Adenomyosis diagnosis method and system based on cervical multi-parameter ultrasound imaging
By using cervical multiparameter ultrasound imaging technology, combined with shear wave elastography and ultramicro blood flow imaging, a multivariate logistic regression model was established. This solved the problems of subjectivity and difficulty in early diagnosis of adenomyosis in existing technologies, and achieved quantitative assessment and highly accurate diagnosis of cervical tissue.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL OF GUANGZHOU MEDICAL UNIV (GUANGZHOU RESPIRATORY CENT)
- Filing Date
- 2026-03-04
- Publication Date
- 2026-06-26
AI Technical Summary
Current ultrasound diagnostic techniques rely on morphological observation of the uterine body, which leads to a high degree of subjectivity in the diagnosis of adenomyosis, making early diagnosis difficult. Furthermore, they cannot quantify and capture changes in the biomechanical stiffness and microcirculation remodeling of cervical tissue, resulting in missed diagnoses and insufficient diagnostic sensitivity.
A multi-parameter ultrasound imaging method based on the cervix was adopted, combined with two-dimensional grayscale, shear wave elastography and ultra-micro blood flow imaging, to establish a multivariate logistic regression model. By quantitatively analyzing the biomechanical and hemodynamic parameters of the cervical parenchyma, the probability value of disease risk was output.
It enables objective and quantitative assessment of cervical tissue, improves the diagnostic accuracy and specificity of adenomyosis, overcomes the shortcomings of traditional methods, and is particularly robust in the diagnosis of early lesions and complex cases.
Smart Images
Figure CN122272075A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ultrasound imaging diagnostic technology, specifically to a method and system for diagnosing adenomyosis based on multi-parameter ultrasound imaging of the cervix. Background Technology
[0002] Adenomyosis is a common benign gynecological disease in women of reproductive age. Its pathological characteristic is the invasion of endometrial glands and stroma into the myometrium, often leading to progressive dysmenorrhea, menorrhagia, and infertility, severely impacting patients' quality of life and reproductive health. In clinical diagnosis and treatment, imaging examinations are crucial for early detection, disease grading, and treatment planning. Among these, transvaginal ultrasound, with its advantages of real-time imaging, radiation-free operation, ease of use, and cost-effectiveness, has become the preferred imaging modality for screening and diagnosing this disease.
[0003] Current ultrasound diagnostic strategies primarily rely on direct observation of the morphological features of the uterine body using two-dimensional grayscale imaging technology. Clinicians typically make judgments based on morphological signs such as uterine spherical enlargement, heterogeneous myometrial echogenicity, asymmetrical thickening of the anterior and posterior walls, palisade-like attenuation within the myometrium, or cystic anechoic areas. Some diagnostic procedures also incorporate color Doppler flow imaging to observe the distribution of blood flow signals within the myometrium, or utilize elastography to assess the stiffness of focal lesions in the uterine body to aid in differentiating adenomyosis from uterine fibroids. The core diagnostic logic revolves around macroscopic anatomical abnormalities in the lesion area of the uterine body.
[0004] While existing technologies have some clinical value in typical cases, their limitations lie primarily in their heavy reliance on macroscopic morphological changes of the lesions and the operator's subjective visual experience. In the early stages of the disease or in the diffuse lesion phase, the uterine morphology may not show significant changes, and two-dimensional sonograms lack specific features, leading to insufficient diagnostic sensitivity. Furthermore, when adenomyosis is accompanied by multiple uterine fibroids, the acoustic shadowing and morphological changes of the fibroids often mask the characteristics of adenomyosis, resulting in missed diagnoses. In addition, existing technologies largely focus on assessing the lesions within the uterine body itself, neglecting the fact that adenomyosis, as a diffuse pathological change affecting the entire uterus, causes tissue fibrosis and microcirculatory remodeling that actually extend to the cervical region. Conventional morphological diagnostic methods cannot quantify and capture these potential changes in biomechanical stiffness and microvascular resistance within the cervical parenchyma, making it difficult to provide objective, standardized, and quantitative diagnostic evidence. Summary of the Invention
[0005] The first aspect of this invention provides a diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging. This solves the problems of high diagnostic subjectivity and difficulty in early diagnosis caused by the reliance on uterine morphological observation in existing technologies.
[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides a method for diagnosing adenomyosis based on multi-parameter cervical ultrasound imaging, the method comprising the following steps: Multimodal ultrasound image data, including two-dimensional grayscale ultrasound images, shear wave elastography images, and ultramicro blood flow imaging images, were acquired using a transvaginal ultrasound probe. Based on the longitudinal axis of the cervical canal in the two-dimensional grayscale ultrasound image, an anatomical coordinate system is established, and multiple independent quantitative analysis sub-regions covering the internal cervical os and the middle segment of the cervix are identified as measurement target areas within the cervical parenchyma. Based on the quantitative analysis sub-region, biomechanical parameters reflecting tissue stiffness and hemodynamic parameters reflecting microcirculation resistance are extracted using the shear wave elastography image and the ultramicro blood flow imaging image, respectively. The biomechanical parameters and hemodynamic parameters are input as a multivariate logistic regression model for the diagnostic model of adenomyosis. The diagnostic model for adenomyosis performs linear weighted summation based on weight coefficients and maps the linear predicted values to disease risk probability values through a logistic function. When the disease risk probability value is higher than a preset diagnostic threshold, it outputs a classification conclusion indicating that the patient has adenomyosis.
[0007] Preferably, the step of establishing an anatomical coordinate system based on the longitudinal axis of the cervical canal in the two-dimensional grayscale ultrasound image, and determining multiple independent quantitative analysis sub-regions within the cervical parenchyma, includes: Using the hyperechoic linear structure of the cervical canal as the reference axis, with the internal cervical os as the origin, the cervical longitudinal axis as the Y-axis, and the straight line perpendicular to the cervical longitudinal axis as the X-axis, a two-dimensional plane rectangular coordinate system is established. Based on the two-dimensional Cartesian coordinate system, the measurement target area is divided into four spatially independent regions: the anterior lip region of the internal cervical os, located on the ventral side of the longitudinal axis of the cervical canal and at the same horizontal level as the internal cervical os; the posterior lip region of the internal cervical os, located on the dorsal side of the longitudinal axis of the cervical canal and at the same horizontal level as the internal cervical os; the anterior lip region of the middle segment of the cervix, located on the ventral side of the longitudinal axis of the cervical canal and at a predetermined midpoint of the longitudinal axis length; and the posterior lip region of the middle segment of the cervix, located on the dorsal side of the longitudinal axis of the cervical canal and at a predetermined midpoint of the longitudinal axis length.
[0008] Preferably, when placing a circular measurement sampling frame within the quantitative analysis sub-region, the following boundary constraints are applied: Set the sampling frame diameter to a preset diameter range; constrain the edge of the sampling frame to a preset first safe distance from the hyperechoic line of the cervical canal mucosa to eliminate interference from cervical mucus and mucosal glands; The sampling frame edge is constrained to a preset first safe distance from the cervical serosal layer or vaginal wall boundary to eliminate boundary hardening artifacts caused by surrounding connective tissue and probe compression; If calcifications or cervical glandular cysts are present at the preset location, the sampling frame position is adjusted by horizontally shifting while maintaining the same horizontal plane.
[0009] Preferably, the step of extracting biomechanical parameters reflecting tissue stiffness using the shear wave elastography image includes: Read the shear wave propagation velocity of each pixel in the quantitative analysis sub-region; calculate the Young's modulus value based on the isotropic medium assumption in elasticity, whereby the Young's modulus value is equal to the product of the elastic coefficient, the cervical tissue density, and the square of the shear wave propagation velocity; The arithmetic mean of the Young's modulus values of all valid pixels in the quantitative analysis sub-region is calculated to obtain the average Young's modulus value as the biomechanical parameter.
[0010] Preferably, the step of extracting hemodynamic parameters reflecting microcirculatory resistance using the ultramicroscopic blood flow imaging image includes: Under the guidance of ultra-micro blood flow imaging, pulsed Doppler spectral data were acquired in the anterior lip region of the internal cervical os; the highest flow velocity envelope of the spectral data was tracked using an edge detection algorithm to identify the peak systolic flow velocity and end-diastolic flow velocity during the cardiac cycle. Calculate the drag index, which is equal to the difference between the peak systolic velocity and the end-diastolic velocity divided by the peak systolic velocity. The arithmetic mean of the resistance indices over at least three consecutive cardiac cycles is used to obtain the anterior lip resistance index of the internal cervical os as the hemodynamic parameter.
[0011] Preferably, before constructing the adenomyosis diagnostic model, a third-order cross-feature selection strategy is performed to determine the input variables, the third-order cross-feature selection strategy including: First-order feature dimensionality reduction: Input the standardized full set of features into the minimum absolute shrinkage and selection operator regression model, select the regularization parameter corresponding to the minimum mean square error through ten-fold cross-validation, and retain the features with non-zero regression coefficients to form the first candidate feature subset; Second-order feature ranking: Input all features into a support vector machine model based on a linear kernel function, evaluate feature importance based on the square of the weight vector, and select the feature combination corresponding to the point with the highest classification accuracy by combining five-fold cross-validation to form the second candidate feature subset; Third-order feature importance assessment: Construct a random forest classifier, calculate the average Gini index reduction caused by each feature when splitting nodes, and select the top 50% of features by importance score to form the third candidate feature subset; Feature intersection: Perform an intersection operation on the first candidate feature subset, the second candidate feature subset, and the third candidate feature subset to determine the final core parameter combination.
[0012] Preferably, the core parameter combination specifically includes five parameters: the average Young's modulus of the anterior lip of the internal cervical os, the average Young's modulus of the posterior lip of the internal cervical os, the average Young's modulus of the anterior lip of the mid-cervical segment, the average Young's modulus of the posterior lip of the mid-cervical segment, and the resistance index of the anterior lip of the internal cervical os.
[0013] Preferably, the step of inputting the extracted biomechanical parameters and hemodynamic parameters as input variables into the adenomyosis diagnostic model includes: The mean and standard deviation of the training set features calculated during the model training phase are called; using the mean and standard deviation of the training set features, the extracted parameters of the current subject are Z-score standardized to make the input data follow a standard normal distribution.
[0014] Preferably, the method for determining the preset diagnostic threshold is as follows: the Youden index is calculated based on the receiver operating characteristic curve of the training set, wherein the Youden index is the sum of sensitivity and specificity minus a baseline constant, and the probability value corresponding to the maximum Youden index is selected as the optimal diagnostic cutoff value.
[0015] Preferably, a second aspect of the present invention also provides a diagnostic system for adenomyosis based on multi-parameter cervical ultrasound imaging, the system comprising: The image acquisition module is used to connect to the ultrasound probe and acquire two-dimensional grayscale ultrasound images, shear wave elastography images, and ultra-micro blood flow imaging images of the cervical anatomical region. The target area localization module is used to identify the internal os and cervical canal structure based on the two-dimensional grayscale ultrasound image, establish an anatomical coordinate system with the longitudinal axis of the cervical canal as the reference, and generate a standardized measurement sampling frame in the cervical parenchyma. The parameter calculation module is used to read the image data within the measurement sampling frame, calculate the average value of Young's modulus based on the shear wave propagation velocity, and calculate the drag index based on the Doppler spectrum envelope. The model inference module stores pre-trained multivariate logistic regression model parameters. It is used to receive the standardized Young's modulus mean and resistance index, and calculate the disease risk probability value through linear weighting and logistic function mapping. The results display module is used to compare the disease risk probability value with the optimal diagnostic cutoff value and output the classification conclusion of adenomyosis.
[0016] This invention provides a method and system for diagnosing adenomyosis based on multi-parameter cervical ultrasound imaging. It has the following beneficial effects: 1. This invention achieves quantitative assessment of the biomechanical stiffness and microcirculatory resistance of the cervical parenchyma by combining shear wave elastography and ultramicrovascular blood flow imaging techniques. Compared with traditional diagnostic methods that rely on morphological changes in the uterine body, the Young's modulus and resistance index extracted by this invention directly reflect the pathophysiological changes in the distal cervical tissue caused by adenomyosis, providing objective quantitative indicators for clinical practice and effectively compensating for the insufficient diagnostic sensitivity of conventional two-dimensional ultrasound in early or atypical cases.
[0017] 2. This invention establishes a standardized anatomical coordinate system based on the longitudinal axis of the cervical canal and strict sampling frame boundary constraints. This standardized positioning strategy effectively avoids artificial hardening artifacts caused by cervical mucus reflection, interference from adjacent tissues, and probe pressure, ensuring the consistency of the spatial distribution of the measurement target area among different subjects and operators, and significantly improving the repeatability and measurement accuracy of ultrasound quantitative data.
[0018] 3. This invention employs an integrated three-tiered cross-feature screening strategy to accurately select core parameter combinations, including Young's modulus and resistance index in specific regions, from multimodal data, and constructs a multivariate logistic regression diagnostic model. This model integrates multidimensional tissue stiffness and blood perfusion characteristics through weighted coefficients, outputting a quantified probability value of disease risk. This represents a leap from single-parameter analysis to multi-parameter comprehensive modeling, enhancing the robustness of the diagnostic system in complex clinical scenarios and improving the specificity and accuracy of adenomyosis diagnosis. Attached Figure Description
[0019] Figure 1 This is a flowchart of a diagnostic method for adenomyosis based on cervical multiparameter ultrasound imaging according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a diagnostic system for adenomyosis based on cervical multiparameter ultrasound imaging according to an embodiment of the present invention; Figure 3 This is a flowchart of cervical multimodal ultrasound image data acquisition according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the anatomical localization of the cervical measurement target area according to an embodiment of the present invention; Figure 5 This is a logic block diagram for quantitative parameter extraction according to an embodiment of the present invention; Figure 6 This is a flowchart of a three-order cross-feature screening strategy according to an embodiment of the present invention; Figure 7 This is a flowchart of the construction of a multi-parameter joint diagnostic model for adenomyosis according to an embodiment of the present invention.
[0020] Among them, 10: image acquisition module; 20: target area positioning module; 21: mark recognition unit; 22: region generation unit; 30: parameter calculation module; 31: hardness calculation unit; 32: resistance calculation unit; 40: model inference module; 50: result display module. Detailed Implementation
[0021] The technical solutions in 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] See attached document Figure 1 , Figure 1 This is a schematic flowchart of a method for diagnosing adenomyosis based on cervical multiparameter ultrasound imaging according to an embodiment of the present invention. The present invention provides a method for diagnosing adenomyosis based on cervical multiparameter ultrasound imaging, comprising the following steps: The S100 uses a transvaginal ultrasound probe to scan the subject, acquiring data on the anatomical region of the cervix, excluding the uterine body. Multimodal ultrasound image data includes two-dimensional grayscale ultrasound images, shear wave elastography images, and ultra-microvascular flow imaging images. The data acquisition process requires obtaining a clear midsagittal section of the cervix showing the internal os, external os, and the entire cervical canal. Specifically, the two-dimensional grayscale ultrasound image is used for anatomical localization, the shear wave elastography image provides information on tissue stiffness distribution, and the ultra-microvascular flow imaging image provides information on microvascular perfusion distribution.
[0023] S200 defines the anatomical extent for quantitative analysis. The cervical measurement target area is located within the cervical parenchyma, excluding the lower segment of the uterine body and surrounding vaginal tissue. The localization strategy uses the longitudinal axis of the cervical canal as the central axis and the horizontal line of the internal cervical os as the boundary, dividing the measurement target area into four independent quantitative analysis sub-regions: the anterior lip region of the internal cervical os, the posterior lip region of the internal cervical os, the anterior lip region of the mid-cervical segment, and the posterior lip region of the mid-cervical segment. The sampling frame position for each region is fixed and its shape is set to circular to ensure consistency of measurement data among different subjects.
[0024] The S300, based on the cervical measurement target area, extracts biomechanical and hemodynamic parameters separately; the system quantitatively calculates the raw signals within the selected target area. Biomechanical parameters are Young's modulus values calculated from shear wave elastography images, in kPa. This parameter quantifies the cervical tissue's resistance to deformation. Specific extracted indicators include the average and maximum Young's modulus values for four regions. Hemodynamic parameters are resistance indices calculated from pulsed Doppler spectra guided by ultra-micro blood flow imaging. Sampling volumes are placed in the region with the densest microvascular distribution shown in the ultra-micro blood flow image to acquire the blood flow spectrum, and the ratio of the difference between the peak systolic velocity and the end-diastolic velocity to the peak systolic velocity is calculated. This parameter reflects the vascular bed resistance status of the cervical microcirculation.
[0025] S400: The extracted biomechanical and hemodynamic parameters are input into a pre-constructed adenomyosis diagnostic model to calculate the probability value of the disease risk. The adenomyosis diagnostic model is a pre-trained multifactor logistic regression model. The input variables of this model are the core parameter combinations optimized through a feature selection strategy. In this embodiment, the input variables specifically include: the average Young's modulus of the anterior lip of the internal cervical os, the resistance index of the anterior lip of the internal cervical os, the average Young's modulus of the posterior lip of the internal cervical os, the average Young's modulus of the anterior lip of the mid-cervical segment, and the average Young's modulus of the posterior lip of the mid-cervical segment. The model performs a linear weighted sum based on the weight coefficients of each input variable, and maps the weighted sum to a probability value between 0 and 1 through a logistic function.
[0026] In step S500, the system compares the calculated probability value with a preset diagnostic threshold. When the probability value is higher than the diagnostic threshold, it outputs a diagnosis indicating adenomyosis; when the probability value is lower than the diagnostic threshold, it outputs a diagnosis indicating no disease. Simultaneously, the system generates a visual nomogram to intuitively display the contribution of each parameter to the total risk.
[0027] See attached document Figure 2 , Figure 2 This is a schematic diagram of the architecture of a diagnostic system for adenomyosis based on multi-parameter cervical ultrasound imaging according to an embodiment of the present invention. Figure 2 As shown, the system includes: The image acquisition module 10 connects to the ultrasound probe hardware, controls the sound beam emission and echo reception, and performs beamforming and signal processing. The image acquisition module 10 includes various imaging algorithm units for generating and outputting grayscale images containing anatomical structures, color-coded elastic images containing tissue stiffness information, and ultramicro-blood flow images containing micro-blood flow signals. The module also includes a storage unit for temporarily storing the acquired raw radio frequency data and the beamformed image data.
[0028] The target area localization module 20, communicatively connected to the image acquisition module 10, is used to receive ultrasound image data and determine the analysis area. The target area localization module 20 includes a landmark recognition unit 21 and a region generation unit 22. The landmark recognition unit 21 is used to identify the hyperechoic linear structures of the internal cervical os and cervical canal in the sagittal section image of the cervix. The region generation unit 22 is used to generate a circular measurement sampling frame within the myometrium of the anterior and posterior lip of the cervix based on the identified anatomical landmarks. This unit uses a boundary constraint algorithm to ensure that the generated sampling frame does not exceed the cervical serosal layer and does not enter the cervical canal mucosa layer.
[0029] The parameter calculation module 30, connected to the target area positioning module 20, is used for quantitative analysis of image data within the sampling frame. The parameter calculation module 30 includes a hardness calculation unit 31 and a drag calculation unit 32. The hardness calculation unit 31 reads the shear wave velocity values of pixels within the sampling frame, converts them into Young's modulus values using Hooke's law, and calculates the average Young's modulus value within the calculation area. The drag calculation unit 32 processes the Doppler spectrum signal, traces the spectral envelope, identifies the systolic peak point and the end-diastolic point, and calculates the drag index.
[0030] The model inference module 40, connected to the parameter calculation module 30, is used to perform disease prediction calculations. The model inference module 40 stores a subset of features selected based on machine learning algorithms and their corresponding model coefficients. This module receives hardness data output from the hardness calculation unit 31 and resistance data output from the resistance calculation unit 32, performs a weighted calculation, and outputs the predicted probability of adenomyosis. The calculation process of this module is independent of the subject's age and obstetric history information, relying only on the measured physical parameters.
[0031] The results display module 50, connected to the model inference module 40 and the image acquisition module 10, is used to present diagnostic information to the user. The results display module 50 displays real-time multimodal images of the cervix, a table of quantitative parameter measurements, and a bar chart or nodal graph of disease probability generated by the model inference module 40 on the screen simultaneously. The results display module 50 also provides an interactive interface, allowing the operator to confirm or fine-tune the sampling frame position generated by the target area localization module 20.
[0032] See attached document Figure 3 , Figure 3 This is a schematic flowchart illustrating the acquisition of cervical multimodal ultrasound image data according to an embodiment of the present invention. To achieve objective and quantitative diagnosis of adenomyosis, acquiring high-quality, standardized ultrasound image data is fundamental for subsequent feature extraction and model calculation. For the specific anatomical structure of the cervix, this invention establishes a strict data acquisition protocol covering the entire process from setting equipment parameters to acquiring specific cross-sections. The specific image acquisition implementation includes the following steps: S101, configures the basic parameters of the ultrasound imaging system; the system uses an intracavitary probe with volumetric imaging capabilities, and the frequency range is set between 2 MHz and 10 MHz. The subject must empty their bladder before the examination to avoid bladder distension altering the position or angle of the cervix. During two-dimensional grayscale imaging, depth gain compensation and overall gain are adjusted to suppress background noise while ensuring clear echoes of tissue structures. For subsequent quantitative analysis, the system must disable image post-processing filtering algorithms or spatial composite imaging functions to prevent distortion of the original acoustic signal and ensure the physical accuracy of subsequent elastic modulus calculations.
[0033] S102, Obtain a standardized midsagittal section image of the cervix; the operator inserts the probe into the vaginal fornix, adjusting the probe angle and depth until the full-thickness structure of the cervix is clearly displayed on the screen. A qualified standard section must simultaneously meet the following anatomical landmark display requirements: the internal cervical os appears as a hyperechoic boundary connecting the uterine cavity line and the cervical canal line; the external cervical os appears as a morphological landmark at the junction of the cervical canal and the vagina; the entire cervical canal appears as a hyperechoic linear structure running through the center of the cervix. During the procedure, the probe needs to be adjusted so that the linear structure of the cervical canal remains as horizontal or nearly horizontal as possible in the image to reduce the influence of anisotropic artifacts caused by the ultrasound incident angle on the measurement results. This grayscale image is a two-dimensional grayscale ultrasound image, used for subsequent anatomical localization and target area delineation.
[0034] S103, Acquire shear wave elastography data; while maintaining standard section stability, switch to shear wave elastography mode. The system emits focused acoustic radiation force pulses through the probe, inducing transversely propagating shear waves within the cervical tissue. At this time, the system adjusts the pulse repetition frequency to adapt to the stiffness range of the cervical tissue, with the upper limit of the elastic modulus range set to 150 kPa. During operation, the probe surface should lightly touch the outer wall of the cervix, avoiding any external pressure to prevent artificial stiffness increase due to mechanical pressure. The operator should keep the probe relatively still for 3 to 5 seconds, waiting for the shear wave to fully propagate within the tissue. Once the elastography quality control chart displayed by the system indicates stable signal and the fill rate reaches the preset standard, freeze the image and save it. The data acquired in this step contains the distribution information of shear wave propagation velocity at various points within the cervical tissue, i.e., the shear wave elastography image.
[0035] S104: Acquire ultra-micro blood flow imaging data. After completing elastic data acquisition, the system switches back to two-dimensional grayscale mode and reconfirms the section position before activating ultra-micro blood flow imaging mode. This mode employs a high frame rate and high-sensitivity Doppler signal processing algorithm, setting the wall filter parameters to a low-pass mode adapted to low-speed blood flow to filter out clutter signals generated by surrounding tissue movement while preserving microvascular signals. Color gain is adjusted until background noise disappears and microvascular signals are abundant. To avoid motion artifacts, the subject must hold their breath for several seconds. After the blood flow signal in the quantitative analysis sub-region stabilizes, the image is frozen and saved. The data acquired in this step reflects the perfusion density and distribution characteristics of the cervical microcirculation, i.e., the ultra-micro blood flow imaging image.
[0036] S105 acquires pulsed Doppler spectral data; guided by ultra-micro blood flow imaging, it focuses on the anterior lip region of the internal cervical os, identifying the area with the richest blood flow signal as the sampling point. Pulsed Doppler mode is activated, and the sampling volume is set to 1 mm to 2 mm, positioned in the center of the target vessel. The angle between the sound beam and the blood flow direction is adjusted to be less than 60 degrees, compensated using the angle correction function. The system records at least three consecutive, stable cardiac cycle spectral waveforms. This spectral data contains information on the temporal changes in systolic and diastolic flow velocities, used for subsequent calculation of hemodynamic parameters.
[0037] For the ultrasound probe operation techniques, beam modulation principles, and basic physical mechanisms of the Doppler effect involved in the above steps, those skilled in the art can refer to relevant medical ultrasound physics textbooks or equipment operation manuals, as these are well-known technologies in the field and will not be elaborated upon here. By performing the standardized procedures S101 to S105 above, a complete set of cervical multimodal ultrasound datasets suitable for subsequent quantitative analysis can be obtained.
[0038] See attached document Figure 4 , Figure 4 This is a schematic diagram illustrating the anatomical location of the cervical measurement target area according to an embodiment of the present invention. To overcome measurement errors caused by the irregular shape and unclear boundaries of the uterus in traditional adenomyosis diagnosis, the present invention employs a standardized target area localization method based on anatomical landmarks. This method locks the quantitative analysis sub-region onto the relatively fixed anatomical structure of the cervical stroma, and achieves objectification and standardization of the measurement location by establishing a geometric coordinate system. The specific localization and definition implementation includes the following steps: S201. Establish an anatomical coordinate system based on the longitudinal axis of the cervical canal. On the acquired midsagittal section image of the cervix, identify the hyperechoic linear structure of the cervical canal as the reference axis. Define the internal cervical os as the junction between the cervical canal and the uterine cavity; this point appears on ultrasound images as the morphological transition between the cervical canal mucosa and the endometrial line. Define the external cervical os as the opening of the distal end of the cervical canal into the vagina. The line connecting the internal and external cervical os is established as the longitudinal axis of the cervix. With the internal cervical os as the origin, the longitudinal axis of the cervix as the Y-axis, and the line perpendicular to the longitudinal axis of the cervix as the X-axis, establish a two-dimensional Cartesian coordinate system. This coordinate system aims to eliminate spatial positional deviations caused by different probe scanning angles, ensuring that all measurement points are distributed relative to the inherent anatomical structure of the cervix.
[0039] S202 defines four standardized quantitative analysis sub-regions. Based on the established anatomical coordinate system, the measurement target area is divided into four spatially independent regions. The first is the anterior lip of the internal cervical os: located ventrally along the longitudinal axis of the cervical canal and at the same horizontal level as the internal cervical os. This region corresponds to the mixed area of smooth muscle and connective tissue at the level of the internal cervical os. The second is the posterior lip of the internal cervical os: located dorsally along the longitudinal axis of the cervical canal and at the same horizontal level as the internal cervical os. This region is adjacent to the rectouterine pouch, but the cervical tissue and rectal wall must be clearly distinguished on ultrasound images. The third is the anterior lip of the mid-segment of the cervix: located ventrally along the longitudinal axis of the cervical canal and at a predetermined midpoint in the vertical direction. The fourth is the posterior lip of the mid-segment of the cervix: located dorsally along the longitudinal axis of the cervix and at a predetermined midpoint in the vertical direction. These four regions cover the main functional parts of the cervix and reflect the overall biomechanical characteristics and blood perfusion of the cervix.
[0040] S203, Perform boundary constraints and placement of the sampling frame for the quantitative analysis sub-region; place circular measurement sampling frames in the four regions defined above. To ensure the accuracy and specificity of the measurement data, the placement of the sampling frames must follow the following geometric constraints: Rule 1: The diameter of the sampling frame is set to a preset diameter range to accommodate the thickness of the cervical myometrium and avoid volume effects. Rule 2: The edge of the sampling frame is at a preset first safe distance from the hyperechoic line of the cervical canal mucosa. This rule is used to eliminate interference from cervical mucus and mucosal glands on the elastic modulus measurement, ensuring that the measurement object is only the cervical stromal layer. Rule 3: The edge of the sampling frame is at a preset first safe distance from the boundary of the cervical serosal layer or vaginal wall. This rule is used to eliminate boundary sclerosis artifacts caused by surrounding connective tissue and probe pressure. Rule 4: The sampling frame must not contain calcifications, cervical glandular cysts, or other focal lesions. If such focal lesions exist at the preset location, the sampling frame position is slightly adjusted horizontally while maintaining the same horizontal plane to avoid the lesion area.
[0041] Through steps S201 to S203, the system can output four standardized quantitative measurement sites. This localization strategy shifts the diagnostic focus of adenomyosis from the highly heterogeneous uterine body to the morphologically stable cervix, utilizing histological changes in the internal os and mid-segment of the cervix as diagnostic criteria, thus achieving repeatability in the diagnostic process. For image segmentation techniques or edge detection algorithms in the anatomical landmark recognition process, those skilled in the art can employ existing image processing techniques, which will not be elaborated upon here.
[0042] See attached document Figure 5 , Figure 5 This is a schematic diagram of a logic block for quantitative parameter extraction according to an embodiment of the present invention. After determining the standardized anatomical quantitative analysis sub-region, the system needs to convert the ultrasound echo signal into objective physical quantitative indicators to construct the input feature vector for subsequent diagnostic models. In this embodiment, biomechanical parameters reflecting tissue stiffness and hemodynamic parameters reflecting microcirculatory resistance are extracted through pixel-level calculations and spectral envelope analysis, respectively. The specific parameter extraction implementation method includes the following steps: S301, calculate the pixel-level Young's modulus value within the quantitative analysis sub-region; the system reads the original shear wave velocity data matrix obtained from shear wave elastic imaging in the memory. For each pixel within the quantitative analysis sub-region, the system performs a conversion based on the isotropic medium assumption in elasticity, utilizing the physical relationship between shear wave propagation velocity and medium hardness. The specific conversion calculation is performed according to the following formula: ; In the formula, This represents Young's modulus, expressed in kilopascals (kPa). The density of cervical tissue is represented and is set to a constant value of 1000 kg per cubic meter in this embodiment. This represents the propagation speed of the shear wave within the tissue, measured in meters per second. Using this calculation, the system generates a Young's modulus distribution map corresponding to the spatial distribution of the quantitative analysis sub-region.
[0043] S302, regional statistical features are extracted as biomechanical parameters. To quantify the overall stiffness characteristics of the cervical stromal layer and reduce noise interference, the system performs statistical aggregation on the data in the distribution map. For the four cervical measurement target areas defined above, the arithmetic mean of the Young's modulus values of all pixels in each area is calculated as a representative stiffness index for that area. The calculation formula is as follows: ; In the formula, This represents the average value of Young's modulus; This indicates the total number of valid pixels contained within the quantitative analysis sub-region; Indicates the first in this region The system calculates the Young's modulus value for each pixel. Additionally, it extracts the maximum Young's modulus value within the region as an auxiliary reference. Through this step, the system outputs the average Young's modulus values for the anterior lip, posterior lip, anterior mid-segment lip, and posterior mid-segment lip regions of the cervix.
[0044] S303, the system analyzes the Doppler spectral envelope to obtain flow velocity indicators; it performs time-domain analysis on pulsed Doppler spectral data acquired under ultra-micro blood flow imaging guidance. An edge detection algorithm is used to automatically track the highest flow velocity envelope of the spectrum and identify key phase points within the cardiac cycle. The system locks the highest frequency point during systole and converts it into peak systolic flow velocity; simultaneously, it locks the lowest frequency point during diastole and converts it into end-diastolic flow velocity. The specific calculation of flow velocity relies on the Doppler frequency shift formula and angle correction coefficient; the physical principles of these formulas are known to those skilled in the art and will not be elaborated here.
[0045] S304, the resistance index is calculated as a hemodynamic parameter; based on the extracted flow velocity index, the system calculates a dimensionless parameter reflecting the compliance and resistance status of the cervical microvascular bed. The formula for calculating the resistance index is as follows: ; In the formula, Indicates the resistance index; This indicates the peak flow velocity during the contraction phase, expressed in centimeters per second. This represents the end-diastolic flow velocity, measured in centimeters per second. The system calculates the resistance index for three or more recorded cardiac cycles and takes the arithmetic mean, outputting the final resistance index of the anterior lip of the internal cervical os. This parameter reflects the compressive effect of tissue fibrosis caused by adenomyosis on microvessels.
[0046] S305, Construct a multidimensional quantitative feature dataset; the system integrates the physical parameters calculated in the above steps. The final output feature dataset includes the following core quantitative indicators: the average Young's modulus of the anterior lip of the internal cervical os, the average Young's modulus of the posterior lip of the internal cervical os, the average Young's modulus of the anterior lip of the mid-cervical segment, the average Young's modulus of the posterior lip of the mid-cervical segment, and the resistance index of the anterior lip of the internal cervical os. This dataset will serve as the input vector for subsequent intelligent diagnostic models, used for calculating the probability of disease risk.
[0047] See attached document Figure 6 , Figure 6This is a flowchart illustrating a three-order cross-feature screening strategy according to an embodiment of the present invention. To screen core features with independent diagnostic value for adenomyosis from extracted high-dimensional biomechanical and hemodynamic parameters, this embodiment of the invention employs an integrated three-order cross-feature screening strategy. This strategy combines algorithms based on three different mathematical principles: regularized regression, recursive feature elimination, and ensemble learning, to eliminate the bias of a single algorithm and ensure the reliability of the final selected features. The specific feature engineering implementation includes the following steps: S401, standardization preprocessing is performed on the original quantitative parameters; since the extracted Young's modulus and drag index differ in magnitude, directly inputting them into the algorithm would affect the allocation of model weights. The system uses the Z-score standardization method to map the feature data of all samples to the standard normal distribution space. The standardization calculation is performed according to the following formula: ; In the formula, Represents the standardized feature values; This represents the original parameter measurement value; This represents the arithmetic mean of the parameter in the training set; This represents the standard deviation of this parameter in the training set. After processing, the mean of all input features is 0, and the standard deviation is 1, thus eliminating the influence of dimensional differences on the calculation of feature weights.
[0048] S402, performs first-order feature dimensionality reduction based on the minimum absolute shrinkage and selection operator; the system inputs the standardized full set of features into the minimum absolute shrinkage and selection operator regression model. This model introduces an L1 regularization penalty term into the loss function, making the regression coefficients of non-critical features zero. During implementation, ten-fold cross-validation is used to construct error curves, and the regularization parameter corresponding to the minimum mean squared error is selected. The system retains features whose regression coefficients are not zero at this point, forming the first candidate feature subset. This step utilizes the sparsity characteristic of L1 regularization, which can effectively remove multicollinear features and initially screen parameters linearly correlated with the disease state.
[0049] S403 executes the second-order feature ranking based on recursive feature elimination using a support vector machine (SVM). The system inputs the entire set of features into an SVM model based on a linear kernel function. This algorithm evaluates feature importance based on the square of the weight vector. Through iterative training, the system removes the feature with the smallest absolute value of the weight coefficient in each training round and retrains the model. During this process, the system uses five-fold cross-validation to calculate the model classification accuracy under different numbers of features, and selects the feature combination corresponding to the highest accuracy point to form the second candidate feature subset. This step focuses on examining the contribution of feature combinations to the construction of the classification hyperplane.
[0050] S404 performs a third-order feature importance assessment based on random forests; the system constructs a random forest classifier containing multiple decision trees. During model training, the algorithm calculates the average Gini index reduction caused by each feature when a node splits. This metric reflects the purity contribution of a feature in distinguishing between diseased and non-diseased samples. The system ranks all features according to the average Gini index reduction and selects the top 50% of features by importance score to form the third candidate feature subset. This step leverages the ability of ensemble learning to handle nonlinear relationships and captures complex interactions between features.
[0051] S405, based on a cross-intersection strategy, determines the final core diagnostic features. To obtain features with high diagnostic value, the system performs an intersection operation on the first, second, and third candidate feature subsets. A three-set intersection strategy is used to determine the final core diagnostic features; only when a parameter is included in all three candidate feature subsets is it confirmed as a core feature with robust diagnostic value. After this three-stage cross-screening, the system ultimately identifies five core parameters: the average Young's modulus of the anterior lip of the internal cervical os, the average Young's modulus of the posterior lip of the internal cervical os, the average Young's modulus of the anterior lip of the mid-cervical segment, the average Young's modulus of the posterior lip of the mid-cervical segment, and the resistance index of the anterior lip of the internal cervical os. This screening strategy mathematically ensures that the selected features possess linear correlation, classification contribution, and non-linear importance, effectively solving the overfitting problem commonly found in small-sample, high-dimensional data.
[0052] See attached document Figure 7 , Figure 7 This is a flowchart illustrating the construction of a multi-parameter joint diagnostic model for adenomyosis according to an embodiment of the present invention. After screening out core feature parameters with independent diagnostic value, this embodiment of the invention constructs a probabilistic prediction model based on multivariate logistic regression. This model establishes a quantitative relationship between cervical biomechanical and hemodynamic characteristics and the risk of adenomyosis, thereby outputting classification results. The specific model construction and output implementation includes the following steps: S501 is the input vector for constructing a multivariate logistic regression model. The system uses the five core parameters selected in the previous stage as independent variables to input into the prediction model. These five independent variables are: the average Young's modulus of the anterior cervical os, the average Young's modulus of the posterior cervical os, the average Young's modulus of the anterior cervical os mid-segment, the average Young's modulus of the posterior cervical os mid-segment, and the resistance index of the anterior cervical os. To ensure the stability of the calculation, the input data needs to be standardized. The system calls the mean and standard deviation of the training set features calculated in the feature engineering stage to perform Z-score standardization on the feature values of the new samples, ensuring that the input values follow a standard normal distribution.
[0053] S502, Model parameters are determined based on maximum likelihood estimation. During the model training phase, using a dataset of labeled clinical samples representing disease states, the intercept term and regression coefficients of each feature in the regression equation are iteratively solved using maximum likelihood estimation. The goal is to find a set of optimal parameters that maximizes the joint probability of the model observing the current sample data. For the specific mathematical derivation of maximum likelihood estimation, those skilled in the art can refer to relevant statistics textbooks; it is a well-known technique in the field and will not be elaborated upon here. The system stores the regression coefficients after training in a database for subsequent prediction of new samples.
[0054] S503, calculate the linear predictive value and the probability of disease; for a new sample to be diagnosed, the system calls the stored regression coefficients and first calculates the linear predictive value. The calculation is performed according to the following formula: ; In the formula, Represents linear predicted values; Indicates the regression intercept; to Represents the regression coefficients for the corresponding features; This represents the average Young's modulus of the anterior lip of the internal cervical os; This represents the average Young's modulus of the posterior lip of the internal cervical os; This represents the average Young's modulus of the anterior lip of the mid-cervix; This represents the average Young's modulus of the posterior lip of the mid-cervix; This indicates the resistance index of the anterior lip of the internal cervical os.
[0055] Subsequently, the system uses the logistic function to map the linear predicted values to the interval between 0 and 1, obtaining the probability of disease. The calculation is performed according to the following formula: ; In the formula, Indicates the predicted probability of developing adenomyosis; The base of the natural logarithm; This represents the linear prediction value.
[0056] S504: Determine the optimal diagnostic threshold and output the classification result. To convert continuous probability values into a binary diagnostic conclusion, the system calculates the Youden index based on the receiver operating characteristic (ROC) curve of the training set. This index is the sum of sensitivity and specificity minus a baseline constant. The probability value corresponding to the maximum Youden index is selected as the optimal diagnostic cutoff value. If the calculated predicted probability is greater than or equal to the optimal diagnostic cutoff value, the system outputs a classification conclusion of "suspected adenomyosis"; if the predicted probability is less than the optimal diagnostic cutoff value, the system outputs a classification conclusion of "no obvious signs of adenomyosis found." This step realizes the transformation from quantitative data to qualitative clinical decision-making.
[0057] Example subject: Patient Ms. Li, 42 years old Chief complaint: Progressive worsening of dysmenorrhea for 3 years, with increased menstrual flow. Traditional examination: Routine two-dimensional ultrasound showed an enlarged spherical uterus with heterogeneous myometrial echoes, but it was impossible to clearly distinguish whether it was due to multiple fibroids or diffuse adenomyosis, resulting in diagnostic uncertainty.
[0058] Diagnostic process using the method of this invention: Step S100: Data Acquisition The doctor uses an ultrasound device equipped with a high-frequency intracavitary probe to obtain a mid-sagittal section of Ms. Li's cervix.
[0059] Grayscale image: Clearly shows the linear structure of the internal and external cervical os and the cervical canal.
[0060] Elastography: The probe gently touches the cervix to collect shear wave elastic data.
[0061] Blood flow imaging: In ultra-micro blood flow mode, the distribution of microvessels in the cervical myometrium is displayed.
[0062] Step S200: The target area positioning system automatically identifies the longitudinal axis of the cervical canal. Using the horizontal line of the internal cervical os as the boundary, it automatically generates circular sampling frames with a diameter of 3mm in the following four areas: Anterior lip region of the internal cervical os Posterior lip region of the internal cervical os Mid-segment anterior lip region of the cervix Note: The system detected a Nabothian cyst near the mucosa of the anterior lip of the cervix and automatically shifted the sampling frame 2mm outward to avoid the fluid area of the cyst.
[0063] Step S300: The parameter extraction and calculation system performs quantitative analysis on the data within the sampling frame and extracts the following core parameter values: X1 (average Young's modulus of the anterior lip of the internal cervical os): 48.5 kPa (normal reference value is about 25-30 kPa, indicating obvious hardening) X2 (average Young's modulus of the posterior lip of the internal cervical os): 45.2 kPa X3 (average Young's modulus of the anterior lip of the mid-cervix): 38.1 kPa X4 (average Young's modulus of the posterior lip of the mid-cervix): 36.8 kPa X5 (Cervical Internal Ovary Anterior Lip Resistance Index RI): 0.78 (Normal reference value is about 0.60-0.65, indicating increased microcirculation resistance) Step S400: Model Inference The system inputs the normalized data above into the pre-trained logistic regression model. Assume the model's regression equation is: ; The linear prediction value was calculated. The probability is calculated using the logistic function: (94.2%). Step S500: Output Results Probability value: 94.2% Diagnostic threshold: set at 65% Systematic conclusion: Suspected adenomyosis Visualization: The screen displays a nomogram, in which the bars for Young's modulus and resistance index of the anterior lip of the internal cervical os are the longest, suggesting that these two factors contribute the most to this diagnosis.
[0064] Clinical verification: Ms. Li subsequently underwent an MRI examination and subsequent surgical treatment. Postoperative pathology confirmed diffuse adenomyosis. The diagnostic results of this invention are consistent with the pathological results.
[0065] Part Two: Experimental Verification and Effect Comparison To verify the effectiveness of the method of the present invention, we conducted a comparative study with the traditional subjective scoring method based on two-dimensional ultrasound of the uterine body.
[0066] 1. Experimental setup Sample size: A total of 160 samples were collected.
[0067] Case group: 80 patients with pathologically confirmed adenomyosis.
[0068] Control group: 80 healthy or non-adenomyotic subjects who were excluded from adenomyosis.
[0069] Comparison method: Method A: Logistic regression model based on 5 core parameters of the cervix.
[0070] Method B: Two senior physicians performed routine two-dimensional ultrasound diagnosis based on characteristics such as uterine morphology and asymmetric thickening.
[0071] 2. Through parameter analysis of the case group and the control group, it was found that the features screened by this invention have significant distinguishing power: Young's modulus: The average Young's modulus of the internal cervical os region in the case group was significantly higher than that in the control group (P<0.001), indicating that adenomyosis caused occult fibrosis of the cervical tissue.
[0072] Resistance index: The cervical microvascular resistance index was significantly elevated in the case group, which is related to pelvic congestion and increased tissue pressure.
[0073] 3. Comparison of Diagnostic Efficacy Results Based on ROC curve (Receiving Receiver Operating Characteristic) analysis, the results are as follows: Conclusion: As shown in the table above, compared with traditional two-dimensional ultrasound diagnosis that relies on the morphological characteristics of the uterine body, the method of this invention has achieved a qualitative leap in all key diagnostic indicators: The diagnostic accuracy of this invention is significantly improved from 70.6% of the traditional method to 86.9%, an increase of 16.3 percentage points. This indicates that by introducing objective physical quantitative parameters, the risk of misdiagnosis caused by traditional ultrasound relying on the doctor's subjective experience to judge the shape of the uterus is effectively avoided.
[0074] In terms of sensitivity, this invention achieves 88.7%, far exceeding the 72.5% of traditional methods. This means that in the same clinical sample, this approach can identify more than 16% more patients with occult or early-stage adenomyosis, which is of decisive significance in solving the clinical diagnostic problem of "severe dysmenorrhea but atypical imaging manifestations".
[0075] As the gold standard for evaluating the effectiveness of diagnostic models, the area under the curve (AUC) jumped from 0.742 to 0.915. An AUC exceeding 0.9 indicates that this invention is no longer an auxiliary screening method, but possesses high reliability as a basis for diagnosis.
[0076] In summary, the data strongly demonstrate that by quantifying the biomechanical stiffness and microcirculatory blood flow resistance of cervical tissue, this invention successfully captures the specific pathological changes induced by adenomyosis in the cervical region, overcoming the bottleneck of traditional methods limited by the variability of uterine body pathologies, and providing a new objective, quantitative, and highly accurate diagnostic paradigm for clinical practice.
[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging, characterized in that, The method includes the following steps: Multimodal ultrasound image data, including two-dimensional grayscale ultrasound images, shear wave elastography images, and ultramicro blood flow imaging images, were acquired using a transvaginal ultrasound probe. Based on the longitudinal axis of the cervical canal in the two-dimensional grayscale ultrasound image, an anatomical coordinate system is established, and multiple independent quantitative analysis sub-regions covering the internal cervical os and the middle segment of the cervix are identified as measurement target areas within the cervical parenchyma. Based on the quantitative analysis sub-region, biomechanical parameters reflecting tissue stiffness and hemodynamic parameters reflecting microcirculation resistance are extracted using the shear wave elastography image and the ultramicro blood flow imaging image, respectively. The biomechanical parameters and hemodynamic parameters are input as a multivariate logistic regression model for the diagnostic model of adenomyosis. The diagnostic model for adenomyosis performs linear weighted summation based on weight coefficients and maps the linear predicted values to disease risk probability values through a logistic function. When the disease risk probability value is higher than a preset diagnostic threshold, it outputs a classification conclusion indicating that the patient has adenomyosis.
2. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 1, characterized in that, The anatomical coordinate system is established based on the longitudinal axis of the cervical canal in the two-dimensional grayscale ultrasound image, and multiple independent quantitative analysis sub-regions are determined within the cervical parenchyma, including: Using the hyperechoic linear structure of the cervical canal as the reference axis, with the internal cervical os as the origin, the cervical longitudinal axis as the Y-axis, and the straight line perpendicular to the cervical longitudinal axis as the X-axis, a two-dimensional plane rectangular coordinate system is established. Based on the two-dimensional Cartesian coordinate system, the measurement target area is divided into four spatially independent regions: the anterior lip region of the internal cervical os, located on the ventral side of the longitudinal axis of the cervical canal and at the same horizontal level as the internal cervical os; the posterior lip region of the internal cervical os, located on the dorsal side of the longitudinal axis of the cervical canal and at the same horizontal level as the internal cervical os; the anterior lip region of the middle segment of the cervix, located on the ventral side of the longitudinal axis of the cervical canal and at a predetermined midpoint of the longitudinal axis length; and the posterior lip region of the middle segment of the cervix, located on the dorsal side of the longitudinal axis of the cervical canal and at a predetermined midpoint of the longitudinal axis length.
3. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 2, characterized in that, When placing a circular measurement sampling frame within the quantitative analysis sub-region, the following boundary constraints apply: Set the sampling frame diameter to a preset diameter range; constrain the edge of the sampling frame to a preset first safe distance from the hyperechoic line of the cervical canal mucosa to eliminate interference from cervical mucus and mucosal glands; The sampling frame edge is constrained to a preset first safe distance from the cervical serosal layer or vaginal wall boundary to eliminate boundary hardening artifacts caused by surrounding connective tissue and probe compression; If calcifications or cervical glandular cysts are present at the preset location, the sampling frame position is adjusted by horizontally shifting while maintaining the same horizontal plane.
4. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 1, characterized in that, The extraction of biomechanical parameters reflecting tissue stiffness using the shear wave elastography image includes: Read the shear wave propagation velocity of each pixel in the quantitative analysis sub-region; calculate the Young's modulus value based on the isotropic medium assumption in elasticity, whereby the Young's modulus value is equal to the product of the elastic coefficient, the cervical tissue density, and the square of the shear wave propagation velocity; The arithmetic mean of the Young's modulus values of all valid pixels in the quantitative analysis sub-region is calculated to obtain the average Young's modulus value as the biomechanical parameter.
5. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 1, characterized in that, The extraction of hemodynamic parameters reflecting microcirculatory resistance using the ultramicrovascular blood flow imaging image includes: Under the guidance of ultra-micro blood flow imaging, pulsed Doppler spectral data were acquired in the anterior lip region of the internal cervical os; the highest flow velocity envelope of the spectral data was tracked using an edge detection algorithm to identify the peak systolic flow velocity and end-diastolic flow velocity during the cardiac cycle. Calculate the drag index, which is equal to the difference between the peak systolic velocity and the end-diastolic velocity divided by the peak systolic velocity. The arithmetic mean of the resistance indices over at least three consecutive cardiac cycles is used to obtain the anterior lip resistance index of the internal cervical os as the hemodynamic parameter.
6. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 1, characterized in that, Before constructing the adenomyosis diagnostic model, a three-order cross-feature selection strategy is performed to determine the input variables. This three-order cross-feature selection strategy includes: First-order feature dimensionality reduction: Input the standardized full set of features into the minimum absolute shrinkage and selection operator regression model, select the regularization parameter corresponding to the minimum mean square error through ten-fold cross-validation, and retain the features with non-zero regression coefficients to form the first candidate feature subset; Second-order feature ranking: Input all features into a support vector machine model based on a linear kernel function, evaluate feature importance based on the square of the weight vector, and select the feature combination corresponding to the point with the highest classification accuracy by combining five-fold cross-validation to form the second candidate feature subset; Third-order feature importance assessment: Construct a random forest classifier, calculate the average Gini index reduction caused by each feature when splitting nodes, and select the top 50% of features by importance score to form the third candidate feature subset; Feature intersection: Perform an intersection operation on the first candidate feature subset, the second candidate feature subset, and the third candidate feature subset to determine the final core parameter combination.
7. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 6, characterized in that, The core parameter combination specifically includes five parameters: the average Young's modulus of the anterior lip of the internal cervical os, the average Young's modulus of the posterior lip of the internal cervical os, the average Young's modulus of the anterior lip of the mid-cervical segment, the average Young's modulus of the posterior lip of the mid-cervical segment, and the resistance index of the anterior lip of the internal cervical os.
8. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 1, characterized in that, The step of inputting the extracted biomechanical parameters and hemodynamic parameters as input variables into the adenomyosis diagnostic model includes: The mean and standard deviation of the training set features calculated during the model training phase are called; using the mean and standard deviation of the training set features, the extracted parameters of the current subject are Z-score standardized to make the input data follow a standard normal distribution. The stored regression intercept and regression coefficients for each input variable are retrieved to calculate the linear prediction value.
9. The diagnostic method for adenomyosis based on multi-parameter cervical ultrasound imaging according to claim 1, characterized in that, The method for determining the preset diagnostic threshold is as follows: the Youden index is calculated based on the receiver operating characteristic curve of the training set. The Youden index is the sum of sensitivity and specificity minus the baseline constant. The probability value corresponding to the maximum Youden index is selected as the optimal diagnostic cutoff value.
10. A diagnostic system for adenomyosis based on multi-parameter cervical ultrasound imaging, characterized in that, The method for diagnosing adenomyosis based on cervical multiparameter ultrasound imaging according to any one of claims 1-9 includes the following system: The image acquisition module is used to connect to the ultrasound probe and acquire two-dimensional grayscale ultrasound images, shear wave elastography images, and ultra-micro blood flow imaging images of the cervical anatomical region. The target area localization module is used to identify the internal os and cervical canal structure based on the two-dimensional grayscale ultrasound image, establish an anatomical coordinate system with the longitudinal axis of the cervical canal as the reference, and generate a standardized measurement sampling frame in the cervical parenchyma. The parameter calculation module is used to read the image data within the measurement sampling frame, calculate the average value of Young's modulus based on the shear wave propagation velocity, and calculate the drag index based on the Doppler spectrum envelope. The model inference module stores pre-trained multivariate logistic regression model parameters. It is used to receive the standardized Young's modulus mean and resistance index, and calculate the disease risk probability value through linear weighting and logistic function mapping. The results display module is used to compare the disease risk probability value with the optimal diagnostic cutoff value and output the classification conclusion of adenomyosis.