An ultrasound underbied uterine conduction band evaluation system

Abstract

The application discloses an ultrasonic uterine junction zone evaluation system, and the subjective process is converted into objective mathematical calculation through image processing and computer vision algorithm. Specifically, after the image quality is optimized by using anisotropic diffusion filtering and CLAHE algorithm, a hybrid segmentation technology based on edge detection and three-dimensional region growth is adopted to accurately extract a three-dimensional geometric model of JZ. On this basis, the integrity of JZ is quantified by defining and calculating a core parameter, namely, continuity index (CI). The algorithm samples along the center line system of JZ, analyzes the mutation of the local width of adjacent points, and accurately represents the continuity degree of the ring structure by using a mathematical proportion (CI value).

Description

Technical Field

[0001] This invention relates to an ultrasound-guided evaluation system for the uterine junction zone, specifically a system and its implementation method that automatically evaluates the morphological characteristics of the endometrial-myometrial interface / junctional zone (JZ) using three-dimensional transvaginal ultrasound (3D-TVUS) images and a quantitative algorithm. Background Technology

[0002] The junctional zone (JZ) is a specialized functional region within the myometrium adjacent to the endometrium, playing a crucial role in regulating uterine peristaltic waves, maintaining normal endometrial growth and differentiation, and controlling embryo implantation and placental formation. Clinical studies have confirmed that abnormal JZ morphology is significantly associated with adenomyosis, unexplained infertility, and poor embryo implantation outcomes in assisted reproductive technologies. Compared to simple JZ thickness measurements, its morphological integrity is considered a more sensitive predictor of assisted reproductive outcomes.

[0003] Currently, clinical assessment of the uterine lesion (JZ) primarily relies on two imaging techniques: First, magnetic resonance imaging (MRI), especially T2-weighted imaging, which clearly shows the JZ as a low-signal band surrounding the endometrium and is considered the "gold standard" for assessment. However, MRI examinations are expensive, time-consuming, have limited accessibility, and are not convenient for repeated use in continuous monitoring cycles of assisted reproduction. Second, three-dimensional transvaginal ultrasound (3D-TVUS), which has the outstanding advantages of being non-invasive, convenient, real-time, and relatively low-cost. Its coronal imaging can provide a panoramic view of the JZ morphology. The international consensus on ultrasound assessment of uterine morphology proposes descriptive classification standards for JZ morphology.

[0004] However, existing JZ assessment models based on 3D-TVUS have significant drawbacks, including a high reliance on the experience and subjective judgment of ultrasound physicians, a lack of objective and unified quantitative standards, leading to large intra- and inter-observer variability and low diagnostic reproducibility; and the inability to provide qualitative or semi-quantitative descriptions of key pathological signs such as "irregularity" and "interruption," failing to accurately quantify the degree of irregularity or the size of the interruption gap, resulting in ambiguous information that hinders accurate diagnosis and efficacy assessment; furthermore, the assessment results are mostly independent qualitative descriptions, which cannot be combined and correlated with massive amounts of clinical parameters and pregnancy outcome data, limiting the construction of predictive models.

[0005] There is a need in this field for a technical solution that can rely on widely available 3D-TVUS equipment to achieve objective, quantitative, automated, and intelligent assessment of JZ morphology, in order to solve the problems of subjectivity and ambiguity in manual interpretation, and provide a more reliable and predictive imaging tool for clinical practice in gynecology and reproductive medicine. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention proposes a quantitative assessment system and method for the morphology of the uterine junction based on three-dimensional ultrasound images, aiming to transform the morphological assessment of the junction from subjective qualitative description to objective quantitative analysis.

[0007] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution: According to one aspect of the present invention, an ultrasound-guided system for evaluating the uterine junction is provided, comprising: The image acquisition module is used to acquire a sequence of three-dimensional vaginal ultrasound images of the target uterine region; The image preprocessing module is used to perform noise reduction, contrast enhancement, and boundary sharpening on the image. Combined with a region segmentation module, it is used to automatically identify and segment the endometrial myometrial junction region; The feature parameter calculation module is used to calculate the continuity feature parameters and thickness uniformity feature parameters of the bonding zone region; An evaluation output module is used to generate a quantitative evaluation report based on the feature parameters.

[0008] Preferably, the image preprocessing module uses an anisotropic diffusion filtering algorithm for noise reduction and a limited contrast adaptive histogram equalization (CLAHE) algorithm for contrast enhancement.

[0009] Preferably, the combined region segmentation module uses a combination of Canny edge detection and a region growing algorithm that combines grayscale and texture features for 3D segmentation.

[0010] Preferably, the continuity characteristic parameter includes a continuity index (CI), which is calculated by analyzing local width abrupt changes at sampling points along the JZ centerline and is used to quantify the integrity of the JZ low-echo band. The closer its value is to 1, the better the continuity.

[0011] Preferably, the thickness uniformity characteristic parameter includes the coefficient of variation of thickness (CV). JZ ) and maximum thickness difference (JZ dif This is used to quantify the degree of fluctuation in the spatial distribution of JZ thickness. CV JZand JZ dif The smaller the value, the better the uniformity.

[0012] Preferably, the feature parameter calculation module further calculates the maximum thickness of JZ (JZ). max ), JZ average thickness (JZ mean ) and the ratio of JZ thickness to total uterine myometrial thickness (R JZ / TWT Comprehensive morphological parameters, including those for morphological parameters.

[0013] Preferably, the evaluation output module has a built-in evaluation model that takes the multiple quantification parameters as input feature vectors and outputs morphological classification results and / or comprehensive morphological abnormality scores (S) through a pre-trained machine learning classifier or a decision tree based on clinical threshold rules. The output morphology classification results include: normal, regular, irregular, interrupted, and blurred.

[0014] Preferably, the system further includes a clinical parameter database and a prompt generation unit, used to match relevant clinical characteristics, pregnancy outcome data and intervention recommendations based on the assessment results, and integrate them into the assessment report.

[0015] According to another aspect of the present invention, a method for evaluating the uterine junction zone under ultrasound is provided, comprising the following steps: S1: Acquire a three-dimensional vaginal ultrasound image sequence; S2: Preprocess the image; S3: Automatically segments the three-dimensional region of the bonding zone; S4: Calculate the quantitative parameters for the continuity and thickness uniformity of the bonding zone region; S5: Automated classification, scoring, and evaluation report generation based on quantitative parameters.

[0016] Preferably, the method further includes step S6: comparing the assessment results with the JZ morphology-function-pregnancy outcome association model, and outputting individualized fertility assessment, disease risk or treatment strategy recommendations.

[0017] The beneficial effects of this invention are: An ultrasound-guided method for evaluating the uterine junctional zone is proposed. By constructing a technical system from image preprocessing and feature quantification to automatic evaluation, this method fundamentally solves some of the problems caused by the reliance on subjective qualitative evaluation in existing technologies, and achieves objective, accurate, repeatable and quantitative analysis of the morphology of the uterine junctional zone. Current clinical practice relies entirely on the visual interpretation of the hypoechoic zone (JZ) by sonographers, using qualitative terms such as "regular," "irregular," or "interrupted." The judgment criteria vary significantly from person to person, exhibiting substantial intra- and inter-observer variability. This invention transforms this subjective process into objective mathematical calculations through image processing and computer vision algorithms. Specifically, after optimizing image quality using anisotropic diffusion filtering and the CLAHE algorithm, a hybrid segmentation technique based on edge detection and 3D region growing is employed to accurately extract the 3D geometric model of the JZ. Based on this, the integrity of the JZ is quantified by defining and calculating the continuity index (CI): the algorithm systematically samples along the JZ centerline, analyzing abrupt changes in local width at adjacent points, and accurately characterizing the continuity of its annular structure using a mathematical proportion (CI value). Simultaneously, the thickness variation coefficient (CV) is calculated... JZ ) and maximum thickness difference (JZ dif This transforms vague descriptions such as "non-uniformity" or "local thickening" into measurable parameters based on statistics and range. These parameters have clear mathematical definitions and calculation procedures, completely eliminating the instability of human visual judgment. This ensures that the assessment results of the same patient at different times, by different operators, and in different medical institutions are highly comparable and reproducible, providing a stable and reliable quantitative tool for clinical diagnosis and scientific research. Traditional focus has often been limited to the maximum or average thickness of the uterine plexus (JZ). However, clinical studies have shown that, compared to absolute thickness, the "continuity" and "uniformity" of JZ morphology are more closely related to uterine peristaltic function, endometrial receptivity, and embryo implantation outcomes. The innovation of this invention lies in avoiding single-dimensional measurement and constructing a quantitative evaluation system centered on "continuity" and "uniformity of thickness spatial distribution." The continuity index (CI) directly maps the integrity of the JZ as a ring-shaped functional structure; its discontinuity may correspond to peristaltic wave conduction disorders or local micro-trauma lesions. The coefficient of variation (CV) of thickness... JZ This statistically characterizes the dispersion of JZ thickness across the entire uterine wall. An increased JZ value suggests significant focal thickening or thinning, reflecting the heterogeneity of glandular and stromal infiltration within the myometrium. This structural heterogeneity may be a major factor affecting local blood perfusion and the embryo implantation microenvironment. Simultaneous acquisition of CI and CV... JZ JZ dif and JZ max JZ mean JZ thickness / total muscle wall thickness ratio (R) JZ / TWT With a set of multiple parameters, the system can perform panoramic, multi-scale digital twin analysis of JZ morphology, providing refined data assistance for the pathophysiological mechanisms of diseases such as adenomyosis.

[0018] The core of this invention lies in its built-in evaluation model, which facilitates the derivation from data to assessment. The system integrates calculated multi-dimensional feature parameters into a feature vector, which is then input into a pre-trained evaluation model. This model learns from a large amount of historical case data, including the correlation between JZ quantification features and known IVF pregnancy outcomes or clinical diagnoses, to uncover the complex mapping relationship between key features and clinical outcomes, ultimately outputting a comprehensive morphological abnormality score (S). This process combines scattered imaging features into a clinically relevant evaluation structure. The score S not only provides an image description but also includes decision-making aids for prognostic prediction.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below. 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.

[0021] Example 1 This embodiment provides a quantitative assessment system for the morphology of the uterine junctional zone based on three-dimensional ultrasound images. This system can be integrated into an ultrasound workstation, a standalone image processing server, or a cloud analysis platform, and mainly includes the following modules: Image acquisition module: Responsible for receiving or importing 3D-TVUS volumetric data of the target uterine region from the 3D vaginal ultrasound diagnostic instrument. This data is usually stored in DICOM (Digital Imaging and Communications in Medicine) format and contains two-dimensional sequences obtained from scans at different angles. After reconstruction, three-dimensional volumetric data including sagittal, coronal, and transverse views can be obtained.

[0022] Image preprocessing module: Connected to the image acquisition module, this module preprocesses the raw 3D volume data to improve image quality and provide clearer JZ boundary information for subsequent segmentation. This module performs at least two key steps: 1. Anisotropic diffusion filtering noise reduction: The Perona-Malik model is used to smooth image noise while preserving important edge information. Its discretization iterative formula is as follows: in:- Indicates the first In the next iteration, voxels The grayscale value. The diffusion coefficient ( This controls the diffusion rate. Voxel representation The six-neighborhood or twenty-six-neighborhood. , indicating from To its neighboring voxels The gradient. The edge stopping function is defined as follows: .in The gradient magnitude threshold is used when the gradient... Greater than At this time, diffusion is suppressed, thus protecting the edge.

[0023] 2. Contrast-Limited Adaptive Histogram Equalization (CLAHE): To enhance the contrast between the JZ region and its adjacent endometrium and superficial myometrium, the CLAHE algorithm is employed. This algorithm divides the image into several small regions (tiles), performs histogram equalization independently on each region, and avoids noise amplification by limiting the contrast stretching amplitude. Finally, bilinear interpolation is used to eliminate block boundary artifacts.

[0024] Combined with the region segmentation module: connected to the preprocessing module, its core task is to accurately separate the JZ region from the preprocessed 3D image. This embodiment adopts a hybrid segmentation strategy combining edge detection and region growing, and the specific steps are as follows: 1. Edge Initialization: First, operations are performed on the coronal plane of the preprocessed 3D volume data, as this plane best displays the overall JZ surrounding the inner membrane. The Canny edge detection operator is applied to each coronal slice sequence. A double thresholding method (high threshold) is used. and low threshold After processing and edge connectivity, potential edge pixels at the endometrial-myolipin junction are initially located, and this set of points is used as the initial seed set for JZ segmentation. .

[0025] 2. Three-dimensional region growth: with Starting from each point in the graph, a 3D region growing process is performed. The growing process considers not only grayscale similarity but also local texture features to improve the robustness of segmentation for JZ regions with non-uniform echoes. For the currently grown region... Its regional characteristics are determined by the average gray level of all its voxels. and average Local Binary Pattern (LBP) texture feature vector Characterization. For a neighborhood voxel to be examined. Its growth criterion function Defined as: in: It is a voxel The grayscale value.

[0026] It is a voxel The LBP texture feature vector is used to describe its local micro-patterns.

[0027] and These are weighting coefficients used to balance the contributions of grayscale and texture features, typically set through grid search or empirical methods, for example... This represents the Euclidean distance between vectors.

[0028] if (Growth threshold), then voxels Included in the region This process is repeated iteratively until no new voxels meet the growth conditions, ultimately yielding a complete JZ 3D binary mask. .

[0029] Feature parameter calculation module: Connected to the segmentation module, it calculates the segmented 3D JZ model. Quantitative analysis was conducted to calculate two main categories of core morphological feature parameters.

[0030] 1. First, extract the center line of the JZ 3D model. The center path can be obtained using skeletonization algorithms or distance-transform-based center path extraction methods.

[0031] Along the center line Equal arc length or equal spacing sampling Let there be 1 point, denoted as 1. .

[0032] At each sampling point At that point, calculate the plane along the point and the JZ model. Local width of intersecting sections This can be obtained by measuring the distance between the two boundaries of the JZ region at that point within the normal plane.

[0033] Width mutation analysis: Traverse all adjacent sampling point pairs Calculate the absolute value of its local width difference. Define a relative mutation threshold. ,in It is the average local width. It is a coefficient (e.g., 0.4). If Then it is believed that in and There are suspected morphological interruptions or irregularities, record. Otherwise, remember .

[0034] The continuity index (CI) is defined as the percentage of all adjacent point pairs that do not exhibit abrupt changes. The closer the value is to 1, the better the integrity of the JZ band on the loop path. 2. Thickness uniformity characteristic parameters: Let JZ 3D model The set of thickness values ​​extracted is ,in This represents the total number of sampling points.

[0035] Coefficient of Variation (CVA) The thickness variation coefficient is used to characterize the dispersion of the JZ thickness distribution. It is defined as the ratio of the thickness standard deviation to the thickness mean, and is calculated as follows: in: The arithmetic mean of the thickness of JZ; The sample standard deviation of the JZ thickness; It is a dimensionless parameter, and the smaller its value, the more uniform the spatial distribution of JZ thickness.

[0036] Maximum Thickness Difference The maximum thickness difference directly reflects The absolute fluctuation range between thickness extrema is defined as: This parameter is intuitively characterized The thickness span between the thickest and thinnest regions.

[0037] Comprehensive morphological parameters provide a complete description In addition to morphological characteristics, the system also calculates the following parameters: Thickness extremes and mean Maximum thickness ( ): Minimum thickness ( ): Average thickness ( ): The ratio of JZ thickness to the total thickness of the uterine myometrium ( To eliminate the influence of individual differences in uterine size, at specific anatomical locations (e.g., fundus, midpoint of the anterior wall, midpoint of the posterior wall) Calculate the ratio of the local JZ thickness to the corresponding total thickness of the uterine myometrium: in: For position JZ thickness at the location; For position Total thickness of the uterine myometrium at that location; The standardized thickness ratio parameter is suitable for comparative analysis across cases or in longitudinal follow-up.

[0038] Evaluation Output Module: This module receives one or more sets of feature parameters from the feature parameter calculation module, forming a feature vector. ,For example Evaluation Model: The evaluation output module has one or more built-in evaluation models used to transform quantified feature vectors into clinically meaningful outputs.

[0039] 1. Rule-based classifier: Set explicit threshold rules: like and If so, it is classified as "Class I: Rules".

[0040] like or If so, it is classified as "Class II: Irregular".

[0041] like If so, it is classified as "Class III: Interruption".

[0042] If incomplete segmentation or unreliable calculation of key parameters is caused by image quality issues, it is classified as "Category IV: Blurred / Unevaluable". 2. Machine Learning Model: The model is trained using a historical dataset containing a large number of patients' JZ quantified features V and their corresponding gold standard diagnostic labels or IVF pregnancy outcome labels "pregnancy / non-pregnancy". A logistic regression model is trained. This model learns a weight vector. and bias terms Used to calculate a comprehensive morphological anomaly score in, For input feature vectors Standardization functions, including but not limited to Z-score standardization, are used to ensure that all features are on the same scale. Scoring The probability of morphological abnormalities can be obtained after transformation using the Sigmoid function. A higher score indicates a worse JZ morphology, and a lower predicted pregnancy success rate. Weight vector The numerical value in the value reflects the contribution of the corresponding feature parameter to the adverse outcome.

[0043] Clinical parameter database: Connected to the assessment model, it is a structured knowledge base that stores clinical information corresponding to different JZ morphological classifications or scoring intervals; Typical ultrasound images and descriptions.

[0044] Clinical pregnancy rate and implantation rate statistics based on literature reviews or data from our hospital.

[0045] Expert consensus or guidelines recommend the following clinical treatment pathways: "For patients with a JZ morphology score S>0.7, it is recommended to undergo at least 2 cycles of GnRH agonist pretreatment followed by reassessment."

[0046] Report Generation Unit: Integrates the results output by assessment model 151 and relevant information retrieved from database 152 to automatically generate a structured graphic assessment report. The report content may include: JZ 3D reconstruction rendering, key parameter numerical tables, morphological classification results, comprehensive abnormality score and risk level, and personalized "clinical tips and suggestions" text.

[0047] Example 2 This embodiment details the complete process of a quantitative assessment method for the morphology of the uterine junction based on three-dimensional ultrasound images. This method is implemented using the system described in Embodiment 1, and through a series of automated and quantitative steps, transforms the raw 3D-TVUS data into an assessment report with clear clinical significance. The specific steps are as follows: S610: Acquisition and Import of 3D Ultrasound Image Data This step is performed by the system's image acquisition module. The operator uses an ultrasound device with a transvaginal probe equipped with three-dimensional volume to perform a standardized three-dimensional scan of the patient's uterus. During the scan, ensure the uterus is completely within the sampling frame and that the patient's bladder is moderately distended to provide a good acoustic window. The scan yields raw ultrasound volume data, typically stored in DICOM 3.0 format. The system supports direct DICOM transfer from the ultrasound workstation or importing stored DICOM files. After data import, the system automatically parses and reconstructs a three-dimensional volume dataset containing sagittal, coronal, and transverse views. ,in For voxel coordinates, the grayscale value represents the tissue echo intensity.

[0048] Image preprocessing and quality enhancement This step is performed by the image preprocessing module. Its purpose is to optimize the image signal-to-noise ratio and enhance the boundary contrast of the target structure (JZ), laying the foundation for subsequent accurate segmentation. Preprocessing includes two operations: 1. Anisotropic diffusion filtering: for Iterative filtering is performed using the Perona-Malik model. The diffusion coefficient is set. Gradient magnitude threshold The edge stopping function is adaptively determined based on the overall noise level of the image, typically the 70th percentile of the image's gray-level gradient histogram. This ensures that diffusion is suppressed in the high-gradient region at the boundary between the JZ and the inner membrane / muscle layer, thereby preserving or even sharpening the edge contour of the JZ while homogenizing noise within the smooth muscle layer. After the iteration, the denoised data is obtained. .

[0049] 2. Contrast-limited adaptive histogram equalization (CLAHE): This will... The data was divided into local blocks of 32×32×8 voxels. For each block, its grayscale histogram was calculated, and the contrast stretching was limited to a factor of 2.0 to prevent excessive enhancement of local areas and amplification of noise. Finally, bilinear interpolation was used to eliminate boundary effects between blocks. This operation significantly improved the grayscale difference between the low-echo JZ and the relatively high-echo endothelium and the moderately echogenic superficial muscle layer, generating contrast-enhanced volumetric data. This step makes the previously blurry JZ boundary clearer and more discernible in the image.

[0050] S630: Fully Automatic 3D Segmentation of the JZ Region This step is performed by the combined region segmentation module, and its goal is to... The complete 3D structure of JZ was accurately extracted. A hybrid segmentation strategy was adopted to ensure robustness: A sequence of coronal views. For each coronal image, the Canny edge detection operator is applied. The Gaussian smoothing kernel size is set to [value missing]. Pixels, dual thresholds are the high threshold and low threshold High threshold This is the foreground scale threshold for the image gradient magnitude, with a default value of 30%. This operation initially delineates the boundary between the endometrium and myometrium, forming an initial set of edge points. .

[0051] Each point in the diagram is a seed, in three-dimensional space. Perform region growth within the range. Growth criterion function. Taking into account both grayscale similarity and texture consistency. Specifically, the current growth region The characteristic is determined by its average gray level and average local binary pattern (LBP, using a circular neighborhood with a radius of 2 pixels) texture histogram Characterization. For the 26-neighborhood voxels to be investigated Calculate its grayscale difference and its LBP histogram and The chi-square distance. Weights are set. growth threshold It is adaptively determined through a function of the regional gray-level variance. When At that time, voxels Included in the region This process is repeated iteratively until no new voxels satisfy the condition. Finally, a 3D binary mask marking the JZ region is output. A voxel with a value of 1 represents a member of JZ.

[0052] S640: Precise Calculation of Multi-Dimensional Morphological Feature Parameters This step is performed by the feature parameter calculation module, which processes the segmented 3D model. Quantitative analysis was conducted to extract two types of core features and a set of comprehensive parameters.

[0053] 1. Calculation of continuous characteristics: First, calculate the continuous characteristics... Morphological refinement was performed to extract its three-dimensional centerline. N points were obtained by sampling at 0.5 mm intervals along the centerline. At each point On the normal plane, calculate The maximum inscribed circle diameter on this cross section is used as the local width. Calculate the absolute difference in width between all adjacent pairs of points. Set a mutation threshold. ( (For average width). Statistically satisfied. The number of adjacent point pairs, according to the formula Calculate the continuity index. Simultaneously, record significant abrupt changes. The location and the range of width variation are used for the positioning description in the report.

[0054] 2. Calculation of thickness uniformity characteristics: In On the entire surface, by calculating the shortest Euclidean distance from each surface voxel to its opposite boundary, a set of thickness values ​​covering the entire JZ is obtained. There are M samples in total. Calculate the mean of this set. Standard deviation This leads to the thickness variation coefficient. Simultaneously, the maximum thickness difference is calculated. 3. Comprehensive morphological parameter calculation: from the set The maximum thickness of JZ is obtained directly from the middle. minimum thickness and average thickness In addition, on a standardized coronal view, the JZ thickness was measured at three pre-defined anatomical locations: the fundus, the midpoint of the anterior wall, and the midpoint of the posterior wall. and the total thickness of the uterine myometrium from the endometrial cavity to the serosal layer And calculate the ratio. .

[0055] S650: Evaluation and Structured Report Generation Based on Intelligent Models This step is performed by the assessment output module, which transforms quantitative parameters into clinical decision support information.

[0056] 1. Feature Integration and Model Evaluation: The CI and CV calculated in step S640 are integrated. JZ JZ dif JZ max And three positions A total of 7 core parameters are used to construct a feature vector. The system inputs this vector into a pre-trained logistic regression evaluation model. This model has been trained based on data from over 500 patients with adenomyosis and a proven IVF-ET outcome, and its weight vector... and bias terms It's already fixed. The model first... Perform Z-score calibration; x+14 yields Then calculate the linear combination. The S-value is mapped to the morphological abnormality probability through the Sigmoid function and directly converted into a comprehensive morphological abnormality score (S) between 0 and 10.

[0057] 2. Report Synthesis and Output: Based on the score S, the system categorizes the risk level as "low risk (S<3)," "medium risk (3≤S<7)," or "high risk (S≥7)." Simultaneously, the system activates the prompt generation unit, retrieving from the clinical parameter database typical imaging feature descriptions corresponding to that risk level, the reported pregnancy rate range from the literature, and expert consensus-recommended clinical management suggestions. Finally, the report generation unit automatically integrates the following content: Image section: 3D surface reconstruction rendering of JZ, with different colors used to highlight the interrupted areas detected in CI analysis.

[0058] Data section: All calculated quantization parameters and their reference ranges are clearly listed in tabular form.

[0059] Conclusion section: Clearly defined morphological risk level (low / medium / high) and comprehensive score S.

[0060] Recommendations section: Personalized clinical tips and suggestions for follow-up steps.

[0061] S660: Clinical decision support and longitudinal comparison. This advanced step can be performed for cases requiring in-depth analysis or follow-up. A second scan assessment is performed after the patient completes one cycle of GnRH-a treatment. The system processes the new scan data through steps S610-S650 to obtain new feature vectors. and rating The system can automatically retrieve the patient's baseline data before treatment. Calculate the improvement rate of key parameters (e.g.) , ) and score reduction value By comparing the treatment response model with those in the database, the system can generate analytical opinions, providing quantitative evidence for clinical adjustments to treatment plans.

[0062] Beneficial effects and experimental verification The system and method of this invention, by applying image processing, computer vision, and machine learning techniques to 3D-TVUS image analysis, achieves objective, accurate, and automated quantitative assessment of JZ morphology. Its advantages are: It eliminates the subjective differences in human assessment and improves the consistency and reproducibility of diagnoses.

[0063] Provided quantization parameters (CI, CV) JZ (etc.) provide new, precisely measurable imaging biomarkers for clinical research.

[0064] The generated composite score is directly correlated with clinical outcomes, enhancing the predictive value of the assessment results.

[0065] Automated report generation and clinical prompts improve clinical efficiency and assist physicians in decision-making.

[0066] In a retrospective validation, the system of this invention was used to analyze preimplantation 3D-TVUS images of a cohort of adenomyosis patients with known IVF-ET outcomes. The system-calculated composite score S was used as a predictor, and logistic regression analysis was performed between the score and actual pregnancy outcomes. The results showed that score S was an independent and significant predictor of clinical pregnancy, with an odds ratio (OR) of 4.2 (95% confidence interval 2.1–8.3, P < 0.001). The area under the receiver operating characteristic (AUC) of S for predictive power was 0.82, which was higher than the AUC of 0.68 for the method based solely on subjective morphological classification by the sonographer.

[0067] Example 3 It also provides an objective, quantitative, multi-dimensional, and highly predictive algorithm and system for assessing the endometrial binding band, including: The image acquisition module is used to acquire multimodal ultrasound images of the uterus; The image processing module is used to automatically segment and identify the endometrial junction region; The feature extraction module is used to calculate at least the following quantitative features of the binding zone region: thickness uniformity index (JZ-UI), thinnest point thickness (JZ-MinT), and binding zone-inner membrane volume ratio (JZ / EM-Vol Ratio). The prediction module, based on a machine learning model, combines the quantitative features with clinical data to output the probability of pregnancy outcomes and clinical decision recommendations.

[0068] The formula for calculating the thickness uniformity index (JZ-UI) is as follows: JZ-UI=[1-(Thickness_SD / Thickness_Mean)]×100% Thickness SD To combine the standard deviation of the tape thickness at multiple standard measurement points, Thickness Mean This represents the average thickness.

[0069] The prediction module adopts a dual-model parallel structure, including: A classification model to output the probability of clinical pregnancy; A regression model is used to output an embryo implantation potential score; A decision optimization layer, which integrates the above outputs, embryo quality, and patient age, generates a transplantation recommendation.

[0070] The endometrial junction assessment algorithm and system also include: 1. Data Acquisition and Preprocessing Module Input data source: 1.1 Image Data: Transvaginal 3D ultrasound grayscale images: acquired on the day of HCG injection, the day of embryo transfer, and the 3rd day after transfer.

[0071] Color / energy Doppler images: used to assess blood perfusion in the binding zone region.

[0072] Ultrasonic elastography data: quantifying the stiffness of the combined tissue band.

[0073] 1.2 Synchronous Integration of Clinical Data: Patient age, BMI, cause of infertility, hormone levels (E2, P); Embryo quality scoring, such as the Gardner score; Previous history of intrauterine procedures; Preprocessing steps: Image standardization, resolution uniformity, and grayscale calibration; Noise filtering and enhancement processing; Automatic identification and image alignment of the uterine cavity midline; 2. Quantitative Characterization System of Endometrial Binding Zone This invention defines and quantifies for the first time the following three categories of 12 core feature parameters: A. Morphological and structural features: 2.1.1 Three-dimensional thickness distribution map: Measure the thickness (in mm) of the commissural band at the fundus, anterior wall, posterior wall, and lateral wall of the uterus, and calculate: Average thickness (JZ-AvgT); Thickness uniformity index (JZ-UI) = (1 - standard deviation of thickness / average thickness) × 100%; The thickness of the thinnest point (JZ-MinT) and its location coordinates; 2.1.2 Three-dimensional volume and surface area: Combined with volume (JZ-Vol); Volume to intima volume ratio (JZ / EM-Vol Ratio). 2.1.3 Regularity of the boundary area: Inner edge sharpness scoring: Gradient magnitude quantization based on edge detection operator (0-1 points); Outer edge serration index: calculated by boundary curvature fluctuation; B. Functional and dynamic characteristics: 2.2.1 Blood perfusion parameters: Combined with regional vascularization index (JZ-VI): the percentage of blood vessels per unit volume. Consistency of blood flow direction: Statistical distribution of Doppler signal direction 2.2.2. Peristaltic conduction characteristics, specifically for dynamic ultrasound sequences: The attenuation rate of the creeping wave as it crosses JZ; JZ contraction frequency (times / minute) and phase difference with endometrial contraction; C. Texture and echo features, calculated based on gray-level co-occurrence matrix, etc. 2.3.1 Echo uniformity; 2.3.2 Texture complexity; 3. Construction of core algorithm model A multi-stage hybrid prediction model is adopted: Pregnancy outcome prediction model = feature selection layer + core prediction layer + decision optimization layer; Phase 1: Feature Selection Layer We used LASSO regression or random forest feature importance ranking to select the feature subset that was most strongly correlated with pregnancy outcomes (p<0.01) from the initial feature pool. Pregnancy outcomes included clinical pregnancy and live birth.

[0074] The expected core features include: JZ-UI (homogeneity index), JZ-MinT (thinnest point thickness), JZ / EM-VolRatio, and JZ-VI (vascularization index).

[0075] Phase Two: Core Prediction Layer, Dual Model Parallel Operation 3.1.1 Classification model for predicting pregnancy status: Algorithm: XGBoost or LightGBM; Output: Clinical pregnancy probability (P1, range 0-1); 3.1.2 Regression model, predicting success rate: Algorithm: Multilayer Perceptron Neural Network; Output: Embryo implantation potential score (IPS, range 0-100), which comprehensively reflects the expected implantation depth and stability.

[0076] Phase 3: Decision Optimization Layer Inputs: P1, IPS, embryo quality score, patient age; Processing: Multi-objective optimization algorithm to balance the probability of pregnancy with the risk of multiple pregnancies; Output: 3.2.1. Transplantation recommendations: "Strongly recommended", "Transplantation recommended", "Consider carefully", "Transplantation not recommended for this cycle"; 3.2.2 Optimal number of embryos to be transferred; 3.2.3 Individualized intervention suggestions: For example, for cases of thin JZ, the suggestion is "It is recommended to perform intrauterine perfusion or estrogen pretreatment"; 4. System Workflow 4.1 Data Input: Import ultrasound DICOM images and clinical data.

[0077] 4.2 Automatic segmentation: The algorithm automatically identifies and delineates the junction area, which can be fine-tuned by the physician.

[0078] 4.3 Feature Extraction: Calculate all feature parameters from Section 3.2.

[0079] 4.4 Model Calculation: Run the prediction model and generate a report.

[0080] 4.5 Report Output: A report with both text and images, including: Combined with 3D reconstruction images; Comparison of characteristic parameter radar chart with normal range; Pregnancy probability and recommendations; Historical cycle comparison.

[0081] In practice, taking the prediction of live birth outcomes as an example... Example 1: Model Training and Validation Training set: Collected ultrasound images and final live birth outcomes data of 1500 patients who completed a full IVF / ICSI-ET cycle.

[0082] Feature selection results: JZ-UI>85%, JZ-MinT>3.8mm, and JZ / EM-Vol Ratio between 0.25 and 0.45 were determined to be the "ideal range".

[0083] Model performance: On the independent test set (n=300), the AUC for predicting live birth reached 0.89, which was significantly higher than the model that only used age and embryo quality (AUC=0.72).

[0084] Example 2: Clinical Decision Making Application Case: A 35-year-old woman underwent two high-quality blastocyst transfers.

[0085] System Analysis: JZ-UI = 78% (slightly lower than the ideal value); JZ-MinT=3.5mm (the fundus is relatively thin); JZ-VI=1.2% (low blood flow); Model output: Clinical pregnancy probability P1=62%; Embryo implantation potential score (IPS) = 68 / 100; Decision recommendation: "Consider carefully. Considering the homogeneity of the endometrial tissue and poor blood perfusion, we recommend considering the transfer of one embryo to reduce the risk, or performing endometrial blood flow pretreatment before transfer." In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0086] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An ultrasound-guided system for evaluating the uterine junction, characterized in that: include: The image acquisition module is used to acquire a three-dimensional vaginal ultrasound (3D-TVUS) image sequence of the target uterine region, the image sequence including at least sagittal, coronal and transverse views of the uterus; An image preprocessing module, connected to the image acquisition module, is used to perform noise reduction, contrast enhancement, and tissue boundary sharpening on the 3D-TVUS image to highlight the endometrial-myolipin junction area. The combined region segmentation module is connected to the image preprocessing module and configured to automatically identify and segment the endometrial myometrial junction zone (JZ) region from the preprocessed image; The feature parameter calculation module, connected to the fusion zone region segmentation module, is configured to calculate at least two types of core morphological feature parameters from the segmented JZ region: a) Continuity characteristic parameter: used to quantify the integrity of the JZ low echo band; b) Thickness uniformity characteristic parameter: used to quantify the degree of spatial thickness variation of the JZ low echo band; An evaluation output module, connected to the feature parameter calculation module, is used to generate a quantitative evaluation report on the morphology of the target uterine junction based on the calculated continuity feature parameters and thickness uniformity feature parameters. The report includes at least one of morphological classification identifier and abnormal risk score.

2. The ultrasound-guided uterine junction evaluation system as described in claim 1, characterized in that: The image preprocessing module specifically performs at least one of the following: Anisotropic diffusion filtering algorithm is used for noise reduction, and its discretization implementation formula is as follows: in, Let be the image grayscale value at time t. Where is the diffusion coefficient. For pixels The neighborhood, For point Time gradient, The edge stopping function is defined as follows: , The gradient magnitude threshold is used; the contrast-limited adaptive histogram equalization (CLAHE) algorithm is used to enhance the contrast between the JZ region and the adjacent endothelium and muscle layer.

3. The ultrasound-guided uterine junction evaluation system according to claim 1 or 2, characterized in that, The combined region segmentation module employs a segmentation algorithm based on edge detection and region growing. Specifically, it includes: First, on the coronal view of the 3D image, the Canny edge detection operator is used to initially locate the potential boundary between the endometrium and myometrium; then, using the detected edges as initial seed points, a region growing algorithm based on grayscale and texture features is used to extend the region to adjacent slices in 3D space until the JZ region is fully delineated. The growth criterion function of the region growing algorithm... Defined as: in, Candidate pixels grayscale value, This represents the average gray level of the currently grown area. For point Local Binary Pattern (LBP) texture feature vectors, This is the mean vector of texture features for the current region. For weighting coefficients, when At that time, the point Included in the JZ region.

4. The ultrasound-guided uterine junction evaluation system as described in claim 1, characterized in that: The continuity characteristic parameters include the continuity index. The calculation method is as follows: In the segmented JZ 3D model, N points are sampled at equal intervals along the JZ centerline. ; at each sampling point At this point, calculate the local width of JZ perpendicular to the centerline. Define width abrupt change point: if the absolute value of the width difference between two adjacent points is... Greater than the preset mutation threshold If the continuity index is positive, then a suspected interruption is considered to exist at that point; Defined as the proportion of non-mutation points: ,in in, The closer the value is to 1, the better the continuity of JZ.

5. The ultrasound-guided uterine junction evaluation system according to claim 1 or 4, characterized in that, The thickness uniformity characteristic parameter includes the thickness variation coefficient. and maximum thickness difference The calculation method is as follows: In the JZ three-dimensional model, the thickness distribution of the whole or a specific sub-region (such as the anterior wall, posterior wall, and fundus) is measured to obtain a set of thickness values. ; Calculate the average thickness and standard deviation : Then the coefficient of variation of thickness for: At the same time, calculate the maximum thickness difference. : in, The smaller the value, The lower the value, the better the uniformity of JZ thickness.

6. The ultrasound-guided uterine junction evaluation system as described in claim 1, characterized in that: The feature parameter calculation module also calculates a set of comprehensive morphological parameters, including: JZ maximum thickness. JZ minimum thickness JZ average thickness And the ratio of JZ thickness to the total thickness of the uterine myometrium at the corresponding location. The evaluation output module has a built-in evaluation model, which will evaluate the continuity index. Coefficient of variation of thickness Maximum thickness difference , , Multiple parameters in the input feature vector The input feature vector is processed through a pre-trained machine learning classifier or a decision tree based on clinical threshold rules. The mapping is to a morphological classification result, and the classification includes at least: Class I (normal and regular), Class III (irregular), Class III (disruptive), and Class N (fuzzy / unevaluable).

7. The ultrasound-guided uterine junction evaluation system as described in claim 6, characterized in that: The evaluation model uses a logistic regression-based scoring system to calculate a comprehensive morphological anomaly score. in, For the input feature vector, For the characteristic normalization function, For the weight vector, The weight vector is the bias term. The weights for each component in the score were determined using multivariate logistic regression analysis based on retrospective clinical study data, resulting in a score that... It is positively correlated with poor morphology or lower pregnancy rates.

8. The ultrasound-guided uterine junction evaluation system as described in claim 1, characterized in that: The assessment output module is also connected to a clinical parameter database and a prompt generation unit; The clinical parameter database stores typical clinical features corresponding to different JZ morphological classifications, correlation data with pregnancy outcomes, and recommended intervention measures. The prompt generation unit is configured to match the current assessment result with the clinical parameter database, automatically generate text content containing diagnostic prompts, prognostic assessments and clinical recommendations, and integrate it into the quantitative assessment report.

9. A method for evaluating the uterine junction zone under ultrasound, characterized in that, Includes the following steps: S1: Acquire a sequence of three-dimensional vaginal ultrasound (3D-TVUS) images of the target uterine region; S2: Preprocess the image sequence to enhance the imaging clarity of the endometrial-myolipin junction region; S3: Automatically segment the three-dimensional region of the myometrial junction (JZ) from the preprocessed image; S4: Calculate at least two core morphological quantization parameters for the JZ region, including a continuity characteristic parameter for evaluating integrity and a thickness uniformity characteristic parameter for evaluating thickness spatial variability; S5: Based on the calculated quantitative parameters, the morphology of the uterine junction is automatically classified or scored, and a quantitative assessment report is generated.

10. The method for evaluating the uterine junction under ultrasound according to claim 9, characterized in that, Step S5 is followed by step S6: The results in the quantitative assessment report were compared and analyzed with known association models of JZ morphology, function, and pregnancy outcomes established based on large-sample clinical studies. Based on the comparative analysis results, we can provide fertility assessment opinions for individual patients, risk levels of adenomyosis or fibroid-related pathological conditions, or optimization suggestions for embryo transfer strategies in assisted reproductive technologies.

Images