the preset threshold T, and then its ratio in the nodule mask M is calculated: Where, I c(x, y) is the blood flow signal intensity at the pixel position (x, y) of the color Doppler image, TT is the preset blood flow intensity threshold, and the preferred value range is the 70th-80th percentile of the blood flow signal intensity distribution; δ(·) is the indicator function, which takes the value of 1 when the condition is met and 0 otherwise.

[0042] In this embodiment, to ensure the spatial consistency of the nodule area in the grayscale ultrasound image and the color Doppler blood flow image, the two modal images are aligned using a mutual information-based image registration technique before segmentation. Specifically, with the grayscale image as a reference, a rigid transformation (translation, rotation, scaling) is applied to the color Doppler image, and parameter optimization is completed by maximizing the mutual information function of the two images. Preferably, the registered image needs to be manually verified using a visualization tool. If the deviation of the center coordinate of the nodule exceeds a preset tolerance (for example, 2 pixels), the re-registration process is triggered.

[0043] Regarding step S3, in this embodiment, step S3 quantifies the dynamic evolution trend of nodule characteristics over time and constructs a comprehensive scoring model in combination with the patient's clinical factors to achieve dynamic assessment of the malignant risk of benign breast nodules.

[0044] In this embodiment, based on the baseline time point t0 and each follow-up time point t i (i=1,2,…,n) morphological and blood flow characteristic values, calculate the percentage change rate of each characteristic relative to the baseline. Specifically, for each follow-up time point t i , calculate the characteristic change rate according to the following formula: in, is the characteristic value of the baseline time point t0, The follow-up time point t i The characteristic values ​​of the image are shown in Figure 1. A, L, S, R, and B represent the area, major diameter, minor diameter, length-to-short ratio, and blood flow percentage, respectively. Calculating the percentage change eliminates the interference of baseline differences between individuals and highlights the relative trends of the characteristics over time.

[0045] In this embodiment, in order to quantify the overall trend of feature changes at multiple time points, a feature change rate matrix is ​​constructed: Its rows correspond to follow-up time points, and its columns correspond to feature types. The specific form is: Further introduce the time decay weight matrix Its diagonal elements are defined as: The off-diagonal elements are zero, where α > 0 is the attenuation coefficient, which is preferably determined by grid search or cross-validation of historical data. This weight matrix gives higher weight to recent follow-up data, reflecting the "time recency effect" in the evolution of malignant risk.

[0046] In this embodiment, the time decay weight matrix W is multiplied by the feature change rate matrix C to obtain the weighted feature matrix C W =W·C. By extracting C W Trace as a comprehensive indicator of dynamic evolution trend: The trace is the sum of the main diagonal elements of the matrix. Its physical meaning is the linear superposition of the weighted characteristic change rates at each time point, which can quantify the overall evolution trend of the nodule malignancy risk.

[0047] In this embodiment, combined with the patient's age A g (Unit: years) and breast cancer high risk factor marker R f ∈{0,1} (such as family history, gene mutation, etc.), and generate the final dynamic malignancy risk score through linear weighting: Score=β1·Trace(C W )+β2·A g +β3·R f Among them, β1, β2, and β3 are predefined weight coefficients, preferably determined by genetic algorithm optimization. By introducing clinical factors, the scoring model can simultaneously capture the synergistic effect of nodule biological behavior and individual patient risk attributes.

[0048] In this embodiment, the optimization goal of the weight coefficients β1, β2, and β3 is to maximize the sum of the sensitivity and specificity of malignant nodule classification. Specifically, the objective function is defined as: Where TP is the number of true positives (malignant nodules correctly warned), FN is the number of false negatives (malignant nodules missed), TN is the number of true negatives (benign nodules correctly ruled out), and FP is the number of false positives (benign nodules misdiagnosed). The optimal weight combination is iteratively searched through the selection, crossover, and mutation operations of the genetic algorithm until the objective function converges.

[0049] Regarding step S4, in this embodiment, step S4 generates a nodule risk level classification and clinical intervention recommendations based on the comparison results of the dynamic malignancy risk score and the preset threshold, and dynamically optimizes the threshold parameters through reinforcement learning to achieve personalized risk warning.

[0050] In this embodiment, the nodule risk is divided into three levels: low risk, medium risk and high risk according to the comparison result of the dynamic malignancy risk score (Score) and the preset threshold θ. Specifically, the buffer threshold Δ θ Based on the impact of smoothing score fluctuations on grade determination, the classification rules are as follows: Among them, the buffer threshold Δ θ The preferred value is 10%-20% of the preset threshold θ. The buffer zone design can avoid frequent switching of warning levels caused by slight fluctuations in the score, thereby improving the stability of clinical decision-making.

[0051] For different risk levels, corresponding clinical intervention measures are recommended in this example: Low risk: Regular follow-up for 6-12 months is recommended to observe changes in nodules; Medium risk: It is recommended to shorten the follow-up period to 3-6 months, or combine ultrasound elastography for further evaluation; High risk: Immediate puncture biopsy or surgical resection is recommended to confirm the pathological diagnosis.

[0052] In this embodiment, the preset threshold θ is dynamically optimized through a deep reinforcement learning model to adapt to the differences in risk distribution among different patient groups. Specifically, the following core steps are included: 1. State space definition: state variable s t By patient age A g , breast cancer high risk factor marker R f And the historical characteristic change rate matrix C, that is: s t =(A g ,R f ,C) By integrating individual patient attributes with dynamic imaging features, personalized status representation can be achieved.

[0053] 2. Action space definition: action variable a t Threshold adjustment amount The update rule is: θ t+1 =θ t +δ Preferably, the action space is discretized into a fixed step size (e.g., δ∈{-5,-2.5,0,+2.5,+5}) to balance exploration efficiency and optimization accuracy.

[0054] 3. Reward function design: Define the immediate reward r based on the pathological diagnosis results (gold standard) t , the specific rules are: Through a strict reward and punishment mechanism, we ensure that the model optimization goals are consistent with clinical diagnostic accuracy.

[0055] 4. Strategy optimization algorithm: Deep Q-Network (DQN) is used for strategy learning. The network structure is preferably a double hidden layer fully connected network with the input layer dimension and the state variable s t Matching, the output layer dimension is consistent with the size of the action space. Through experience replay and target network technology, the training process is stabilized to maximize the cumulative discounted reward: Among them, γ∈[0,1) is the discount factor, and the preferred value is 0.9-0.99, which is used to balance current rewards and long-term benefits.

[0056] In this embodiment, the threshold optimization process is performed as follows: Data collection: During clinical use, the patient's early warning results, pathological diagnosis and status variables are recorded in real time; Strategy update: When a sufficient number of new samples (e.g., 100) are accumulated, the DQN model is triggered to conduct online training and update the threshold adjustment strategy. Threshold release: The optimized threshold θ t+1 Synchronize to the early warning system for subsequent patient use.

[0057] In general, the present invention obtains grayscale ultrasound images and color Doppler blood flow images of patients at baseline and multiple follow-up time points, uses an improved U-Net model to automatically segment the nodule area, and extracts morphological features (area, major diameter, minor diameter, aspect ratio) and blood flow features (blood flow proportion); constructs a feature change rate matrix based on the percentage change rate of the features at each time point relative to the baseline, quantifies the dynamic evolution trend in combination with the time attenuation weight matrix, and generates a dynamic malignant risk score by coupling the patient's age and breast cancer high-risk factors; further dynamically optimizes the risk judgment threshold through a deep reinforcement learning model, divides the risk into low, medium, and high levels according to the comparison results between the score and the threshold, and outputs differentiated follow-up or clinical intervention recommendations, thereby realizing accurate malignant risk warning by integrating the dynamic evolution of images with individualized clinical characteristics.

[0058] The follow-up data processing and nodule risk warning system for benign breast nodules described below and the follow-up data processing and nodule risk warning method for benign breast nodules described above can be referenced to each other.

[0059] Please see the attached Figure 2 The present invention also provides a follow-up data processing and nodule risk warning system for benign breast nodules, comprising: An image acquisition module 10 is used to collect grayscale ultrasound images and color Doppler blood flow images at baseline and multiple follow-up time points; an image processing module 20 is used to automatically segment nodule areas and extract morphological and blood flow characteristics; Dynamic scoring module 30, used to calculate the characteristic change rate and generate a malignancy risk score in combination with clinical factors; The early warning module 40 is used to output a visual early warning report based on the comparison result between the score and the threshold.

[0060] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0061] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing follow-up data of benign breast nodules and early warning of nodule risk, characterized in that: The method comprises the following steps: obtaining grayscale ultrasound images and color Doppler blood flow images of the patient at a baseline time point and at least two follow-up time points; Automatically segmenting the nodule area in the grayscale ultrasound image and the color Doppler blood flow image, and extracting the morphological features and blood flow features at the baseline time point and each follow-up time point; Based on the characteristic values ​​at each follow-up time point, the characteristic change rate relative to the baseline time point was calculated, and combined with the patient's age and high-risk factors for breast cancer to generate a dynamic malignancy risk score; According to the comparison result of the dynamic malignancy risk score and the preset threshold, a nodule risk warning signal is output.

2. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 1, characterized in that: The at least two follow-up time points include at least three time points among 6 months, 12 months, 18 months and 24 months, and the time interval is an integer multiple of 6 months.

3. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 1, characterized in that: In the step of extracting the morphological characteristics and blood flow characteristics at the baseline time point and each follow-up time point: The morphological characteristics include: Area A: the total number of pixels of the nodule area in the grayscale ultrasound image, converted to the actual area according to the image scale; Long diameter L: the maximum length of the first principal component direction of the principal component analysis of the nodule area; Short diameter S: the maximum length of the second principal component direction of the nodule area; Length ratio R = L / S; The blood flow feature is the pixel ratio B of the color Doppler blood flow signal in the nodule area, which is calculated as follows: Where M is the binary mask of the nodule area, M(x,y)∈{0,1} indicates whether the pixel (x,y) belongs to the nodule; I c (x, y) is the blood flow signal intensity at pixel (x, y) in the color Doppler image; T is the preset blood flow signal intensity threshold; δ(·) is the indicator function, which takes the value 1 when the condition is met and 0 otherwise.

4. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 3, characterized in that: In the step of automatically segmenting the nodule area in the grayscale ultrasound image and the color Doppler blood flow image: Automatic segmentation is achieved through an improved U-Net model, with grayscale ultrasound images and color Doppler blood flow images as input and binary masks as output; Before segmentation, a mutual information-based image registration algorithm was used to align grayscale ultrasound images and color Doppler blood flow images to ensure that the spatial position of the same nodule was consistent in both images.

5. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 1, characterized in that: The step of calculating the characteristic change rate relative to the baseline time point based on the characteristic value of each follow-up time point includes: for each follow-up time point t i , calculate the percentage change rate of each feature relative to the baseline time point t0: in, is the characteristic value at the baseline time point t0; The follow-up time point t i A, L, S, R, and B are the area, major diameter, minor diameter, aspect ratio, and blood flow ratio, respectively.

6. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 5, characterized in that: The step of generating a dynamic malignancy risk score by combining the patient's age and breast cancer high-risk factors includes: Construct the feature change rate matrix C, whose elements are the feature change rates at each time point; Construct the time-attenuated weight matrix W, whose diagonal elements are: The off-diagonal elements are 0, where α>0 is the attenuation coefficient optimized by historical data; Calculate the weighted feature matrix C W =W·C, and extract its trace as a dynamic evolution trend indicator, the trace is defined as the matrix C W Sum of the main diagonal elements: Combined with the patient's age A g and breast cancer risk factor marker R f ∈{0,1}, generate the final score: Score=β1·Trace(C W )+β2·A g +β3·R f Among them, β1, β2, and β3 are predefined weight coefficients.

7. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 6, characterized in that: The weight coefficients β1, β2, and β3 are determined by genetic algorithm optimization, and the objective function is to maximize the sum of the sensitivity and specificity of malignant nodule classification.

8. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 1, characterized in that: The step of outputting a nodule risk warning signal based on the comparison result of the dynamic malignancy risk score with the preset threshold value includes: nodule risk level classification, which is determined based on the comparison result of the dynamic malignancy risk score Score with the preset threshold value θ: Among them, Δ θ is the preset buffer threshold; Clinical intervention measures are recommended, including at least one of continued follow-up, needle biopsy, or surgical resection.

9. The method for follow-up data processing and nodule risk early warning of benign breast nodules according to claim 8, characterized in that: The preset threshold θ is dynamically optimized through a deep reinforcement learning model, specifically including: State space definition: state variable s t Including patient age A g , high-risk factor identification R f and the historical characteristic change rate matrix C; Action space definition: action variable a t is the threshold adjustment δ, and the update rule is: i t+1 =θ t +d Reward function design: Define the immediate reward r based on the pathological diagnosis results t : Strategy Optimization: Maximizing Cumulative Rewards with Deep Q-Networks Where γ is the discount factor.

10. A follow-up data processing and nodule risk warning system for benign breast nodules, used to execute the method according to any one of claims 1 to 9, characterized in that: include: An image acquisition module, used to acquire grayscale ultrasound images and color Doppler blood flow images at baseline and multiple follow-up time points; Image processing module, used to automatically segment nodule areas and extract morphological and blood flow characteristics; A dynamic scoring module is used to calculate the rate of change of features and generate a malignancy risk score in combination with clinical factors; The early warning module is used to output a visual early warning report based on the comparison results between the score and the threshold.

Images