Method and system for automatic estimation of labor progress angles based on confidence guidance and reliability constraints

Abstract

An automatic estimation method and system for the angle of labor progression (AoP) based on confidence-guided and reliability-constrained approaches are proposed. The method includes: acquiring and preprocessing transperineal two-dimensional ultrasound images during labor as the original input images; constructing and training a joint segmentation and geometric modeling network; inputting the original input images into the trained joint segmentation and geometric modeling network to generate segmentation results and a spatial confidence map; performing confidence-guided geometric modeling based on the spatial confidence map to obtain the pubic symphysis axis and the fetal head tangent; calculating the angle of labor progression based on the pubic symphysis axis and the fetal head tangent, and obtaining the confidence score of the angle of labor progression; adaptively performing lightweight testing with geometric reliability constraints during the target domain testing phase; and outputting the final segmentation results of the pubic symphysis, the final segmentation results of the fetal head, the angle of labor progression, and the confidence score of the angle of labor progression. This invention can obtain stable, interpretable, and quantifiable reliable automatic AoP measurement results, and is particularly suitable for cross-device and cross-center labor ultrasound scenarios.

Description

Technical Field

[0001] This invention relates to the field of medical ultrasound image processing and intelligent obstetric auxiliary assessment technology, and in particular to an automatic estimation method and system for the angle of labor progression based on confidence guidance and reliability constraints. Background Technology

[0002] Assessing labor progress is a crucial step in obstetric clinical decision-making, directly impacting the choice of delivery method, the timing of assisted delivery procedures, and the development of relevant intervention plans. Transperineal ultrasound during labor can acquire real-time information on fetal head descent at the bedside, providing relatively objective imaging evidence for labor assessment. Among these, the Angle of Progression (AoP) is an important quantitative indicator widely used in current clinical practice and research. AoP is determined by the geometric relationship between the fetal head contour and the pubic symphysis, typically requiring initial identification of the pubic symphysis and fetal head structure before further geometric measurements.

[0003] Existing automatic angle measurement technologies can be broadly categorized into two types: direct angle regression methods, which directly input ultrasound images into a neural network and output angle values; and post-segmentation geometric measurement methods, which first segment the pubic symphysis and fetal head, and then extract axes, boundaries, or tangents based on the segmentation results to calculate the AoP. While the former method is simpler, it lacks sufficient interpretability and makes it difficult to pinpoint the source of the angle; the latter method has stronger interpretability but is extremely sensitive to the accuracy of structural boundary positioning. Because AoP is a cascaded geometric measurement, even with high overall segmentation accuracy, any local deviation at key boundaries will amplify errors in subsequent ellipse fitting, tangent determination, and angle calculation, resulting in significant angle deviations.

[0004] Furthermore, ultrasound images taken during labor often suffer from speckle noise, low contrast, blurred boundaries, and acoustic shadowing. The pubic symphysis is a narrow and small structure, making it susceptible to noise and occlusion; the fetal head boundary often exhibits localized reflection loss and discontinuous contours. This can lead to false positives, boundary adhesions, localized breaks, or positioning drift in existing segmentation models in the pubic symphysis and fetal head boundary region, thereby affecting the stability of AoP geometric measurements.

[0005] Furthermore, practical clinical deployment faces significant domain shift issues. Differences in image grayscale distribution, texture style, and structural visibility arise from variations between hospitals, ultrasound equipment, operators, and acquisition parameters. Existing models are mostly trained in a single source domain, often exhibiting performance degradation in the target domain, which typically lacks labeled data, making supervised fine-tuning difficult. Therefore, improving cross-domain measurement stability in the absence of target domain annotations remains a challenge in current technologies.

[0006] Due to the aforementioned problems, there is an urgent need for an automatic estimation technology for the process progression angle that can simultaneously take into account structural segmentation accuracy, geometric measurement stability, result interpretability, and cross-domain robustness. Summary of the Invention

[0007] To address the problems in existing technologies, such as unstable key boundary positioning, amplification of local errors after geometric operations, and insufficient robustness under conditions of unlabeled target domains, this invention provides an automatic estimation method and system for the angle of labor progression (AoP) based on confidence guidance and reliability constraints. By constructing a three-branch local structure enhancement architecture on the basis of the existing two-dimensional segmentation backbone network, and combining boundary point confidence weighting, fetal head weighted ellipse fitting, AoP confidence assessment, and a test-time adaptive strategy of geometric reliability constraints, stable, interpretable, and quantifiable reliability-based automatic AoP measurement results are obtained, which are particularly suitable for cross-device and cross-center ultrasound scenarios during labor.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect is an automatic estimation method for the labor progression angle based on confidence guidance and reliability constraints, which includes the following steps: S1. Acquire and preprocess transperineal two-dimensional ultrasound images during labor as the raw input images; S2. Construct and train the joint segmentation and geometric modeling network; input the original input image into the trained joint segmentation and geometric modeling network to generate segmentation results and spatial confidence maps; S3. Based on the spatial confidence map, perform confidence-guided geometric modeling to obtain the pubic symphysis axis and the tangent of the fetal head; S4. Calculate the labor progression angle based on the pubic symphysis axis and the tangent of the fetal head, and obtain the confidence level of the labor progression angle; S5. Adaptive testing of geometric reliability constraints during the target domain testing phase; S6. Output the final segmentation results of the pubic symphysis, the final segmentation results of the fetal head, the labor progression angle, and the confidence level of the labor progression angle.

[0009] Preferably, S2 includes: The joint segmentation and geometric modeling network consists of a two-dimensional segmentation backbone, a three-branch shared coding structure, a spatial gating unit, a decoder, and a confidence prediction branch. The three-branch shared coding structure includes a global context branch for preserving the overall structural information of the original image and two local structural enhancement branches for highlighting key regions through spatial gating. Two spatial gating maps are predicted on shallow high-resolution features and multiplied pixel by pixel with the original input image to generate two complementary local enhanced views. The original input image and two local augmented views are simultaneously input into a shared encoder for multi-scale feature extraction, fused at each scale, and then fed into a decoder to obtain the segmentation result and spatial confidence output.

[0010] As a preferred embodiment, S2 also includes: For the segmentation results Perform softmax operation to obtain the probability map. : in, Indicates the first Line number Column pixels belong to category The probability of; Denotes a set of categories, where, Indicates background, Indicates the pubic symphysis. Indicates the fetal head; Represents an exponential function; By analyzing the segmentation results The discrete segmentation label is obtained by taking the maximum response category pixel by pixel. : in, This indicates that the category with the largest response value is selected in the category dimension.

[0011] Discrete segmentation labels The pubic symphysis region is obtained by extracting the maximum connected components from the regions belonging to the pubic symphysis category and the fetal head category respectively. and fetal head area Extract the pubic symphysis region separately. and fetal head area The set of boundary points is used to obtain the set of boundary points of the pubic symphysis. Set of fetal head boundary points : in, Indicates the first One pubic symphysis boundary point, This represents the total number of boundary points of the pubic symphysis; Indicates the first The boundary point of the fetal head, This represents the total number of boundary points of the fetal head.

[0012] Preferably, S3 includes: Set of fetal head boundary points Each point in In the spatial confidence map Upsampling yields the corresponding confidence value: in, Represents boundary points The weights satisfy ; The tire head boundary is fitted to an ellipse using weighted least squares; the general quadratic curve expression of the ellipse is given by: in, , , , , and These are the parameters to be determined; Represents two-dimensional plane coordinates. When the following conditions are met... When this occurs, the corresponding quadratic curve is an ellipse; The parameters of the fitted ellipse for the tire head boundary are obtained by minimizing the following weighted error function: in, Represents boundary points The residual value after substituting into the ellipse equation; The parameters of a general conic section are converted into the center, major and minor axes, and rotation angle of an ellipse, denoted as: in, To fit the coordinates of the center of the ellipse; The length of the semi-major axis; The length of the semi-minor axis; Let be the rotation angle of the ellipse relative to the horizontal axis.

[0013] As a preferred option, S3 also includes: Set of boundary points of the pubic symphysis By performing least-squares linear fitting, the axis of the pubic symphysis was obtained. Determine the upper and lower endpoints along the pubic symphysis axis, denoted as ; and A tangent line on the outer side of the fetal head fitting ellipse, corresponding to the geometric configuration, intersects the axis of the pubic symphysis. The geometric relationships required to establish the labor progression angle measurement are defined, where the key points related to the tangent and the axis are denoted as follows: .

[0014] Preferably, S4 includes: The angle of labor progression is calculated based on the triangular relationship formed by the axis of the pubic symphysis and the tangent of the fetal head. in, Indicates the angle of labor progress, in degrees; Point With point The distance between them; Point With point The distance between them; Point With point The distance between them; Represents the inverse cosine function; Pi; As a preferred option, S4 also includes: Calculate the confidence level of the angle of labor progression : in, Indicates the confidence level of the AoP measurement; The number of key boundary sampling points of the tire head involved in geometric modeling; Represents boundary points Confidence values ​​on the spatial confidence map.

[0015] As a preferred embodiment, S5 includes: The optimization objective during unsupervised testing is firstly defined as the entropy minimization loss. in, A probability map generated for the segmentation logits of the current test image; This represents pixel-level prediction entropy loss, meaning it encourages the network to output more definitive category predictions. The second part is the spatial smoothing loss: in, Used to constrain the continuity of predictions for adjacent pixels and reduce local fragmentation predictions; The third part is the geometric reliability loss: in, The stability constant is greater than zero, for example, it can be taken as... ; The overall optimization goal during testing is: in, , and This is the loss weighting coefficient.

[0016] As a preferred option, S5 also includes: Perform gradient updates three times for each target domain image. After each update, perform forward inference again and refresh the confidence level of the labor progress angle. And the segmentation probability graph.

[0017] Secondly, an automatic estimation method system for the angle of labor progression based on confidence guidance and reliability constraints includes: The data acquisition and preprocessing module is used to acquire and preprocess transperineal two-dimensional ultrasound images during labor as the raw input images. The structure-enhanced segmentation module is used to construct and train a joint segmentation and geometric modeling network; the original input image is input into the trained joint segmentation and geometric modeling network to generate segmentation results and spatial confidence maps; The geometric modeling module is used for confidence-guided geometric modeling based on spatial confidence maps to obtain the pubic symphysis axis and the tangent of the fetal head; and to calculate the labor progression angle based on the pubic symphysis axis and the tangent of the fetal head. The reliability assessment module is used to calculate the confidence level of the labor progress angle; The test-time adaptive module is used for lightweight test-time adaptation with geometric reliability constraints during the target domain testing phase. The results output module is used to output the final segmentation results of the pubic symphysis, the final segmentation results of the fetal head, the labor progression angle, and the confidence level of the labor progression angle. The method for automatically estimating the angle of labor progression based on confidence guidance and reliability constraints is used to implement the automatic estimation system for the angle of labor progression based on confidence guidance and reliability constraints as described in the first aspect, and its steps.

[0018] Compared with the prior art, the beneficial effects of the present invention are reflected in: (1) Integrated collaborative modeling of structural segmentation and geometric measurement was achieved.

[0019] This invention does not only output the segmentation results of the pubic symphysis and the fetal head, but also introduces boundary point confidence weighting, fetal head weighted ellipse fitting and explicit geometric angle solution on the basis of the segmentation results. This gives the AoP results a clear anatomical source and geometric interpretation, avoiding the problem of insufficient interpretability of existing direct angle regression methods.

[0020] (2) It significantly reduces the amplification effect of local boundary errors on AoP.

[0021] This invention introduces a spatial confidence map to assign different weights to boundary points involved in geometric modeling. The influence of low-confidence boundary points on tire head contour fitting is explicitly weakened, thus effectively suppressing the interference of noise, local pseudo-boundaries, and fuzzy boundaries on AoP calculation and improving the stability of key geometric quantities.

[0022] (3) Improved the ability to identify the critical boundary between the pubic symphysis and the fetal head.

[0023] This invention employs a three-branch shared coding and spatial gating enhancement structure, which strengthens the representation of local key boundaries while preserving the global structural context. It is particularly suitable for slender structures such as the pubic symphysis and structures with easily blurred boundaries and local discontinuities such as the fetal head, thereby improving the boundary accuracy and stability of the segmentation results.

[0024] (4) For the first time, the reliability of geometric measurement is directly incorporated into the adaptive optimization objective during testing.

[0025] This invention proposes an AoP confidence C_{AoP} and its corresponding geometric reliability loss L_{aop}, and combines them with entropy minimization loss and spatial smoothing loss to form the overall optimization objective during testing. This enables the model to not only improve segmentation determinism under unlabeled conditions in the target domain, but also to directly adjust towards "higher geometric availability", thereby enhancing the robustness of AoP measurement in cross-device and cross-center scenarios.

[0026] (5) It can output the measured values ​​and their reliability quantification indicators simultaneously.

[0027] Most existing automatic AoP measurement methods only output a single angle result, while this invention, while outputting the AoP value, also provides the AoP confidence level C_{AoP} directly corresponding to the geometric measurement, providing additional reliability reference information for clinical applications, which is beneficial for doctors to screen and judge the automatic measurement results.

[0028] (6) It has the potential for lightweight online deployment.

[0029] This invention updates only the normalized layer affine parameters and linear layer parameters during the target domain testing phase, while freezing the remaining network parameters. This reduces the computational burden of online updates while maintaining adaptability, making it more suitable for actual deployment in clinical bedside ultrasound scenarios. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall framework of the method of the present invention; Figure 2 This is a schematic diagram showing the qualitative comparison results of the present invention with other advanced algorithms on intrapartum ultrasound images; Figure 3 This is a schematic diagram illustrating the contribution of each component in this invention to the accuracy of labor progress angle estimation; Detailed Implementation

[0031] To make the technical means, inventive features, objectives, and effects of the invention readily understandable, the invention is further described below with reference to specific illustrations. However, the invention is not limited to the embodiments described below.

[0032] It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings of this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0033] This invention provides an automatic estimation method and system for the labor progress angle based on structure-enhanced segmentation, confidence-guided geometric modeling, and adaptive geometric reliability constraints. It is applicable to the automatic identification of the pubic symphysis and fetal head structure in transperineal ultrasound images during labor, the automatic measurement of the labor progress angle, and stable inference under cross-device and cross-center conditions.

[0034] Example 1: Automatic AoP estimation process based on a single transperineal ultrasound image during labor An automatic estimation method for the angle of labor progression based on confidence guidance and reliability constraints includes the following steps: S1. Acquire and preprocess transperineal two-dimensional ultrasound images during labor; This embodiment uses a transperineal two-dimensional ultrasound image during childbirth as an example to illustrate the method of the present invention in detail.

[0035] Let the input grayscale ultrasound image be denoted as . Its spatial dimensions are standardized after preprocessing. .in, Represents coordinates in the image grayscale value at that location Indicates the horizontal pixel coordinates. This represents the vertical pixel coordinates. Image preprocessing includes removing invalid edge regions, grayscale normalization, and resizing. As an implementable example, the original image can first be cropped to a valid region including the pubic symphysis and fetal head, and then scaled using bilinear interpolation. Finally, the grayscale values ​​are normalized to The range. This preprocessing procedure falls under the category of standard image processing workflows in this field.

[0036] S2. Construct and train a joint segmentation and geometric modeling network; input the transperineal two-dimensional ultrasound image of childbirth into the trained joint segmentation and geometric modeling network to generate segmentation results and spatial confidence maps; S21. Construct and train a joint segmentation and geometric modeling network; This embodiment preferably uses a two-dimensional U-Net as the basic segmentation backbone. Two-dimensional U-Net is an existing technology used for basic segmentation of three types of pixels: background, pubic symphysis, and fetal head. Building upon this, this invention further introduces a three-branch shared coding and spatial gating enhancement mechanism to improve the representation capability of key local boundary structures. This step is one of the core innovative steps of this invention.

[0037] The joint segmentation and geometric modeling network consists of a two-dimensional segmentation backbone, a three-branch shared coding structure, a spatial gating unit, a decoder, and a confidence prediction branch.

[0038] The two-dimensional segmentation backbone can use U-Net as the basic architecture, which is an existing technology. The innovation of this invention lies in introducing a three-branch shared coding and gating enhancement mechanism on this basis to improve the representation ability of key local boundary structures.

[0039] The three-branch shared coding structure includes: one global context branch to preserve the overall structural information of the original image; and two local structure enhancement branches to highlight key regions through spatial gating. The two spatially gating maps predicted on shallow high-resolution features are denoted as follows: and and compared with the original input image respectively. Pixel-wise multiplication creates two complementary local interest views; where, and The values ​​are all located in Interval. Two locally enhanced views are obtained through pixel-by-pixel multiplication: in, This represents a pixel-by-pixel multiplication operation; and These represent two locally enhanced input views.

[0040] Then, , and Simultaneously, a shared encoder is used for multi-scale feature extraction, which is then fused at each scale before being fed into the decoder to obtain the final segmentation output and spatial confidence output. The global context branch and the two local structure enhancement branches share the same encoder, extracting multi-scale features and fusing them at each scale before feeding them into the decoder to output the final segmentation result. Both the gating graph prediction and the multi-scale feature fusion process are trained end-to-end using the segmentation loss.

[0041] During the training phase, it is preferable to first perform supervised training on labeled data from the source domain. As a feasible training configuration, the AdamW optimizer can be used, with a batch size of 8, a training epoch count of 30, and an initial learning rate of [missing information]. The weight decay coefficient is set to 0.1. During training, the network outputs segmentation logits and a spatial confidence map, where the segmentation logits are denoted as... ,in , The number of categories is 3, corresponding to the background, pubic symphysis, and fetal head respectively. The spatial confidence plot is denoted as follows. ,in Represents pixels The value indicates the reliability of the prediction result; the higher the value, the higher the reliability.

[0042] S22. Perform forward inference on a single ultrasound image to be tested to obtain segmentation results and spatial confidence maps; The preprocessed single image to be tested Input the trained structure-reinforcement segmentation network to obtain the segmentation results. Spatial confidence plot . The larger the value, the higher the prediction confidence at the corresponding position. To obtain pixel-level class probabilities from logits, the segmentation results are... Perform softmax operation to obtain the probability map. : in, Indicates the first Line number Column pixels belong to category The probability of; Represents a set of categories, where Indicates background, Indicates the pubic symphysis. Indicates the fetal head; This represents an exponential function. The segmentation results are analyzed... The discrete segmentation label is obtained by taking the maximum response category pixel by pixel. Its expression is: in, This indicates that the category with the largest response value is selected in the category dimension.

[0043] In this embodiment, discrete segmentation labels are used. The maximum connected components are extracted from regions belonging to the pubic symphysis and fetal head categories to remove scattered noise and false positive regions. Maximum connected component extraction itself is an existing image post-processing technique; however, combining it with subsequent confidence-weighted geometric modeling for automatic AoP measurement represents the overall innovative solution of this invention.

[0044] After the maximum connected component extraction is completed, the pubic symphysis region is obtained. and fetal head area Furthermore, the boundary point sets of both are extracted separately, denoted as... and .in: in, Indicates the first One pubic symphysis boundary point, This represents the total number of boundary points of the pubic symphysis; Indicates the first The boundary point of the fetal head, This represents the total number of fetal head boundary points. Taking an image from this embodiment as an example, after maximum connected component extraction, a clear pubic symphysis connected region and a complete fetal head connected region can be obtained, where the number of fetal head boundary sampling points can be, for example, taken as... The number of sampling points at the pubic symphysis boundary can be taken as an example. The number of sampling points is merely an example and does not constitute a limitation of the present invention.

[0045] S3. Based on the spatial confidence map, perform confidence-guided geometric modeling to obtain the pubic symphysis axis and the tangent of the fetal head. This step is one of the key innovative steps of this invention. Existing AoP measurement methods often treat all boundary points equally, while this invention utilizes spatial confidence maps. Different weights are assigned to the boundary points involved in geometric fitting to suppress the influence of noise points, blurred boundary points, and local outliers on geometric measurements. Specifically, the set of tire head boundary points... Each point in In the spatial confidence map Upsampling yields the corresponding confidence value: in, Represents boundary points The weights satisfy A higher confidence level indicates a more reliable geometric location of the boundary point in the current image, and it should have a greater weight in subsequent fitting; conversely, its influence should be weakened.

[0046] In this embodiment, weighted least squares method is preferably used to fit the tire head boundary to an ellipse. Let the general quadratic curve expression of the ellipse be: in, , , , , and These are the parameters to be determined; Represents two-dimensional plane coordinates. When the following conditions are met... At this point, the quadratic curve corresponds to an ellipse. The parameters of the fitted ellipse for the tire head boundary are obtained by minimizing the following weighted error function: in, Represents boundary points The residual value after substituting into the ellipse equation. After solving, the parameters of a general conic section can be converted into the center, major and minor axes, and rotation angle of an ellipse, denoted as: in, To fit the coordinates of the center of the ellipse; The length of the semi-major axis; The length of the semi-minor axis; Let be the rotation angle of the ellipse relative to the horizontal axis.

[0047] In the complete example of this embodiment, after performing weighted ellipse fitting on the fetal head boundary in a single ultrasound image, the resulting ellipse parameters can be, for example, the ellipse center. semi-major shaft Pixels, semi-short axis Pixels, rotation angle These values ​​are merely demonstration parameters for a complete processing case and do not constitute a limitation on the scope of protection of this invention.

[0048] For the pubic symphysis area In this embodiment, the axis is preferably determined by a region principal direction fitting method. Specifically, the set of boundary points of the pubic symphysis can be used. By performing least-squares linear fitting, the axis of the pubic symphysis was obtained. Then, determine the upper and lower endpoints on this axis, denoted as... and The tangent line on the outer side of the fetal head fitting ellipse, corresponding to the geometric configuration, forms the geometric relationship required for AoP measurement with the axis of the pubic symphysis. The key points related to the tangent line and the axis are denoted as follows: The specific numerical algorithm for determining the pubic symphysis axis and solving the tangent can adopt existing geometric analytical methods in this field. However, the key to this invention is that the pubic symphysis axis and the fetal head tangent are not directly generated by equal weighting of the original boundary, but are based on the aforementioned confidence-guided structural extraction and weighted fitting results.

[0049] S4. Calculate AoP based on the pubic symphysis axis and the tangent to the fetal head, and obtain the AoP confidence level. In this embodiment, AoP is calculated based on the triangular relationship formed by the axis of the pubic symphysis and the tangent of the fetal head in the image plane, and the expression is: in, Indicates the angle of labor progress, in degrees; Point With point The distance between them; Point With point The distance between them; Point With point The distance between them; Represents the inverse cosine function; Pi is the mathematical constant of a circle.

[0050] In the complete case of this embodiment, after analyzing the pubic symphysis axis and the tangent of the fetal head, the following can be obtained: Pixels Pixels Pixels. Substituting the above values ​​into the formula, we get: Therefore, the automatically estimated labor progression angle for this example image is approximately... .

[0051] To quantify the reliability of this AoP measurement result, this embodiment further calculates the AoP confidence level. Its expression is: in, Indicates the confidence level of the AoP measurement; The number of key boundary sampling points of the tire head involved in geometric modeling; Represents boundary points The confidence value on the spatial confidence map. In this embodiment, if the average confidence value of the 48 key tire head boundary points is 0.908, then: This indicates that the overall reliability of the current AoP results is relatively high. Unlike existing technologies that only output a single angle value, this invention can simultaneously output the AoP and its corresponding reliability quantification indicators, providing clinical users with more comprehensive reference information.

[0052] S5. Adaptive testing of geometric reliability constraints during the target domain testing phase. When the image to be tested comes from a hospital, equipment, or acquisition condition different from the training data, this embodiment preferably adapts during lightweight testing in the inference phase to enhance the model's robustness to domain shifts. This step is also one of the important innovative steps of this invention. Specifically, a probability map is generated from the segmentation logits of the current test image. Next, the optimization objective for unsupervised testing is constructed. The first part is the entropy minimization loss: in, The first part represents the pixel-level prediction entropy loss, which encourages the network to output more certain category predictions. The second part is the spatial smoothing loss. in, This is used to constrain the continuity of predictions for adjacent pixels and reduce local fragmentation predictions. The third part is the geometric reliability loss: in, The stability constant is greater than zero, for example, it can be taken as... The overall optimization goal during testing is: in, , and These are the loss weighting coefficients. As an implementable example, one could take... , , It should be noted that the above weights are parameters for the embodiment and do not constitute a limitation of the present invention.

[0053] In this embodiment, it is preferable to update only the affine parameters of the normalized layer and the linear layer parameters in the network, while keeping the other parameters frozen. As a specific implementation, three gradient updates can be performed on each target domain image, with forward inference restarted and the image refreshed after each update. And the segmentation probability graph. This lightweight update strategy of "updating only a few parameters" balances adaptability and online deployment stability.

[0054] In the complete case of this embodiment, assuming that the AoP confidence of the image is 0.842 when the adaptive test is not performed, the corresponding geometric reliability loss is approximately: After three lightweight tests and adaptive updates, the AoP confidence level increased to 0.908, and the corresponding geometric reliability loss became: At this point, it indicates that the geometric availability of the model on the current target domain samples has improved, and the final output AoP has increased from the initial estimate, for example... Stable to .

[0055] S6. Output the final result After completing the above steps, the system output includes: the final segmentation result of the pubic symphysis, the final segmentation result of the fetal head, the AoP value, and the AoP confidence level. In the complete case of this embodiment, the final output result is as follows: One pubic symphysis segmentation mask; One tire head segmentation mask; Labor progress angle ; Confidence level of labor progress angle .

[0056] like If the result is higher than a preset reliability threshold, such as 0.85, the measurement result can be marked as "highly reliable"; if it is lower than the threshold, the doctor is alerted to potential problems such as blurred boundaries, sound occlusion, or structural incompleteness in the image. This dual output format of "angle value + reliability" is a significant advantage of this invention in clinical applications.

[0057] Example 2: An automatic estimation method system for the angle of labor progression based on confidence guidance and reliability constraints includes: The data acquisition and preprocessing module is used to acquire and preprocess transperineal two-dimensional ultrasound images during labor as the raw input images. The structure-enhanced segmentation module is used to construct and train a joint segmentation and geometric modeling network; the original input image is input into the trained joint segmentation and geometric modeling network to generate segmentation results and spatial confidence maps; The geometric modeling module is used for confidence-guided geometric modeling based on spatial confidence maps to obtain the pubic symphysis axis and the tangent of the fetal head; and to calculate the labor progression angle based on the pubic symphysis axis and the tangent of the fetal head. The reliability assessment module is used to calculate the confidence level of the labor progress angle; The test-time adaptive module is used for lightweight test-time adaptation with geometric reliability constraints during the target domain testing phase. The results output module is used to output the final segmentation results of the pubic symphysis, the final segmentation results of the fetal head, the labor progression angle, and the confidence level of the labor progression angle.

[0058] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for automatic estimation of labor progress angles based on confidence-guided and reliability-constrained, characterized in that, Includes the following steps: S1. Acquire and preprocess transperineal two-dimensional ultrasound images during labor as the raw input images; S2. Construct and train the joint segmentation and geometric modeling network; input the original input image into the trained joint segmentation and geometric modeling network to generate segmentation results and spatial confidence maps; S3. Based on the spatial confidence map, perform confidence-guided geometric modeling to obtain the pubic symphysis axis and the tangent of the fetal head; S4. Calculate the labor progression angle based on the pubic symphysis axis and the tangent of the fetal head, and obtain the confidence level of the labor progression angle; S5. Adaptive testing of geometric reliability constraints during the target domain testing phase; S6. Output the final segmentation results of the pubic symphysis, the final segmentation results of the fetal head, the labor progression angle, and the confidence level of the labor progression angle.

2. The automatic estimation method for the production progress angle based on confidence guidance and reliability constraints according to claim 1, characterized in that, S2 include: The joint segmentation and geometric modeling network consists of a two-dimensional segmentation backbone, a three-branch shared coding structure, a spatial gating unit, a decoder, and a confidence prediction branch. The three-branch shared coding structure includes a global context branch for preserving the overall structural information of the original image and two local structural enhancement branches for highlighting key regions through spatial gating. Two spatial gating maps are predicted on shallow high-resolution features and multiplied pixel by pixel with the original input image to generate two complementary local enhanced views. The original input image and two local augmented views are simultaneously input into a shared encoder for multi-scale feature extraction, fused at each scale, and then fed into a decoder to obtain the segmentation result and spatial confidence output.

3. The automatic estimation method for the labor progress angle based on confidence guidance and reliability constraints according to claim 2, characterized in that, S2 also includes: For the segmentation results Perform softmax operation to obtain the probability map. : in, Indicates the first Line 1 Column pixels belong to category The probability of; Denotes a set of categories, where, Indicates background, Indicates the pubic symphysis. Indicates the fetal head; Represents an exponential function; By analyzing the segmentation results The discrete segmentation label is obtained by taking the maximum response category pixel by pixel. : in, This indicates that the category with the largest response value is selected in the category dimension; Discrete segmentation labels The pubic symphysis region is obtained by extracting the maximum connected components from the regions belonging to the pubic symphysis category and the fetal head category respectively. and fetal head area Extract the pubic symphysis region separately. and fetal head area The set of boundary points is used to obtain the set of boundary points of the pubic symphysis. set of fetal head boundary points : in, Indicates the first One pubic symphysis boundary point, This represents the total number of boundary points of the pubic symphysis; Indicates the first The boundary point of the fetal head, This represents the total number of boundary points of the fetal head.

4. The automatic estimation method for the production progress angle based on confidence guidance and reliability constraints according to claim 3, characterized in that, S3 include: Set of fetal head boundary points Each point in In the spatial confidence map Upsampling yields the corresponding confidence value: in, Represents boundary points The weights satisfy ; The tire head boundary is fitted to an ellipse using weighted least squares; the general quadratic curve expression of the ellipse is given by: in, , , , , and These are the parameters to be determined; Represents two-dimensional planar coordinates; when satisfying When this occurs, the corresponding quadratic curve is an ellipse; The parameters of the fitted ellipse for the tire head boundary are obtained by minimizing the following weighted error function: in, Represents boundary points The residual value after substituting into the ellipse equation; The parameters of a general conic section are converted into the center, major and minor axes, and rotation angle of an ellipse, denoted as: in, To fit the coordinates of the center of the ellipse; The length of the semi-major axis; This is the length of the semi-minor axis; Let be the rotation angle of the ellipse relative to the horizontal axis.

5. The automatic estimation method for the labor progress angle based on confidence guidance and reliability constraints according to claim 4, characterized in that, S3 also includes: Set of boundary points of the pubic symphysis By performing least-squares linear fitting, the axis of the pubic symphysis was obtained. Determine the upper and lower endpoints along the pubic symphysis axis, denoted as ; and A tangent line on the outer side of the fetal head fitting ellipse, corresponding to the geometric configuration, intersects the axis of the pubic symphysis. The geometric relationships required to establish the labor progression angle measurement are defined, where the key points related to the tangent and the axis are denoted as follows: .

6. The automatic estimation method for the production progress angle based on confidence guidance and reliability constraints according to claim 1, characterized in that, S4 include: The angle of labor progression is calculated based on the triangular relationship formed by the axis of the pubic symphysis and the tangent of the fetal head. in, Indicates the angle of labor progress, in degrees; Point With point The distance between them; Point With point The distance between them; Point With point The distance between them; Represents the inverse cosine function; Pi is the mathematical constant of a circle.

7. The automatic estimation method for the labor progress angle based on confidence guidance and reliability constraints according to claim 6, characterized in that, S4 also includes: Calculate the confidence level of the angle of labor progression : in, Indicates the confidence level of the AoP measurement; The number of key boundary sampling points of the tire head involved in geometric modeling; Represents boundary points Confidence values ​​on the spatial confidence map.

8. The automatic estimation method for the production progress angle based on confidence guidance and reliability constraints according to claim 1, characterized in that, S5 include: The optimization objective during unsupervised testing is firstly defined as the entropy minimization loss. in, A probability map generated for the segmentation logits of the current test image; This represents pixel-level prediction entropy loss, meaning it encourages the network to output more definitive category predictions. The second part is the spatial smoothing loss: in, Used to constrain the continuity of predictions for adjacent pixels and reduce local fragmentation predictions; The third part is the geometric reliability loss: in, The stability constant is greater than zero, for example, it can be taken as... ; The overall optimization goal during testing is: in, , and This is the loss weighting coefficient.

9. The automatic estimation method for the labor progress angle based on confidence guidance and reliability constraints according to claim 8, characterized in that, S5 also includes: Perform gradient updates three times for each target domain image. After each update, perform forward inference again and refresh the confidence level of the labor progress angle. And the segmentation probability graph.

10. A system for automatically estimating the angle of labor progression based on confidence-guided and reliability-constrained methods, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess transperineal two-dimensional ultrasound images during labor as the raw input images. The structure-enhanced segmentation module is used to build and train a joint segmentation and geometric modeling network; The original input image is input into the trained joint segmentation and geometric modeling network to generate segmentation results and spatial confidence maps. The geometric modeling module is used for confidence-guided geometric modeling based on spatial confidence maps to obtain the pubic symphysis axis and the tangent of the fetal head; and to calculate the labor progression angle based on the pubic symphysis axis and the tangent of the fetal head. The reliability assessment module is used to calculate the confidence level of the labor progress angle; The test-time adaptive module is used for lightweight test-time adaptation with geometric reliability constraints during the target domain testing phase. The results output module is used to output the final segmentation results of the pubic symphysis, the final segmentation results of the fetal head, the labor progression angle, and the confidence level of the labor progression angle. The method for automatically estimating the angle of labor progression based on confidence guidance and reliability constraints is used to implement the automatic estimation system for the angle of labor progression based on confidence guidance and reliability constraints as described in claim 1, and its steps.

Images