Semi-supervised learning-based labor process ultrasonic quantitative measurement method and system

Abstract

The invention provides a parturition process ultrasonic quantitative measurement method and system based on semi-supervised learning. The method comprises the following steps: constructing an image subset with pixel-level segmentation annotations, an image subset with key point annotations and an unannotated image subset based on a perineal ultrasound image; combining the medical image segmentation network pre-trained by using the image subsets with pixel-level segmentation labels with morphological and geometric algorithms to obtain a labor-process ultrasonic quantitative measurement model, and performing full-supervised training by using the image subsets with key point labels; constructing a plurality of semi-supervised birth process ultrasonic quantitative measurement models based on the birth process ultrasonic quantitative measurement model after full supervised training, and performing iterative training by using an image subset with pixel-level segmentation annotation and an unannotated image subset and adopting a sequential cross model strategy; and inputting the perineal ultrasonic image to be measured into the final semi-supervised birth process ultrasonic quantitative measurement model, and outputting a predicted birth process ultrasonic quantitative measurement result.

Description

Technical Field

[0001] This invention belongs to the field of medical imaging diagnosis, specifically relating to a method and system for quantitative measurement of labor ultrasound based on semi-supervised learning. Background Technology

[0002] Transperineal ultrasound (TPU) is the most direct, non-invasive, and repeatable imaging method for assessing fetal head descent in labor management. By imaging the relative position of the fetal head and pubic symphysis in real time, it provides objective geometric evidence of labor progress and is a safer and more quantifiable alternative to vaginal examination. Clinical studies have shown that subtle changes in fetal head position are closely related to delivery outcomes. Angulation progression reflects the angular trend of fetal head rotation and descent, while the head-pubic distance quantifies the depth of descent of the fetal head relative to the birth canal. These two indicators are of significant reference value in predicting the time of delivery, determining the mode of delivery, and assessing midwifery interventions.

[0003] However, traditional AoP and HSD measurements typically rely on operators manually marking key points or measuring angles on static ultrasound images. This process is significantly affected by experience, image clarity, and probe position, resulting in large measurement errors, poor repeatability, and high subjectivity. In complex or slow-progressing labor, different doctors may obtain significantly different measurement results for the same case, affecting clinical judgment. With the increasing demand for labor monitoring, automated and standardized quantitative measurement methods based on transperineal ultrasound have become an important direction for improving the efficiency and safety of labor management.

[0004] While existing deep learning-based automated measurement methods have made some progress in labor ultrasound analysis, they still face the following three key challenges:

[0005] Key Challenge 1: Image noise and structural complexity make target localization difficult: TPU image quality is significantly affected by factors such as speckle noise, acoustic artifacts, and probe orientation changes, leading to blurred structural boundaries. The pubic symphysis (PS) typically appears near a much larger and more brightly lit tissue area than the fetal head, making it difficult for the model to distinguish anatomical boundaries and achieve stable localization. Furthermore, the fetal head contour changes with the stage of labor, further increasing the complexity of automated detection.

[0006] Key Challenge Two: Scarcity of High-Quality Annotated Data, Facing the Needs of Fully Supervised Learning: The annotation process for labor ultrasound data must be completed by experienced obstetric ultrasound physicians, and different operators may differ in their segmentation of key points and structural boundaries. High annotation costs and strong subjectivity result in an extremely limited supply of high-quality pixel-level segmentation data. Currently available publicly available or clinically applicable data is insufficient to support end-to-end fully supervised model training, making the model prone to overfitting and lacking generalization ability.

[0007] Key Challenge 3: Available datasets contain only sparse landmark annotations, lacking structural information: Most existing datasets only provide sparse annotations at the keypoint or line segment level, lacking complete segmentation masks. Manually segmenting TPU images is not only labor-intensive and time-consuming, but also requires expert-level domain knowledge, greatly limiting the feasibility of large-scale accurate annotation. This point-based labeling cannot fully capture the geometric and topological relationship between the fetal head and pubic symphysis, while segmentation masks can encode richer anatomical structures and contextual information, which is crucial for robust measurement and geometric interpretation. Summary of the Invention

[0008] To overcome the limitations of existing deep learning-based automated labor measurement methods, such as difficulties in target localization due to image noise and structural complexity, and insufficient structural information due to sparse landmark annotations in available datasets, resulting in limited accuracy in structural recognition and geometric measurement in low-label scenarios, this invention provides a semi-supervised learning-based quantitative ultrasound measurement method and system for labor. This method aims to achieve accurate segmentation of the fetal head and pubic symphysis and automatic extraction of key points under limited labeled sample conditions, and further calculate two quantitative clinical indicators of labor: AoP and HSD. The method significantly improves the accuracy of structural recognition and geometric measurement in low-label scenarios through combined error-weighted fully supervised mask optimization and consistency-constrained pseudo-label iterative learning.

[0009] According to one aspect of the present invention, a method for quantitative measurement of labor ultrasound based on semi-supervised learning is provided, comprising:

[0010] Based on transperineal ultrasound images, image subsets with pixel-level segmentation annotations, image subsets with key point annotations, and unannotated image subsets were constructed;

[0011] The medical image segmentation network is pre-trained using a subset of images with pixel-level segmentation annotations. The pre-trained medical image segmentation network is then combined with morphological and geometric algorithms to obtain a quantitative measurement model of labor ultrasound. The quantitative measurement model of labor ultrasound is then trained in a fully supervised manner using a subset of images with key point annotations. The quantitative measurement model of labor ultrasound processes transperineal ultrasound images and outputs a dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0012] Based on the fully supervised training of the quantitative measurement model of labor ultrasound, several semi-supervised quantitative measurement models of labor ultrasound are constructed. Using image subsets with pixel-level segmentation annotations and unlabeled image subsets, a sequential cross-model strategy is adopted for iterative training. After training, the final semi-supervised quantitative measurement model of labor ultrasound is output.

[0013] The transperineal ultrasound image to be tested is input into the final semi-supervised quantitative ultrasound measurement model of labor, which outputs the predicted dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0014] As a further technical solution, the key points include: the upper end point of the pubic symphysis, the lower end point of the pubic symphysis, and the incision point of the fetal head.

[0015] As a further technical solution, the process of fully supervised training of the quantitative measurement model of labor ultrasound using a subset of images with key point annotations includes:

[0016] Each image from the subset of images with keypoint annotations is input into the pre-trained medical image segmentation network, which outputs the initial fetal head and pubic symphysis segmentation mask for each image in the subset of images with keypoint annotations.

[0017] The initial fetal head and pubic symphysis mask is post-processed by hole filling, connected component denoising and boundary smoothing to obtain the fetal head and pubic symphysis dense segmentation mask of each image in the image subset with key point annotation.

[0018] Based on morphological and geometric algorithms, the predicted key points of the dense segmentation mask of the fetal head and pubic symphysis of each image in the image subset with key point annotation are extracted. The dense segmentation mask of the fetal head and pubic symphysis of each image in the image subset with key point annotation and its predicted key points are used as training samples for the fully supervised stage.

[0019] Based on the distance error between the predicted key points and the labeled key points, the adaptive sample weights of each training sample in the fully supervised stage are calculated. The training samples are then used to train the quantitative measurement model of labor ultrasound, and the trained quantitative measurement model of labor ultrasound is output.

[0020] As a further technical solution, the steps for extracting key points in the dense segmentation mask of the fetal head and pubic symphysis based on morphological and geometric algorithms include:

[0021] Connectivity analysis was performed on the pubic symphysis mask region in the dense segmentation mask of the fetal head and pubic symphysis. The contour point set of the largest connected region was extracted, and the Euclidean distance between all pairs of points on the contour was calculated. The pair of points with the largest Euclidean distance was selected as the upper endpoint and lower endpoint of the pubic symphysis, respectively.

[0022] Extract the outline of the fetal head mask region in the dense segmentation mask of the fetal head and pubic symphysis, and find the tangent point and the foot of the fetal head on the outline; wherein, the tangent point of the fetal head satisfies the following: the line connecting the lower end of the pubic symphysis to the tangent point of the fetal head is perpendicular to the normal vector of the fetal head outline at the tangent point of the fetal head.

[0023] As a further technical solution, the formula for calculating the adaptive sample weights of each training sample in the fully supervised phase is as follows:

[0024]

[0025] In the formula, This represents the sample weight of the i-th training sample; This is the lower bound of the sample weights; Let Euclidean distance error be the predicted keypoint between the i-th training sample and the labeled keypoint. This represents the median of the global Euclidean distance error. It is the half-quartile difference.

[0026] As a further technical solution, an image subset with pixel-level segmentation annotations and an unannotated image subset are used, and a sequential cross-model strategy is employed for iterative training. Upon completion of training, the final semi-supervised quantitative measurement model of labor ultrasound is output. The steps include:

[0027] The unlabeled image set is grouped and alternately used as the training image subset for the semi-supervised iterative training process. The image subset with pixel-level segmentation annotation is used as the gold standard data for the semi-supervised iterative training process. Training is started, and a semi-supervised quantitative ultrasound measurement model for labor is output for each iteration. When the semi-supervised iterative training reaches the preset number of iterations or the preset training target, training is stopped. The final semi-supervised quantitative ultrasound measurement model for labor is output as the final semi-supervised quantitative ultrasound measurement model for labor.

[0028] As a further technical solution, the semi-supervised iterative training process includes:

[0029] In the first iteration, the dense segmentation mask of the fetal head and pubic symphysis obtained by processing the training image subset of the first semi-supervised labor ultrasound quantitative measurement model is used as the pseudo-label of the first iteration process and assigned a uniform weight. The training image subset of the first iteration process, the weighted pseudo-label, and the gold standard data are combined to train the second semi-supervised labor ultrasound quantitative measurement model. The trained second semi-supervised labor ultrasound quantitative measurement model is output as the current semi-supervised labor ultrasound quantitative measurement model of the first round.

[0030] Using the first semi-supervised ultrasound quantitative measurement model of labor as the initial reference model for the second round, and the current semi-supervised ultrasound quantitative measurement model of labor in the first round as the pseudo-label generation model for the second round, the second round of iteration process begins. The steps include: processing the current training image subset using the initial reference model and pseudo-label generation model of the second round, obtaining the output dense segmentation mask of the fetal head and pubic symphysis and key point annotations, and calculating the pseudo-label weights of the second round of iteration process using a cross-consistency evaluation mechanism. The dense segmentation mask of the fetal head and pubic symphysis output by the pseudo-label generation model of the second round is retained as the pseudo-label of the second round of iteration process. The training image subset of the second round of iteration process, the weighted pseudo-labels, and the gold standard data are combined to train the third semi-supervised ultrasound quantitative measurement model of labor, and the trained third semi-supervised ultrasound quantitative measurement model of labor is output as the current semi-supervised ultrasound quantitative measurement model of labor in the second round.

[0031] In subsequent iterations, the current semi-supervised ultrasound quantitative measurement model of labor from the two rounds prior to the current round is used as the initial reference model for the current round, and the current semi-supervised ultrasound quantitative measurement model of labor from the previous round is used as the pseudo-label generation model for the current round. The steps of the second round of iterations are repeated to output the current semi-supervised ultrasound quantitative measurement model of labor for the current round.

[0032] As a further technical solution, the pseudo-label weight is a dynamic confidence weight calculated based on the consistency score of the output results of the initial reference model and the pseudo-label generation model processing the same unlabeled image based on the current iteration round.

[0033] According to another aspect of this specification, a semi-supervised learning-based quantitative measurement system for labor ultrasound is provided, comprising:

[0034] The training data acquisition module is used to construct image subsets with pixel-level segmentation annotations, image subsets with key point annotations, and unannotated image subsets based on transperineal ultrasound images.

[0035] The fully supervised training module is used to pre-train a medical image segmentation network using a subset of images with pixel-level segmentation annotations. The pre-trained medical image segmentation network is then combined with morphological and geometric algorithms to obtain a quantitative measurement model of labor ultrasound. The quantitative measurement model of labor ultrasound is then trained in a fully supervised manner using a subset of images with key point annotations. The quantitative measurement model of labor ultrasound processes transperineal ultrasound images and outputs a dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0036] The semi-supervised training module is used to construct several semi-supervised quantitative measurement models of labor ultrasound based on the fully supervised training model. It uses image subsets with pixel-level segmentation annotations and unlabeled image subsets, and adopts a sequential cross model strategy for iterative training. After training, the final semi-supervised quantitative measurement model of labor ultrasound is output.

[0037] The quantitative measurement module for labor ultrasound is used to input the transperineal ultrasound image to be measured into the final semi-supervised quantitative measurement model for labor ultrasound, and output the predicted dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0038] According to another aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing program instructions executed by the processor, the processor invoking the program instructions to perform a semi-supervised learning-based quantitative measurement method for labor ultrasound.

[0039] According to another aspect of this specification, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a semi-supervised learning-based quantitative measurement method for labor ultrasound.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] The proposed method optimizes both fully supervised weighted loss and semi-supervised consistency loss during training, forming a joint constraint mechanism of anatomical segmentation and geometric measurement. Using this method, a semi-supervised quantitative ultrasound measurement model for labor can achieve robust learning under conditions of limited labeled samples and complex image noise. Finally, by inputting a single frame of transperineal ultrasound image during the inference phase, the model automatically outputs the coordinates of key points of the fetal head and pubic symphysis, along with corresponding AoP and HSD measurements, enabling real-time, objective, and repeatable assessment of labor progress.

[0042] This invention significantly reduces the reliance on manual annotation by combining prior anatomical structures with a pseudo-label consistency screening mechanism; it effectively improves the geometric measurement accuracy and stability of the model under low-annotation conditions through dual optimization of error weighting and consistency constraints; and it reduces the operational burden on doctors by automating the calculation of AoP and HSD indicators, providing objective, standardized and interpretable quantitative evidence for clinical assessment during labor.

[0043] This invention, in the face of limited labeled data and complex image noise, makes full use of unlabeled sample information to achieve automated learning that balances structural consistency constraints and geometric measurement accuracy. It enables automatic, objective, and operator-independent quantitative analysis of ultrasound images during labor, and has strong versatility and clinical application value. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart illustrating a method for quantitative measurement of labor ultrasound based on semi-supervised learning, provided as an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the dense segmentation mask of the fetal head and pubic symphysis in an embodiment of the present invention, as well as the key point annotations and quantitative clinical indicators of labor.

[0047] Figure 3 A schematic diagram of a quantitative ultrasound measurement system for labor based on semi-supervised learning, provided as an embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0049] It should be noted that:

[0050] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0051] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0053] like Figure 1 As shown, this invention discloses a semi-supervised learning-based quantitative measurement method for labor ultrasound, applicable to the automated analysis of transperineal ultrasound images during labor and the quantitative measurement of fetal head descent, especially for the intelligent estimation and visualization of Angle of Progression (AoP) and Head-Symphysis Distance (HSD). The method includes:

[0054] Step 1: Construct image subsets with pixel-level segmentation annotations, image subsets with key point annotations, and unannotated image subsets based on transperineal ultrasound images;

[0055] Step 2: Pre-train the medical image segmentation network using a subset of images with pixel-level segmentation annotations. Combine the pre-trained medical image segmentation network with morphological and geometric algorithms to obtain a quantitative measurement model of labor ultrasound. Then, use a subset of images with key point annotations to perform fully supervised training on the quantitative measurement model of labor ultrasound. The quantitative measurement model of labor ultrasound processes transperineal ultrasound images and outputs a dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0056] Step 3: Based on the fully supervised training of the quantitative measurement model of labor ultrasound, construct several semi-supervised quantitative measurement models of labor ultrasound, and use image subsets with pixel-level segmentation annotations and unlabeled image subsets to iteratively train using a sequential cross model strategy. After training, output the final semi-supervised quantitative measurement model of labor ultrasound.

[0057] Step 4: Input the transperineal ultrasound image to be tested into the final semi-supervised quantitative ultrasound measurement model of labor, and output the predicted dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0058] Step 1 essentially involves data integration and annotation expansion. Based on transperineal ultrasound images, a unified labor measurement data resource is constructed, including: a subset of images with pixel-level segmentation annotations, a subset of images with only key point annotations, and a subset of unannotated images.

[0059] In the image subset with pixel-level segmentation annotations, each image is annotated with a pixel-level segmentation mask for the fetal head and pubic symphysis; in the image subset with only keypoint annotations, the keypoints include: the upper end point of the pubic symphysis. The lower end of the pubic symphysis and the fetal head incision point .

[0060] In step 2, the medical image segmentation network is pre-trained using a subset of images with pixel-level segmentation annotations. This gives the medical image segmentation network the ability to extract basic anatomical features of the pubic symphysis and fetal head region, which is used to establish the anatomical structure prior of the medical image segmentation network, enabling the medical image segmentation network to output pixel-level segmentation masks of the pubic symphysis and fetal head.

[0061] Furthermore, in step 2, the process of fully supervised training of the quantitative measurement model of labor ultrasound using a subset of images with key point annotations includes:

[0062] Step 2-1: Input each image of the image subset with key point annotations into the pre-trained medical image segmentation network, and output the initial fetal head and pubic symphysis segmentation mask for each image in the image subset with key point annotations.

[0063] Step 2-2: Perform hole filling, connected component denoising and boundary smoothing post-processing on the initial fetal head and pubic symphysis mask to obtain the fetal head and pubic symphysis dense segmentation mask of each image in the image subset with key point annotations.

[0064] Steps 2-3: Based on morphological and geometric algorithms, extract the predicted key points of the dense segmentation mask of the fetal head and pubic symphysis in each image of the image subset with key point annotation, and calculate the quantitative clinical indicators of labor. Use the dense segmentation mask of the fetal head and pubic symphysis in each image of the image subset with key point annotation, as well as its predicted key points and quantitative clinical indicators of labor, as training samples for the fully supervised stage.

[0065] Steps 2-4: Calculate the adaptive sample weights of each training sample in the fully supervised stage based on the distance error between the predicted key points and the labeled key points. Use the weighted training samples with the adaptive sample weights to train the quantitative measurement model of labor ultrasound and output the trained quantitative measurement model of labor ultrasound.

[0066] In step 2-1, the pre-trained medical image segmentation has a priori network anatomical structure, which is used to infer the initial fetal head and pubic symphysis segmentation mask for each image based on the subset of images with only key point annotations. This mask is a coarse segmentation result.

[0067] In step 2-2, for the initial fetal head and pubic symphysis segmentation mask obtained in step 2-1 with only keypoint annotations, a dense segmentation mask for the fetal head and pubic symphysis is generated through artificial expansion (post-processing) to obtain a high-quality pseudo-segmentation mask that can be used as a training label. Optionally, the post-processing includes: hole filling: repairing closed void regions inside the mask; connected component denoising: based on connected component analysis, only the anatomical region with the largest area in the mask is retained, and isolated small noise points are removed; boundary smoothing: morphological processing of the mask contour to conform to anatomical geometric rules.

[0068] Preferably, step 2-2 further includes deriving and adding ground truth labels for the corresponding angle progression (AoP) and head-symphysis distance (HSD) based on the principles of anatomical geometry, thereby forming a hybrid fully supervised dataset that combines anatomical segmentation information with dual geometric measurement indicators of AoP and HSD, to support the joint learning of anatomical segmentation and geometric measurement.

[0069] Specifically, in steps 2-3, the steps of extracting key points from the dense segmentation mask of the fetal head and pubic symphysis based on morphological and geometric algorithms include:

[0070] like Figure 2 As shown, the key points of the pubic symphysis include the superior end point of the pubic symphysis. With the lower end of the pubic symphysis The line connecting these two points forms the long axis of the pubic symphysis. The extraction steps include:

[0071] Step 2-3-1: Perform connected component analysis on the pubic symphysis mask region in the dense segmentation mask of the fetal head and pubic symphysis, extract the contour point set of the largest connected region, and calculate the Euclidean distance between all pairs of points on the contour. Select the pair of points with the largest Euclidean distance as the upper endpoint and lower endpoint of the pubic symphysis, respectively.

[0072] Key points of the fetal head include the fetal head incision point, which is located at the lower end of the pubic symphysis. Draw a tangent line to the outline of the tire head mask area. - ( (This is the cutting point), which is the cutting point of the fetal head. The steps for extracting key points of the fetal head include:

[0073] Step 2-3-2: Extract the outline of the fetal head mask region in the dense segmentation mask of the fetal head and pubic symphysis, and find the fetal head tangent point and the fetal head foot point from the outline; wherein, the fetal head tangent point satisfies the following: the line connecting the lower end of the pubic symphysis to the fetal head tangent point is perpendicular to the normal vector of the fetal head outline at the fetal head tangent point.

[0074] In step 2, the fully supervised training of the quantitative measurement model of labor ultrasound can output a densely segmented mask of the fetal head and pubic symphysis in the transperineal ultrasound image, and can automatically extract key points and calculate quantitative clinical indicators of labor based on morphological and geometric algorithms.

[0075] Furthermore, such as Figure 2 As shown, the clinical indicators for quantitative measurement of labor by ultrasound include angular progression (AoP) and head-pubic distance (HSD). These two indicators together reflect the descent of the fetal head and the progress of labor, enabling automated and quantitative ultrasound measurement.

[0076] The specific calculations include: First, the angle of progression (AoP) is defined as the angle between the long axis of the pubic symphysis and the tangent of the fetal head, and the calculation formula is:

[0077]

[0078] Secondly, the head-pubic distance (HSD) is defined as the Euclidean distance from the lower end of the pubic symphysis to the point of fetal head drop, and the calculation formula is as follows:

[0079]

[0080] In the formula, This is the point where the fetal head drops to the foot.

[0081] Among them, the fetal head foot point is an auxiliary point for calculating the head-pubic distance, such as... Figure 2 As shown, the extraction process is as follows: [Extending to the lower end of the pubic symphysis] Draw a straight line perpendicular to the long axis of the pubic symphysis, and calculate the intersection of this line and the outline of the fetal head. Record this intersection as the point where the fetal head descends to its foot. The point where the fetal head drops to the foot of the pubic symphysis satisfies the following condition: the line connecting the lower end of the pubic symphysis to the point where the fetal head drops to the foot of the pubic symphysis is perpendicular to the line connecting the lower end of the pubic symphysis to the upper end of the pubic symphysis.

[0082] Specifically, in steps 2-4, after back-extracting predicted key points from the pseudo segmentation mask (fully supervised mask) generated in step 2-3, the Euclidean distance error between the predicted key points and the labeled real key points (labels in the image subset with key point annotations) is calculated. Based on this error, adaptive sample weighting coefficients for each training sample are constructed for error weighting optimization.

[0083] The formula for calculating the adaptive sample weights of each training sample in the fully supervised phase is as follows:

[0084]

[0085] In the formula, The sample weights represent the training samples in the fully supervised phase, where i is the training sample number. This is the lower bound of the sample weights; Let Euclidean distance error be the predicted keypoint between the i-th training sample and the labeled keypoint. This represents the median of the global Euclidean distance error. It is the half-quartile difference.

[0086] Furthermore, to improve keypoint matching accuracy and suppress the influence of outliers, the aforementioned adaptive sample weighting coefficients are utilized. For the current batch The loss of all samples within the range is weighted and averaged to calculate the final batch training loss:

[0087]

[0088] In the formula, Indicates batch training loss; Let represent the cross-entropy loss of the i-th training sample; Let B represent the Dice loss of the i-th training sample; B represents the set of training sample numbers within the batch.

[0089] In step 2, during the fully supervised training phase, the quantitative measurement model of labor ultrasound is trained using limited labeled data through mask generation, and the sample weights are adaptively adjusted based on the key point prediction error. By defining a weighting function, abnormal training samples are suppressed and training stability is improved. This mechanism ensures that the network can still converge stably when the sample differences are significant, thereby enhancing the robust recognition ability of anatomical structures.

[0090] Step 3, which involves iteratively training using a sequential crossover model strategy with an unlabeled subset of images, includes:

[0091] The unlabeled image set is grouped and alternately used as the training image subset for the semi-supervised iterative training process. The image subset with pixel-level segmentation annotation is used as the gold standard data for the semi-supervised iterative training process. Training is started, and a semi-supervised quantitative ultrasound measurement model for labor is output for each iteration. When the semi-supervised iterative training reaches the preset number of iterations or the preset training target, training is stopped. The final semi-supervised quantitative ultrasound measurement model for labor is output as the final semi-supervised quantitative ultrasound measurement model for labor.

[0092] The iterative process includes:

[0093] In the first iteration, the dense segmentation mask of the fetal head and pubic symphysis obtained by processing the training image subset of the first semi-supervised labor ultrasound quantitative measurement model is used as the pseudo-label of the first iteration process and assigned a uniform weight. The training image subset of the first iteration process, the weighted pseudo-label, and the gold standard data are combined to train the second semi-supervised labor ultrasound quantitative measurement model. The trained second semi-supervised labor ultrasound quantitative measurement model is output as the current semi-supervised labor ultrasound quantitative measurement model of the first round.

[0094] Using the first semi-supervised ultrasound quantitative measurement model of labor as the initial reference model for the second round, and the current semi-supervised ultrasound quantitative measurement model of labor in the first round as the pseudo-label generation model for the second round, the second round of iteration process begins. The steps include: processing the current training image subset using the initial reference model and pseudo-label generation model of the second round, obtaining the output dense segmentation mask of the fetal head and pubic symphysis and key point annotations, and calculating the pseudo-label weights of the second round of iteration process using a cross-consistency evaluation mechanism. The dense segmentation mask of the fetal head and pubic symphysis output by the pseudo-label generation model of the second round is retained as the pseudo-label of the second round of iteration process. The training image subset of the second round of iteration process, the weighted pseudo-labels, and the gold standard data are combined to train the third semi-supervised ultrasound quantitative measurement model of labor, and the trained third semi-supervised ultrasound quantitative measurement model of labor is output as the current semi-supervised ultrasound quantitative measurement model of labor in the second round.

[0095] In subsequent iterations, the current semi-supervised ultrasound quantitative measurement model of labor from the two rounds prior to the current round is used as the initial reference model for the current round, and the current semi-supervised ultrasound quantitative measurement model of labor from the previous round is used as the pseudo-label generation model for the current round. The steps of the second round of iterations are repeated to output the current semi-supervised ultrasound quantitative measurement model of labor for the current round.

[0096] The pseudo-label weight is calculated based on the consistency score of the output results of the first semi-supervised ultrasound quantitative measurement model and the second semi-supervised ultrasound quantitative measurement model in processing the same unlabeled image.

[0097] Specifically, the consistency score of the output results for the same unlabeled image. Defined jointly by the pixel overlap ratio (IoU) of the mask and the spatial distance of the keypoints:

[0098]

[0099] in, This represents a function for calculating pixel-level overlap rate. and Let represent the dense segmentation masks of the fetal head and pubic symphysis of the i-th training sample generated by the initial reference model and the pseudo-label generation model, respectively, in the current iteration round. and Here are the coordinates of the j-th key point in the dense segmentation mask corresponding to the fetal head and pubic symphysis. Key points; The median of the global spatial distance of the j-th keypoint; This is the balance coefficient.

[0100] income This is used to control the confidence weight of pseudo-labels in semi-supervised training, and to weight pseudo-label samples in the loss function, thereby ensuring that high-confidence samples dominate model updates. Through this batch-wise, sequential, intergenerational consistent weighted iterative optimization, high-confidence samples dominate model updates, thus improving model segmentation performance while continuously enhancing the predictive consistency between models.

[0101] Optionally, in step 3, the preset training objective in the semi-supervised iterative training process is: training stops when the consistency score between the two semi-supervised models no longer improves significantly.

[0102] As an optional implementation, the iterative training steps using a sequential crossover model strategy include: First, using the weights of the fully supervised training optimized quantitative ultrasound measurement model for labor, several structurally identical deep learning models (consistent with the quantitative ultrasound measurement model for labor) are initialized as semi-supervised quantitative ultrasound measurement models for labor. The unlabeled image set (utilizing large-scale data to improve the model's generalization ability) is then divided into three non-overlapping image subsets. , ),according to The order of each iteration is as follows: A subset of images is selected in each iteration. The specific iteration process is as follows:

[0103] Based on the unlabeled image subset described in step 1, a sequential semi-supervised training process is constructed.

[0104] Subsequently, iterative training was performed in the following order:

[0105] First iteration (guided phase): Using the first semi-supervised labor ultrasound quantitative measurement model with the initial reference model. For the first unlabeled subset Inference is performed, and the generated binarized prediction mask (dense segmentation mask of fetal head and pubic symphysis) is used as the pseudo-label for the first round of iteration. This batch of pseudo-labels is assigned a uniform low confidence weight (0.5), and then mixed with the gold standard data (the subset of images with pixel-level segmentation annotations from step 1) to train the second semi-supervised quantitative ultrasound measurement model of labor. The trained second semi-supervised quantitative ultrasound measurement model of labor serves as the first-generation current model (denoted as...). ).

[0106] Second iteration (consistency constraint phase I): Utilizing the initial reference model (first semi-supervised labor ultrasound quantitative measurement model) ) and the first generation current model Together on the second unlabeled subset Perform reasoning. (Keep) The predicted mask is used as a pseudo-label, and its correlation with the target mask is calculated. The cross-consistency score between the outputs is mapped to a dynamic confidence weight, and this weighted pseudo-label is used in conjunction with the gold standard data to train a third-half-supervised ultrasound quantitative measurement model of labor. The trained third-half-supervised ultrasound quantitative measurement model of labor serves as the second-generation current model (denoted as...). ).

[0107] Third iteration (Consistency Constraint Phase II): Utilizing the first-generation current model Compared with the current second-generation model Common on the third unlabeled subset Perform reasoning. (Keep) The predicted mask is used as a pseudo-label, and its correlation with the target mask is calculated. The cross-consistency score between the output results is used. This score is mapped to a dynamic confidence weight, and the weighted pseudo-labels are combined with the gold standard data to train a fourth-generation semi-supervised ultrasound quantitative measurement model. The trained fourth-generation semi-supervised ultrasound quantitative measurement model is the current third-generation model (denoted as...). ).

[0108] Similarly, training stops when the semi-supervised iterative training reaches the preset number of iterations (e.g., 4 rounds) or when the consistency score between models no longer improves significantly.

[0109] The transperineal ultrasound image to be tested is input into the final optimized segmentation model, and the output is a binarized image containing the pubic symphysis and the fetal head region, which is the final segmentation mask.

[0110] Step 4 essentially involves calculating geometric parameters and outputting labor measurement data. After the semi-supervised iterative training is completed, the transperineal ultrasound image to be measured is input into the final semi-supervised quantitative labor ultrasound measurement model. The model outputs a densely segmented mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor, which are used for real-time monitoring and objective assessment of labor, reducing human measurement errors and improving the reliability of clinical decisions.

[0111] The implementation of the various embodiments of the present invention is based on programmed processing through a system with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a system A, which is used to execute one of the methods described in the above method embodiments.

[0112] See Figure 3 The system includes:

[0113] The training data acquisition module is used to construct image subsets with pixel-level segmentation annotations, image subsets with key point annotations, and unannotated image subsets based on transperineal ultrasound images.

[0114] The fully supervised training module is used to pre-train a medical image segmentation network using a subset of images with pixel-level segmentation annotations. The pre-trained medical image segmentation network is then combined with morphological and geometric algorithms to obtain a quantitative measurement model of labor ultrasound. The quantitative measurement model of labor ultrasound is then trained in a fully supervised manner using a subset of images with key point annotations. The quantitative measurement model of labor ultrasound processes transperineal ultrasound images and outputs a dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0115] The semi-supervised training module is used to construct several semi-supervised quantitative measurement models of labor ultrasound based on the fully supervised training model. It uses image subsets with pixel-level segmentation annotations and unlabeled image subsets, and adopts a sequential cross model strategy for iterative training. After training, the final semi-supervised quantitative measurement model of labor ultrasound is output.

[0116] The quantitative measurement module for labor ultrasound is used to input the transperineal ultrasound image to be measured into the final semi-supervised quantitative measurement model for labor ultrasound, and output the predicted dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

[0117] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0118] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 4As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0119] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0120] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0121] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0122] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0123] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0124] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0125] Based on the same technical concept as the foregoing embodiments, the present invention provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute a semi-supervised learning-based quantitative measurement method for labor ultrasound.

[0126] In summary, this invention discloses a semi-supervised learning method and system for quantitative measurement of fetal head descent in ultrasound images during labor. This method proposes a two-stage learning framework combining fully supervised and semi-supervised training for the automatic measurement tasks of Angle of Progression (AoP) and Head-Symphysis Distance (HSD). In the first stage, sparse keypoint information is transformed into dense anatomical segmentation labels through external segmentation pre-training and point-annotation-based mask generation. An error weighting mechanism is introduced to dynamically adjust the sample contribution, thereby obtaining an initial model representation with stronger structural consistency. In the second stage, a pseudo-label generation mechanism with multi-model interaction is introduced. Parallel inference is performed on unlabeled images, and the confidence of pseudo-labels is determined through pixel-level and keypoint-level consistency calculations. An adaptive weighting strategy is used for iterative optimization to gradually improve the quality of pseudo-labels and the model's generalization ability. During the inference stage, the model automatically extracts the upper and lower endpoints of the pubic symphysis and the tangent points of the fetal head, calculates AoP and HSD based on geometric relationships, and achieves synchronous and automated measurement of the fetal head descent angle and distance. The method of this invention can make full use of unlabeled ultrasound data under limited labeling conditions, significantly improve the stability and accuracy of fetal head descent measurement, and realize automatic, objective and operator-independent quantitative analysis of ultrasound images during labor. It has strong versatility and clinical application value.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for quantitative measurement of labor by ultrasound based on semi-supervised learning, characterized in that, include: Based on transperineal ultrasound images, image subsets with pixel-level segmentation annotations, image subsets with key point annotations, and unannotated image subsets were constructed; The medical image segmentation network is pre-trained using a subset of images with pixel-level segmentation annotations. The pre-trained medical image segmentation network is then combined with morphological and geometric algorithms to obtain a quantitative measurement model of labor ultrasound. The quantitative measurement model of labor ultrasound is then trained in a fully supervised manner using a subset of images with key point annotations. The quantitative measurement model of labor ultrasound processes transperineal ultrasound images and outputs a dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor. Based on the fully supervised training of the quantitative measurement model of labor ultrasound, several semi-supervised quantitative measurement models of labor ultrasound are constructed. Using image subsets with pixel-level segmentation annotations and unlabeled image subsets, a sequential cross-model strategy is adopted for iterative training. After training, the final semi-supervised quantitative measurement model of labor ultrasound is output. The transperineal ultrasound image to be tested is input into the final semi-supervised quantitative ultrasound measurement model of labor, which outputs the predicted dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

2. The method for quantitative measurement of labor ultrasound based on semi-supervised learning as described in claim 1, characterized in that, The process of fully supervised training of a quantitative measurement model of labor ultrasound using a subset of images with key point annotations includes: Each image from the subset of images with keypoint annotations is input into the pre-trained medical image segmentation network, which outputs the initial fetal head and pubic symphysis segmentation mask for each image in the subset of images with keypoint annotations. The initial fetal head and pubic symphysis mask is post-processed by hole filling, connected component denoising and boundary smoothing to obtain the fetal head and pubic symphysis dense segmentation mask of each image in the image subset with key point annotation. Based on morphological and geometric algorithms, the predicted key points of the dense segmentation mask of the fetal head and pubic symphysis of each image in the image subset with key point annotation are extracted. The dense segmentation mask of the fetal head and pubic symphysis of each image in the image subset with key point annotation and its predicted key points are used as training samples for the fully supervised stage. Based on the distance error between the predicted key points and the labeled key points, the adaptive sample weights of each training sample in the fully supervised stage are calculated. The training samples are then used to train the quantitative measurement model of labor ultrasound, and the trained quantitative measurement model of labor ultrasound is output.

3. The method for quantitative measurement of labor ultrasound based on semi-supervised learning as described in claim 2, characterized in that, The upper end of the pubic symphysis, the lower end of the pubic symphysis, and the incision point of the fetal head; The steps for extracting key points from the dense segmentation mask of the fetal head and pubic symphysis based on morphological and geometric algorithms include: Connectivity analysis was performed on the pubic symphysis mask region in the dense segmentation mask of the fetal head and pubic symphysis. The contour point set of the largest connected region was extracted, and the Euclidean distance between all pairs of points on the contour was calculated. The pair of points with the largest Euclidean distance was selected as the upper endpoint and lower endpoint of the pubic symphysis, respectively. Extract the outline of the fetal head mask region in the dense segmentation mask of the fetal head and pubic symphysis, and find the tangent point and the foot of the fetal head on the outline; wherein, the tangent point of the fetal head satisfies the following: the line connecting the lower end of the pubic symphysis to the tangent point of the fetal head is perpendicular to the normal vector of the fetal head outline at the tangent point of the fetal head.

4. The method for quantitative measurement of labor ultrasound based on semi-supervised learning as described in claim 3, characterized in that, The formula for calculating the adaptive sample weights of each training sample in the fully supervised phase is as follows: ； In the formula, This represents the sample weight of the i-th training sample; This is the lower bound of the sample weights; Let Euclidean distance error be the predicted keypoint between the i-th training sample and the labeled keypoint. This represents the median of the global Euclidean distance error. It is the half-quartile difference.

5. The method for quantitative measurement of labor ultrasound based on semi-supervised learning as described in claim 1, characterized in that, Using image subsets with pixel-level segmentation annotations and unlabeled image subsets, a sequential crossover model strategy is employed for iterative training. Upon completion of training, the final semi-supervised quantitative ultrasound measurement model for labor is output. The steps include: The unlabeled image set is grouped and alternately used as the training image subset for the semi-supervised iterative training process. The image subset with pixel-level segmentation annotation is used as the gold standard data for the semi-supervised iterative training process. Training is started, and a semi-supervised quantitative ultrasound measurement model for labor is output for each iteration. When the semi-supervised iterative training reaches the preset number of iterations or the preset training target, training is stopped. The final semi-supervised quantitative ultrasound measurement model for labor is output as the final semi-supervised quantitative ultrasound measurement model for labor.

6. The method for quantitative measurement of labor ultrasound based on semi-supervised learning as described in claim 5, characterized in that, in, The semi-supervised iterative training process includes: In the first iteration, the dense segmentation mask of the fetal head and pubic symphysis obtained by processing the training image subset of the first semi-supervised labor ultrasound quantitative measurement model is used as the pseudo-label of the first iteration process and assigned a uniform weight. The training image subset of the first iteration process, the weighted pseudo-label, and the gold standard data are combined to train the second semi-supervised labor ultrasound quantitative measurement model. The trained second semi-supervised labor ultrasound quantitative measurement model is output as the current semi-supervised labor ultrasound quantitative measurement model of the first round. Using the first semi-supervised ultrasound quantitative measurement model of labor as the initial reference model for the second round, and the current semi-supervised ultrasound quantitative measurement model of labor in the first round as the pseudo-label generation model for the second round, the second round of iteration process begins. The steps include: processing the current training image subset using the initial reference model and pseudo-label generation model of the second round, obtaining the output dense segmentation mask of the fetal head and pubic symphysis and key point annotations, and calculating the pseudo-label weights of the second round of iteration process using a cross-consistency evaluation mechanism. The dense segmentation mask of the fetal head and pubic symphysis output by the pseudo-label generation model of the second round is retained as the pseudo-label of the second round of iteration process. The training image subset of the second round of iteration process, the weighted pseudo-labels, and the gold standard data are combined to train the third semi-supervised ultrasound quantitative measurement model of labor, and the trained third semi-supervised ultrasound quantitative measurement model of labor is output as the current semi-supervised ultrasound quantitative measurement model of labor in the second round. In subsequent iterations, the current semi-supervised ultrasound quantitative measurement model of labor from the two rounds prior to the current round is used as the initial reference model for the current round, and the current semi-supervised ultrasound quantitative measurement model of labor from the previous round is used as the pseudo-label generation model for the current round. The steps of the second round of iterations are repeated to output the current semi-supervised ultrasound quantitative measurement model of labor for the current round.

7. The method for quantitative measurement of labor ultrasound based on semi-supervised learning as described in claim 6, characterized in that, The pseudo-label weight is a dynamic confidence weight calculated based on the consistency score of the output results of the initial reference model and the pseudo-label generation model processing the same unlabeled image in the current iteration round.

8. A quantitative ultrasound measurement system for labor based on semi-supervised learning, characterized in that, include: The training data acquisition module is used to construct image subsets with pixel-level segmentation annotations, image subsets with key point annotations, and unannotated image subsets based on transperineal ultrasound images. The fully supervised training module is used to pre-train a medical image segmentation network using a subset of images with pixel-level segmentation annotations. The pre-trained medical image segmentation network is then combined with morphological and geometric algorithms to obtain a quantitative measurement model of labor ultrasound. The quantitative measurement model of labor ultrasound is then trained in a fully supervised manner using a subset of images with key point annotations. The quantitative measurement model of labor ultrasound processes transperineal ultrasound images and outputs a dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor. The semi-supervised training module is used to construct several semi-supervised quantitative measurement models of labor ultrasound based on the fully supervised training model. It uses image subsets with pixel-level segmentation annotations and unlabeled image subsets, and adopts a sequential cross model strategy for iterative training. After training, the final semi-supervised quantitative measurement model of labor ultrasound is output. The quantitative measurement module for labor ultrasound is used to input the transperineal ultrasound image to be measured into the final semi-supervised quantitative measurement model for labor ultrasound, and output the predicted dense segmentation mask of the fetal head and pubic symphysis, as well as key point annotations and quantitative clinical indicators of labor.

9. An electronic device, characterized in that, The method includes a memory and a processor, the memory storing program instructions that are executed by the processor, the processor invoking the program instructions to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method described in any one of claims 1 to 7.

Images