Method for acquiring image of standard section of limbs in prenatal ultrasound in early pregnancy period

Abstract

The application relates to an early pregnancy prenatal ultrasound four-limb standard section image acquisition method, which comprises the following steps: in the early pregnancy prenatal detection process, inputting four-limb section ultrasound images in a real-time ultrasound video stream into a target prediction model to detect each sub-limb structure in the four-limb section ultrasound images and predict a first image quality score of each sub-limb structure; determining a corresponding sub-section type of the four-limb section ultrasound images under a four-limb section type according to the detected each sub-limb structure; determining an image quality score of each four-limb section ultrasound image based on the first image quality score of each sub-limb structure in the four-limb section ultrasound image; in the case that the four-limb section ultrasound images corresponding to the same sub-section type in the video stream are multiple, selecting a standard section image under the sub-section type from the multiple four-limb section ultrasound images of the same sub-section type based on the image quality score. The method can automatically select high-quality early pregnancy four-limb standard sections in real time.

Description

Technical Field

[0001] This application relates to the field of image processing technology, specifically to a method for acquiring standard cross-sectional images of the limbs via prenatal ultrasound during early pregnancy. Background Technology

[0002] In early pregnancy prenatal testing, standard sections of the fetal limbs are important auxiliary diagnostic tools for doctors. In traditional methods, when using an ultrasound scanner, doctors review the video stream generated during the scan and, based on their subjective experience, determine the standard sections related to the limbs by using techniques such as frame pausing and backtracking.

[0003] As a result, doctors need to maintain a high level of concentration for a long time during the scanning process, repeatedly looking back at the frame to determine the standard cross-sections related to the limbs, which is time-consuming and very inefficient. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for acquiring standard cross-sectional images of the limbs via prenatal ultrasound in early pregnancy, in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a method for acquiring standard cross-sectional images of the limbs via prenatal ultrasound during early pregnancy, the method comprising: During prenatal testing in early pregnancy, the ultrasound images of the limbs in the real-time ultrasound video stream are input into the target prediction model to detect the structure of each sub-limb in the ultrasound images of the limbs and predict the first image quality score of each sub-limb structure. Based on the detected sub-limb structures, determine the sub-section type corresponding to the limb section ultrasound image under the limb section type; For each limb section ultrasound image, an image quality score is determined based on a first image quality score of each of the sub-limb structures in the limb section ultrasound image; Based on the image quality score, a standard section image under the same sub-section type is selected from multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream; wherein, the image quality score of the standard section image is higher than that of the non-standard section image.

[0006] Secondly, this application provides a device for acquiring standard cross-sectional images of the limbs via prenatal ultrasound during early pregnancy, the device comprising: The model prediction module is used to input the limb cross-sectional ultrasound images from the real-time acquired ultrasound video stream into the target prediction model during prenatal testing in early pregnancy, so as to detect the structure of each sub-limb in the limb cross-sectional ultrasound images and predict the first image quality score of each sub-limb structure. The sub-section type identification module is used to determine the sub-section type corresponding to the ultrasound image of the limb section under the limb section type based on the detected sub-limb structures; An image quality assessment module is used to determine the image quality score of each limb section ultrasound image based on a first image quality score of each sub-limb structure in the limb section ultrasound image. The standard section selection module is used to select a standard section image of the same sub-section type from multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream based on the image quality score; wherein the image quality score of the standard section image is higher than that of the non-standard section image.

[0007] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in the first aspect.

[0008] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0009] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0010] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for acquiring standard limb section images via prenatal ultrasound in early pregnancy innovatively propose a novel target prediction model. During prenatal testing in early pregnancy, this model can automatically detect key sub-limb structures in the limb section ultrasound image frame-by-frame in the ultrasound video stream, and automatically predict the first image quality score for each sub-limb structure. Furthermore, based on the detected key sub-limb structures, the sub-section type corresponding to the limb section ultrasound image under the limb section type can be automatically determined. Based on the first image quality score of each sub-limb structure, combined with a scoring quantification mechanism following medical image interpretation standards, the image quality score of the limb section ultrasound image can be objectively and reasonably quantified and evaluated. Then, from multiple limb section ultrasound images corresponding to the same sub-section type, the limb section ultrasound image with the higher image quality score under that sub-section type is selected as the standard section image. In this way, during prenatal testing in early pregnancy, standard cross-sections of the limbs in early pregnancy that meet the standard definition of medical images and have high image quality can be automatically obtained in real time from the acquired ultrasound video stream, which greatly reduces the doctor's operating costs and improves the efficiency of obtaining standard cross-sections. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a method for acquiring standard cross-sectional images of the limbs via prenatal ultrasound in early pregnancy, as described in one embodiment. Figures 2 to 4 This is a schematic diagram of the output detection frame in a cross-sectional ultrasound image of the limbs in one embodiment; Figure 5 This is a schematic diagram illustrating the principle of a feature fusion gating unit generating a first weighted feature map in one embodiment. Figure 6 This is a schematic diagram of the target prediction model in one embodiment; Figure 7 and Figure 8 This is a schematic diagram comparing image quality in one embodiment; Figure 9 This is a flowchart illustrating the multi-round iterative training steps of a target prediction model in one embodiment; Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0013] It should be understood that, unless otherwise defined, all technical terms used herein have the same meaning as commonly understood by those skilled in the art. The technical terms used herein are for the purpose of describing specific examples only and are not intended to limit the scope of this application.

[0014] like Figure 1 As shown, in one embodiment, a method for acquiring standard cross-sectional images of the limbs via prenatal ultrasound in early pregnancy is provided. This method is applied to a computer device and specifically includes the following steps: S11, during prenatal testing in early pregnancy, the ultrasound images of the limbs in the real-time acquired ultrasound video stream are input into the target prediction model to detect the structure of each sub-limb in the ultrasound images of the limbs and predict the first image quality score of each sub-limb structure.

[0015] For example, the target prediction model can output the first detection box and the corresponding structure category of each detected sub-limb structure, which can be used as a basis for judgment for doctors to review.

[0016] In some embodiments, in addition to detecting individual sub-limb structures in the limb sectional ultrasound image, the target prediction model can also detect the sectional contour (such as the detection box of the output sectional contour) and the image quality score of the sectional contour in the limb sectional ultrasound image based on the first feature map. The detection box of the detected sectional contour (including the position of the detection box and the contour category corresponding to the detection box) can be finally output and displayed as a basis for judgment by the doctor; the image quality score of the sectional contour can be used as intermediate data for subsequent calculation of the final image quality score of the output limb sectional ultrasound image.

[0017] Figures 2 to 4 The detection boxes in the images indicate the sub-limb structures and cross-sectional contours detected in the limb cross-sectional ultrasound images. It should be understood that... Figures 2 to 4 The largest bounding box is the bounding box for the cross-sectional contour, while the other smaller bounding boxes are used to mark the sub-limb structures.

[0018] In some embodiments, during prenatal testing in early pregnancy, a computer device can initially screen limb sectional ultrasound images from a real-time acquired ultrasound video stream. Exemplarily, the real-time acquired ultrasound video stream is input into a preliminary section classification model. The preliminary section classification model can identify the section type for each ultrasound image in the ultrasound video stream to preliminarily determine the section type of each ultrasound image. If the section type of an ultrasound image is identified as belonging to the limb section type, the ultrasound image is determined to be a limb sectional ultrasound image, and the limb sectional ultrasound image is input into a target prediction model.

[0019] It should be understood that the preliminary section classification model is another model used for preliminary auxiliary identification of limb section ultrasound images. The sample ultrasound images of the preliminary section classification model have corresponding labeled limb section types. Based on these labeled sample ultrasound images, supervised deep learning training is performed iteratively to obtain the preliminary section classification model.

[0020] The target prediction model is a deep learning model pre-trained through multiple rounds of iterative training using a set of sample limb cross-sectional ultrasound images. The specific iterative training process will be described below. Compared with traditional object detection models, the target prediction model in this embodiment can not only identify target objects, but also perform quality assessment on local regions of the image. Therefore, after inputting the limb cross-sectional ultrasound images into the target prediction model, the target prediction model can detect each sub-limb structure in the limb cross-sectional ultrasound images (e.g., output detection boxes and structure categories for sub-limb structures) and predict the image quality score for each sub-limb structure.

[0021] In some embodiments, the computer device inputs limb cross-sectional ultrasound images from a real-time acquired ultrasound video stream into a target prediction model. The target prediction model can extract a first feature map from the limb cross-sectional ultrasound image, and predict a first structure detection result of sub-limb structures in the limb cross-sectional ultrasound image based on the first feature map. The first structure detection result includes a first detection box for locating and marking the sub-limb structure, a first confidence level of the first detection box, a probability distribution of a first structure category corresponding to the first detection box, and a feature map of a first region of interest indicated by the first detection box. A first feature gating coefficient is calculated based on the first confidence level and the maximum probability value in the first structure category probability distribution. The first region of interest feature map is weighted based on the first feature gating coefficient to obtain a first weighted feature map. A first image quality score of the sub-limb structure marked by the first detection box is predicted based on the first weighted feature map.

[0022] It should be understood that when there is no sub-limb structure in the image or the sub-limb structure features are not obvious, resulting in a low second confidence level for detection, the first feature gating coefficient α approaches 0. After the first feature map is weighted based on the first feature gating coefficient α, the feature values ​​that participate in the quality assessment process are suppressed, thereby blocking the transmission of noise features to the quality assessment process at the physical level.

[0023] In some embodiments, the computer device can also predict the probability distribution of the first section contour category in the limb section ultrasound image based on the first feature map. Then, the first feature gating coefficient α is calculated based on the first confidence level C (i.e., the confidence level of the first detection box output by the substructure detection branch), the maximum probability value in the first structure category probability distribution (hereinafter referred to as the maximum structure category probability) P1, and the maximum probability value in the first section contour category probability distribution (hereinafter referred to as the maximum section contour category probability) P2.

[0024] For example, such as Figure 5 As shown, the feature fusion gating unit in the target prediction model multiplies the first confidence level C, the maximum structure class probability P1, and the maximum cross-sectional contour class probability P2 to generate the first feature gating coefficient α, i.e., α = C × P1 × P2. This first feature gating coefficient α is then broadcast and combined with the first region of interest (ROI) feature map output by the substructure detection branch (i.e., the ROI feature map, which can be denoted as...) through a broadcasting mechanism. Perform element-wise multiplication (i.e., broadcast multiplication as shown in the diagram) to obtain the first weighted feature map, for example, the first weighted feature map. Furthermore, the first weighted feature map can be input into the quality assessment branch (such as the quality assessment head or quality assessment module) in the target prediction model to predict the first image quality score of the sub-limb structure marked by the first detection box.

[0025] In some examples, the target prediction model includes a trained feature extraction unit, a substructure detection branch, a feature fusion gating unit, and a quality evaluation branch. The feature extraction unit extracts a first feature map; the substructure detection branch predicts a first structure detection result based on the first feature map; the feature fusion gating unit calculates first feature gating coefficients and weights the first region of interest feature map based on the first feature gating coefficients to output a first weighted feature map; and the quality evaluation branch predicts a first image quality score based on the first weighted feature map.

[0026] In some examples, the target prediction model also includes a trained cross-sectional contour classification branch. This branch is used to predict detection boxes for cross-sectional contours based on the first feature map, as well as the probability distribution of the first cross-sectional contour category corresponding to those detection boxes.

[0027] Figure 6 This illustrates the model structure used to demonstrate the target prediction model. For example... Figure 6 As shown, the feature extraction unit of the target prediction model includes a backbone network, a multi-head self-attention (MHSA) module, and a neck network (i.e., a feature pyramid fusion module). A MHSA module is cascaded at the end of the backbone network, enabling it to perform well even when dealing with long, curved echo bands in ultrasound images (corresponding to structures like arms and legs), extracting more useful features and reducing the impact of background noise. After multi-layer feature convolution based on the backbone network and MHSA module, the first feature map is output through the neck network.

[0028] Please continue reading. Figure 6 The first feature map can be input into two branches: the substructure detection branch and the section contour classification branch. The substructure detection branch predicts the first structure detection result based on the first feature map, including the first detection box of the sub-limb structure, the first region of interest feature map indicated by the first detection box, the first confidence score of the first detection box, and the probability distribution of the first structure category corresponding to the first detection box. The section contour classification branch is used to predict the detection box of the section contour and the probability distribution of the first section contour category.

[0029] The feature fusion gating unit outputs a first weighted feature map based on the input first region of interest (ROI) feature map, first confidence level, the maximum probability value in the first structure class probability distribution (i.e., maximum structure class probability), and the maximum probability value in the first section contour class probability distribution (i.e., maximum section contour class probability). For specific processing of the output first weighted feature map, please refer to the above text. Figure 5 The relevant descriptions will not be repeated here.

[0030] Furthermore, the first weighted feature map can be input into the quality assessment branch to output the first image quality score of the sublimb structure. In addition, based on the first detection box of the sublimb structure and the probability distribution of the first structure category corresponding to the first detection box in the first structure detection result of the substructure detection branch, the first detection box and structure category of the sublimb structure can be output. Furthermore, based on the detection box of the cross-sectional contour predicted by the cross-sectional contour classification branch and the probability distribution of the first cross-sectional contour category corresponding to the detection box, the detection box and contour category of the cross-sectional contour can be output.

[0031] S12, Based on the detected sub-limb structures, determine the sub-section type corresponding to the limb section ultrasound image under the limb section type.

[0032] Specifically, the detected sub-limb structures can be matched with the medical classification standard rules corresponding to the sub-section types under the limb section type to obtain the sub-section type corresponding to the limb section ultrasound image.

[0033] In some embodiments, for each sub-section type under a limb-type section, if the structural categories and numbers of the detected sub-limb structures match the core structural categories and numbers under that sub-section type, then the limb section ultrasound image is determined to correspond to that sub-section type. The core structural categories under a sub-section type refer to important and representative sub-limb structures that characterize that sub-section type. There is at least one core structural category under a sub-section type.

[0034] In some embodiments, if the first image quality score of the detected sub-limb structure is lower than a preset score threshold (e.g., lower than 0.25), the sub-limb structure is considered invalid. Subsequently, when determining the sub-section type, the invalid sub-limb structure is ignored, and the corresponding sub-section type is determined only based on the valid sub-limb structures (whose first image quality score is higher than the preset score threshold).

[0035] For example, the limb-type section type includes eight sub-section types, namely forearm long axis section, upper arm long axis section, upper limb long axis section, both upper limb long axis section, lower leg long axis section, thigh long axis section, lower limb long axis section, and both lower limb long axis section.

[0036] For example, the sub-section type corresponding to the limb section ultrasound image can be determined according to the following medical classification standard rules corresponding to each sub-section type.

[0037] 1. If the sub-limb structures detected in the limb section ultrasound image include core upper limb structures such as two forearms, two upper arms, two hands, two wrist joints, and two elbow joints, the sub-section type is determined to be a bi-upper limb long axis section.

[0038] 2. If the detected sub-limb structures only contain a single core upper limb structure such as a forearm, upper arm, hand, wrist joint, or elbow joint, its sub-section type is determined to be an upper limb long axis section.

[0039] 3. If the detected sub-limb structure only has a single upper arm, elbow joint or other core structures related to the upper arm, its sub-section type is determined to be the upper arm long axis section.

[0040] 4. If the detected sub-limb structure only contains a single forearm, wrist joint, elbow joint, hand, or other core structures related to the forearm, its sub-section type is determined to be the forearm long axis section.

[0041] 5. If the detected sub-limb structures include two thighs, two calves, two knee joints, two ankle joints, and two feet, the sub-section type is determined to be a bi-lower limb long axis section.

[0042] 6. If the detected sub-limb structure only contains a single core lower limb structure such as the thigh, calf, knee joint, ankle joint, or foot, its sub-section type is determined to be the long axis section of the lower limb.

[0043] 7. If the detected sub-limb structure only contains a single thigh, knee joint or other core structures related to the thigh, its sub-section type is determined to be the long axis section of the thigh.

[0044] 8. If the detected sub-limb structure only contains a single lower leg, knee joint, ankle joint, foot, or other core structures related to the lower leg, its sub-section type is determined to be the long axis section of the lower leg.

[0045] S13, for each limb sectional ultrasound image, determine the image quality score of the limb sectional ultrasound image based on the first image quality score of each sub-limb structure in the limb sectional ultrasound image.

[0046] Specifically, for each limb section ultrasound image, the first image quality scores of each sub-limb structure in the limb section ultrasound image are weighted and summed to obtain the image detail score; the core sub-limb structure in each sub-limb structure is determined; the baseline quality score is obtained based on the matching of the determined core sub-limb structure with the core structure category under the corresponding sub-section type of the limb section ultrasound image; and the image quality score of the limb section ultrasound image is determined based on the baseline quality score and the image detail score.

[0047] For example, the first image quality scores of each sub-limb structure can be weighted and summed to obtain the image detail score, or the weighted sum can be added to the adjustment coefficient to obtain the image detail score.

[0048] In some embodiments, when the sub-section type corresponding to the limb sectional ultrasound image is the first sub-section type, the core sub-limb structure in each sub-limb structure and subsequent steps are performed to determine the baseline quality score, and the image quality score of the limb sectional ultrasound image is determined based on the baseline quality score and the image detail score.

[0049] In some examples, for the first sub-section type, the first image quality score of each sub-limb structure in the limb section ultrasound image and the image quality score of the section contour in the limb section ultrasound image can be weighted and summed to obtain the image detail score.

[0050] For example, the target prediction model (e.g., the section contour classification branch in the target detection model) can predict the detection box for each section contour in the limb section ultrasound image based on the first feature map, and the first section contour category probability distribution corresponding to the detection box; the contour category indicated by the maximum probability value in the first section contour category probability distribution is taken as the final contour category of the section contour indicated by the detection box. Furthermore, the image quality score of the section contour can be quantified based on the maximum probability value in the first section contour category probability distribution. For example, assuming that a detection box for a section contour is output for a limb section ultrasound image, and the maximum probability value of its corresponding first section contour category is 90%, indicating a 90% probability that the section contour is considered to be the overall contour of both upper limbs, then the image quality score of this portion of the image content corresponding to the overall contour of both upper limbs can be 0.9.

[0051] When the sub-section type corresponding to the limb sectional ultrasound image is the second sub-section type, the image quality score of the limb sectional ultrasound image can be directly determined based on the image detail score. In some examples, the image quality score is determined based on the ratio of the image detail score to the total weight used in calculating the image detail score. For example, image quality score = image detail score ÷ total weight × 100. Optionally, to avoid the total score exceeding 100, if the image detail score is greater than the total weight, the image detail score will be adjusted to a value close to but less than the total weight according to a preset rule, and the image quality score will be calculated based on the ratio of the adjusted image detail score to the total weight. In other examples, the image detail score can be directly used as the image quality score.

[0052] The second sub-section type has more core structure categories than the first sub-section type. For example, the first sub-section type includes long-axis sections of both upper limbs and / or long-axis sections of both lower limbs, while the second sub-section type includes long-axis sections of the forearm, upper arm, upper limb, lower leg, thigh, and lower limb.

[0053] To facilitate understanding, the following examples illustrate the calculation strategy for image quality scores for each sub-section type.

[0054] A1. Long axis section of both upper limbs (a1-1) Basic quality score: allocated based on the detection status of core structures: 1. Hand + Forearm + Upper Arm: 75 points (i.e., if the hand, forearm, and upper arm are detected in the ultrasound image of the limb section, the baseline mass score is assigned a preset score of 75 points). 2. Forearm + Upper Arm: 55 points; 3. Other situations: 35 points; (a1-2) Image detail score = Image quality score of overall outline × 4.0 (i.e. weight) + Image quality score of upper arm × 1.0 (i.e. weight) + Image quality score of forearm × 2.5 (i.e. weight) + Image quality score of elbow joint × 1.0 (i.e. weight) + Image quality score of wrist joint × 1.0 (i.e. weight) + Image quality score of hand × 2.5 (i.e. weight); (a1-3) Image quality score = base quality score + image detail score.

[0055] A2. Long axis section of the upper limb (a2-1) Image detail score = image quality score of upper arm × 5 (i.e. weight) + image quality score of forearm × 5 (i.e. weight) + image quality score of hand × 5 (i.e. weight) + image quality score of elbow joint × 1 (i.e. weight) + image quality score of wrist joint × 1 (i.e. weight); thus, the total weight is 17. (a2-2) Image quality score = Image detail score ÷ Total weight 17 × 100; For example, if the image detail score (i.e., the weighted summation score) is ≥17, the image detail score is directly adjusted to 16.5 and substituted into the image quality score calculation formula in (a2-2). If the image detail score is <17, the image detail score is directly substituted into the above formula to calculate the image quality score. This scheme ensures the rationality of the scoring through score range constraints, that is, it can avoid the image quality score from exceeding 100 points. It should be understood that this scheme is also applicable to constraining the rationality of image quality scores for other sub-section types (e.g., subordinate A3 to A4, B2 to B4), which will not be elaborated further below.

[0056] A3. Long axis section of the forearm (a3-1) Image detail score = Image quality score of forearm × 5 (i.e. weight) + Image quality score of hand × 5 (i.e. weight) + Image quality score of other structures × 2, so the total weight is 14; (a3-2) Image quality score = Image detail score ÷ Total weight 14 × 100; A4. Long axis section of the upper arm (a4-1) Image detail score = upper arm image quality score × 6 (i.e. weight) + elbow joint image quality score × 4 (i.e. weight), so the total weight is 10; (a4-2) Final image quality score = Image detail score ÷ Total weight 10 × 100; B1. Long-axis section of both lower limbs: (b1-1) Basic quality score: allocated based on the detection status of core structures: 1. Foot + calf + thigh: 75 points 2. Calf + Thigh: 55 points 3. Other situations: 35 points (b1-2) Image detail score: Image quality score of overall outline × 4.0 (i.e., weight) + Image quality score of lower leg × 1.0 (i.e., weight) + Image quality score of thigh × 2.5 (i.e., weight) + Image quality score of knee joint × 1.0 (i.e., weight) + Image quality score of ankle joint × 1.0 (i.e., weight) + Image quality score of foot × 2.5 (i.e., weight); (b1-3) Image quality score = base quality score + image detail score; B2. Long-axis section of the lower limb (including thigh and calf) Image quality score = 15 (adjustment factor) + 20 (weight) × image quality score of thigh + 20 (weight) × image quality score of calf + Σ(5 + 5 × image quality score of basic structure); where the first 5 in (5 + 5 × image quality score of basic structure) is the adjustment factor corresponding to the basic structure, and the "5" in "5 × image quality score of basic structure" is the weight corresponding to the basic structure. B3. Long axis section of the thigh Image quality score = 46 + 20 (i.e., weight) × image quality score of thigh + 6 + 6 (i.e., weight) × image quality score of basic structure; where 46 and 6 added in the formula are the adjustment coefficients corresponding to the core structure (thigh) and basic structure, respectively.

[0057] B4. Long axis section of the lower leg Image quality score = 45 + 25 (i.e., weight) × image quality score of the calf + 6 + 6 (i.e., weight) × image quality score of the basic structure; where 45 and 6 added in the formula are the adjustment coefficients corresponding to the core structure (calf) and the basic structure, respectively.

[0058] It should be noted that the weight values, adjustment coefficients, and preset scores in the above strategies are all illustrative examples and do not constitute limitations. Furthermore, the score calculation strategies for the second sub-section types of upper limb classes (such as A2 to A4) and lower limb classes (such as B2 to B4) can be the same or different; the above examples are for illustrative purposes only and should not be construed as limiting.

[0059] It should be understood that the above scoring strategy for the A1 bi-upper limb long axis section and the B1 bi-lower limb long axis section uses "structural integrity" as the core dimension of the basic score and "detail clarity" as the bonus / adjustment dimension, which can output a quality score result that is both scientific and practical.

[0060] The above scheme combines the image's sub-section type, the detected key sub-limb structures, and the structural detail quality assessment results (i.e., the image quality score of the sub-limb structure) to output an objective, interpretable score that conforms to medical image interpretation standards, providing a reliable quantitative basis for image quality control (QC).

[0061] S14. Based on the image quality score, select the standard section image under the sub-section type from multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream; the image quality score of the standard section image is higher than that of the non-standard section image.

[0062] It should be understood that if there are multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream, step S14 is executed. If there is only one limb section ultrasound image corresponding to the same sub-section type, it can be directly used as the standard section image under that sub-section type.

[0063] Specifically, for multiple limb section ultrasound images corresponding to the same sub-section type, the limb section ultrasound image with the higher image quality score is selected as the standard section image. For example... Figure 7 and Figure 8 As shown, both images are long-axis sections of the upper limb. Figure 7 The elbow joint was not clearly visible in the long-axis section of the upper limb, but other structures were clear and detectable, thus only meeting the basic standard. Figure 8 The overall structure is clear and complete, therefore the image quality score evaluated for it is higher than that for other images. Figure 7 Higher, meeting the standard, can serve as a higher quality standard cross-sectional image.

[0064] In some examples, at least one standard section image is selected from multiple limb sectional ultrasound images in descending order of image quality score to provide a basis for subsequent diagnostic procedures. For example, the top-K limb sectional ultrasound images with the highest image quality scores are selected as the standard section images, where K is an integer greater than or equal to 1.

[0065] In some examples, if multiple standard cross-sectional images are selected, a short video corresponding to the sub-section type can be generated based on the multiple standard cross-sectional images and stored for doctors to view later.

[0066] The above method innovatively proposes a novel target prediction model. During prenatal testing in early pregnancy, this model can automatically detect key sub-limb structures in limb-sectional ultrasound images frame-by-frame in the ultrasound video stream, and automatically predict the first image quality score for each sub-limb structure. Furthermore, based on the detected key sub-limb structures, it can automatically determine the sub-section type corresponding to the limb-sectional ultrasound image under the limb-sectional type. Based on the first image quality score of each sub-limb structure, combined with a scoring quantification mechanism that follows medical image interpretation standards, the image quality score of the limb-sectional ultrasound image can be objectively and reasonably quantified and evaluated. Then, from multiple limb-sectional ultrasound images corresponding to the same sub-section type, the limb-sectional ultrasound image with the higher image quality score under that sub-section type is selected as the standard section image. In this way, during prenatal testing in early pregnancy, standard limb sections that meet the standard definition of medical images and have high image quality can be automatically obtained in real time from the acquired ultrasound video stream, greatly reducing the doctor's operational costs and improving the efficiency of obtaining standard sections.

[0067] Furthermore, the above method allows for more accurate and reasonable selection of standard sections, reducing the risk of missed sections due to insufficient clinical experience among some doctors. Moreover, in traditional methods, manual operation by doctors may cause them to miss high-quality standard sections in the video stream, retaining sections of relatively low quality, thus affecting subsequent diagnosis and analysis. The above method not only selects standard sections under various important sub-section types in real time but also ensures high-quality images, providing crucial and effective reference information for subsequent diagnostic analysis.

[0068] As mentioned above, the target prediction model is obtained by multiple rounds of iterative training using a set of sample limb cross-sectional ultrasound images. The training process of the target prediction model will be described in detail below.

[0069] like Figure 9 As shown, in one embodiment, the multi-round iterative training steps of the target prediction model include: S91, in each round of iterative training, based on the second feature map of the limb section ultrasound image of each sample, predict the second structure detection result of the sample sub-limb structure in the limb section ultrasound image of the output sample; the second structure detection result includes the second detection box used to locate and mark the sample sub-limb structure, the second confidence of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the feature map of the second region of interest indicated by the second detection box.

[0070] Specifically, limb cross-sectional ultrasound images (including the eight sub-section types mentioned above) from early pregnancy can be selected in advance as sample ultrasound images. Label design, data annotation, and other work can be completed to construct training sets, validation sets, and test sets for deep learning.

[0071] In some embodiments, the ultrasound images of the sample limb cross sections are annotated with sub-limb structure annotation boxes, and the sub-limb structure annotation boxes are annotated with structural classification labels and image quality scores; the annotated image quality scores are used to characterize the quality of the image content within the sub-limb structure annotation boxes.

[0072] For example, structural classification labels include: upper arm, elbow joint, forearm, wrist joint, hand, thigh, knee joint, lower leg, ankle joint, foot, etc. It should be understood that any limb structure that conforms to the medical definition requires a structural classification label.

[0073] In some embodiments, each round of the prediction model to be trained includes a feature extraction unit, a substructure detection branch, a feature fusion gating unit, and a quality assessment branch. The feature extraction unit is used to extract features from the ultrasound images of the sample limb sections to obtain a second feature map. The substructure detection branch is used to predict the second structure detection result of the sample sublimb structure. The feature fusion gating unit is used to execute step S92 to output a second weighted feature map. The quality assessment branch is used to execute step S93 to output a second image quality score for the sample sublimb structure.

[0074] S92, calculate the second feature gating coefficient based on the second confidence level of the second detection box and the maximum probability value in the probability distribution of the second structure category, and weight the feature map of the second region of interest based on the second feature gating coefficient to obtain the second weighted feature map.

[0075] In some embodiments, during each round of training iterations, the probability distribution of the second section contour category can also be predicted based on the second feature map of the limb section ultrasound image of each sample. Furthermore, the second feature gating coefficient is calculated based on the second confidence level of the second detection box, the maximum probability value in the second structure category probability distribution, and the maximum probability value in the second section contour category probability distribution.

[0076] S93, predict the second image quality score of the sample sub-limb structure marked by the second detection box based on the second weighted feature map.

[0077] S94. Calculate the target loss based on the second detection box of the sample sub-limb structure, the second confidence level of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the second image quality score of the sample sub-limb structure.

[0078] It should be understood that the target loss is calculated based on the difference between the second detection box, the second confidence level of the second detection box, the probability distribution of the second structural category corresponding to the second detection box, and the second image quality score of the sample sub-limb structure, and their respective corresponding true annotation values.

[0079] Specifically, the structure detection loss is determined. The structure detection loss is determined based on the bounding box regression loss, confidence loss, and class probability loss.

[0080] The bounding box regression loss represents the difference between the predicted second detection box and the corresponding sub-limb structure annotation box. For example, the bounding box regression loss can be determined based on the difference in at least one dimension between the second detection box and the sub-limb structure annotation box, such as the overlap area, center point distance, and aspect ratio.

[0081] The confidence loss is calculated based on the second confidence score of the second detection box and the corresponding sub-limb structure annotation box. For example, a binary cross-entropy loss can be used; the true value is 1 for an annotated sub-limb structure annotation box and 0 for an unannotated sub-limb structure annotation box. The second confidence score of the predicted second detection box represents the probability of the presence of a sub-limb structure within the predicted second detection box. The larger the difference between the second confidence score and the true value, the greater the confidence loss; conversely, the smaller the difference, the smaller the confidence loss.

[0082] The category probability loss characterizes the difference between the probability distribution of the second structural category corresponding to the second detection box and the structural classification label annotated for the corresponding sublimb structure. It should be understood that the second structural category probability distribution is essentially a conditional probability distribution. That is, if the output of the second detection box confirms the existence of a certain sublimb structure, then, under this premise, the probability that the sublimb structure belongs to a specific structural category is predicted, thus obtaining the second structural category probability distribution.

[0083] Furthermore, the quality assessment loss is calculated based on the difference between the second image quality score of the sample sub-limb structure marked by the second detection box and the image quality score marked by the corresponding sub-limb structure annotation box.

[0084] In some examples, for the second detection box of each sample sub-limb structure, the intersection-over-union ratio (IoU) between the second detection box and the corresponding sub-limb structure annotation box is calculated.

[0085] If the intersection-over-union ratio (IoU) is greater than a preset threshold (e.g., IoU > 0.5), and the predicted probability distribution of the second structure category matches the corresponding structure classification label (i.e., the structure category prediction is correct), then the quality assessment loss is calculated by comparing the second image quality score of the sample sub-limb structure indicated by the second detection box with the image quality score annotated by the corresponding sub-limb structure annotation box. For example, the mean squared error loss function can be used to calculate the quality assessment loss. The quality assessment loss calculated in this way can also be called the annotation score constraint loss, used to supervise the model's prediction score fitting the labeled image quality score. For example, the mask value corresponding to the second detection box can be assigned a value of 1, i.e., the label mask value M = 1. For the sample sub-limb structure located by the second detection box with a mask value of 1, the relevant steps of the annotation score constraint loss described above are performed.

[0086] If the intersection-union ratio (IU) is less than or equal to a preset threshold, or if the predicted probability distribution of the second structure category does not match the corresponding structure classification label (i.e., the structure category prediction is inaccurate), then the quality assessment loss is determined based on the difference between the second image quality score of the sample sub-limb structure indicated by the second detection box and zero. The quality assessment loss calculated in this way can also be called the zero-value suppression loss, used to supervise the model's predicted score as it approaches 0. For example, the mask value corresponding to the second detection box can be assigned to 0, i.e., the mask value M=0. For the sample sub-limb structure located by the second detection box with a mask value of 0, the steps for calculating the zero-value suppression loss are performed.

[0087] It should be understood that using the location information (detection box) of the substructure detection branch to determine the spatial mask can effectively filter out the interference of background noise in ultrasound images on quality assessment.

[0088] Furthermore, the target loss is calculated based on the structural inspection loss and the quality assessment loss.

[0089] In some embodiments, the limb cross-sectional ultrasound images of the samples are also labeled with cross-sectional contour classification labels. These labels are used to classify limb contours that are larger than the sublimb structures within the cross-section. For example, the cross-sectional contour classification labels include: overall contour of both upper limbs, overall contour of the upper limbs, overall contour of both lower limbs, and overall contour of the lower limbs. It should be understood that the cross-sectional contour classification labels are primarily determined based on the general features of the image. For instance, if the head contour is clearly visible or there are clear hand features, it is labeled as an overall contour of both upper limbs or an overall contour of the upper limbs. If the leg structure is clearly visible or there is a feature of the legs together, it is labeled as an overall contour of both lower limbs or an overall contour of the lower limbs.

[0090] Therefore, the target loss can be calculated based on structure detection loss, facet contour classification loss, and quality assessment loss. The facet contour classification loss characterizes the difference between the probability distribution of the second facet contour category and the corresponding facet contour classification label. For example, the cross-entropy loss function can be used to calculate the facet contour classification loss. The facet contour classification loss supervises the model's learning of global image features and is used to consider the model's facet contour classification judgments.

[0091] In some embodiments, the prediction model to be trained in each round further includes a cross-sectional contour classification branch to be trained. The cross-sectional contour classification branch to be trained is used to output the detection boxes of the cross-sectional contours detected in this round, as well as the second cross-sectional contour category probability distribution for each detection box / cross-sectional contour. For example, the maximum probability value in the second cross-sectional contour category probability distribution can be used as the contour category of the corresponding cross-sectional contour.

[0092] S95, adjust the parameters of the prediction model to be trained in this round based on the target loss, and carry out the next round of iteration training until the iteration stops, and obtain the trained target prediction model.

[0093] Specifically, in each round of training iteration, the parameters of the feature extraction unit, substructure detection branch, cross-section contour classification branch, feature fusion gating unit, and quality assessment branch in the prediction model to be trained can be adjusted based on the target loss calculated in this round. The prediction model with adjusted parameters in this round is then used as the prediction model to be trained in the next round of training iteration. This iterative training continues until the number of iterations meets a preset threshold or the prediction model meets a preset convergence condition (such as the target loss in this round being less than a preset threshold), at which point the iteration stops, and the trained target prediction model is obtained.

[0094] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0095] Based on the same inventive concept, this application also provides a device for acquiring standard limb cross-sectional images via prenatal ultrasound during early pregnancy, as described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for acquiring standard limb cross-sectional images via prenatal ultrasound during early pregnancy provided below can be found in the limitations of the method for acquiring standard limb cross-sectional images via prenatal ultrasound during early pregnancy described above, and will not be repeated here.

[0096] In one embodiment, a device for acquiring standard limb cross-sectional images via prenatal ultrasound during early pregnancy is provided. This device is used to perform the method for acquiring standard limb cross-sectional images via prenatal ultrasound during early pregnancy as described in this application embodiment. The device includes: The model prediction module is used to input the limb cross-sectional ultrasound images from the real-time acquired ultrasound video stream into the target prediction model during prenatal testing in early pregnancy, in order to detect the structure of each sub-limb in the limb cross-sectional ultrasound image and predict the first image quality score of each sub-limb structure. The sub-section type identification module is used to determine the sub-section type corresponding to the limb section ultrasound image under the limb section type based on the detected sub-limb structures; The image quality assessment module is used to determine the image quality score of each limb sectional ultrasound image based on the first image quality score of each sub-limb structure in the limb sectional ultrasound image. The standard section selection module is used to select a standard section image of a sub-section type from multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream based on the image quality score; wherein, the image quality score of the standard section image is higher than that of the non-standard section image.

[0097] In some embodiments, the target prediction model is obtained through multiple rounds of iterative training using a set of sample limb cross-sectional ultrasound images. The device further includes: The model training module is used to predict the second structure detection result of the sample sub-limb structure in the ultrasound image of the sample limb cross-section in each iteration training, based on the second feature map of each sample limb cross-section ultrasound image. The second structure detection result includes the second detection box used to locate and mark the sample sub-limb structure, the second confidence of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the feature map of the second region of interest indicated by the second detection box. The second feature gating coefficient is calculated based on the second confidence of the second detection box and the maximum probability value in the probability distribution of the second structure category. The second region of interest feature map is weighted based on the second feature gating coefficient to obtain the second weighted feature map. The second image quality score of the sample sub-limb structure marked by the second detection box is predicted based on the second weighted feature map. The target loss is calculated based on the second detection box of the sample sub-limb structure, the second confidence of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the second image quality score of the sample sub-limb structure. The parameters of the prediction model to be trained in this round are adjusted based on the target loss, and the next iteration training is carried out until the iteration stops, and the trained target prediction model is obtained.

[0098] The modules in the aforementioned prenatal ultrasound limb standard section image acquisition device for early pregnancy can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0099] In one embodiment, a computer device is provided, the internal structure diagram of which can be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for acquiring standard limb cross-sectional images via prenatal ultrasound in early pregnancy.

[0100] Those skilled in the art will understand that Figure 10The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0101] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the embodiments of this application.

[0102] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the embodiments of this application.

[0103] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the embodiments of this application.

[0104] It should be noted that the user information (including but not limited to user device information, user attribute content, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0105] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited thereto.

[0106] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The above embodiments are merely illustrative of several implementation methods of this application and should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for acquiring standard cross-sectional images of the limbs via prenatal ultrasound in early pregnancy, characterized in that, The method includes: During prenatal testing in early pregnancy, the ultrasound images of the limbs in the real-time ultrasound video stream are input into the target prediction model to detect the structure of each sub-limb in the ultrasound images of the limbs and predict the first image quality score of each sub-limb structure. Based on the detected sub-limb structures, determine the sub-section type corresponding to the limb section ultrasound image under the limb section type; For each limb section ultrasound image, an image quality score is determined based on a first image quality score of each of the sub-limb structures in the limb section ultrasound image; Based on the image quality score, a standard section image under the same sub-section type is selected from multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream; wherein, the image quality score of the standard section image is higher than that of the non-standard section image.

2. The method according to claim 1, characterized in that, The step of inputting the real-time acquired ultrasound video stream of limb cross-sectional ultrasound images into the target prediction model to detect the structures of each sub-limb in the limb cross-sectional ultrasound images and predict the first image quality score of each sub-limb structure includes: The limb section ultrasound images are input into the target prediction model to extract the first feature map of the limb section ultrasound images; Based on the first feature map, the first structural detection result of the sub-limb structure in the limb section ultrasound image is predicted and output; the first structural detection result includes a first detection box for locating and marking the sub-limb structure, a first confidence level of the first detection box, a first structural category probability distribution corresponding to the first detection box, and a first region of interest feature map indicated by the first detection box. The first feature gating coefficient is calculated based on the first confidence level and the maximum probability value in the probability distribution of the first structure category. The first feature gating coefficient is then used to weight the feature map of the first region of interest to obtain the first weighted feature map. Based on the first weighted feature map, a first image quality score is predicted for the sub-limb structure marked by the first detection box.

3. The method according to claim 2, characterized in that, The method further includes: Based on the first feature map, predict and output the probability distribution of the first section contour category in the ultrasound image of the limb section; The step of calculating the first feature gating coefficient based on the first confidence level and the maximum probability value in the first structure category probability distribution includes: The first feature gating coefficient is calculated based on the first confidence level, the maximum probability value in the probability distribution of the first structure category, and the maximum probability value in the probability distribution of the first cross-sectional contour category.

4. The method according to claim 1, characterized in that, The target prediction model is obtained by multiple rounds of iterative training using a set of sample limb cross-sectional ultrasound images; The method further includes: In each round of iterative training, based on the second feature map of the limb section ultrasound image of each sample, the second structure detection result of the sample sub-limb structure in the limb section ultrasound image of the sample is predicted and output; the second structure detection result includes a second detection box for locating and marking the sample sub-limb structure, the second confidence of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the second region of interest feature map indicated by the second detection box. The second feature gating coefficient is calculated based on the second confidence level of the second detection box and the maximum probability value in the probability distribution of the second structure category. The second feature gating coefficient is then used to weight the feature map of the second region of interest to obtain the second weighted feature map. A second image quality score is predicted based on the second weighted feature map to represent the sub-limb structure of the sample indicated by the second detection box. Based on the second detection box of the sample sub-limb structure, the second confidence of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the second image quality score of the sample sub-limb structure, the target loss is calculated. The parameters of the prediction model to be trained in this round are adjusted based on the target loss, and the next round of iterative training is carried out until the iteration stops, so as to obtain the target prediction model after training.

5. The method according to claim 4, characterized in that, The ultrasound images of the sample limb cross sections are labeled with sub-limb structure annotation boxes. Each sub-limb structure annotation box is labeled with a structural classification label and an image quality score. The labeled image quality score is used to characterize the quality of the image content within the sub-limb structure annotation box. The target loss is calculated based on the second detection box of the sample sub-limb structure, the second confidence level of the second detection box, the probability distribution of the second structure category corresponding to the second detection box, and the second image quality score of the sample sub-limb structure, including: The structure detection loss is determined based on bounding box regression loss, confidence loss, and category probability loss. The bounding box regression loss represents the difference between the predicted second detection box and the corresponding sub-limb structure annotation box. The confidence loss is calculated based on the second confidence of the second detection box and the corresponding sub-limb structure annotation box. The category probability loss represents the difference between the second structure category probability distribution corresponding to the second detection box and the structure classification label annotated for the corresponding sub-limb structure annotation box. The quality assessment loss is calculated based on the difference between the second image quality score of the sample sub-limb structure marked by the second detection box and the image quality score marked by the corresponding sub-limb structure annotation box. The target loss is calculated based on the structural detection loss and the quality assessment loss.

6. The method according to claim 5, characterized in that, The ultrasound images of the limb sections of the sample also have section contour classification labels marked on them; In each iteration of training, the method further includes: Predict the probability distribution of the second section contour category based on the second feature map of the limb section ultrasound image of each sample; The step of calculating the second feature gating coefficient based on the second confidence level of the second detection box and the maximum probability value in the probability distribution of the second structure category includes: The second feature gating coefficient is calculated based on the second confidence level of the second detection box, the maximum probability value in the probability distribution of the second structure category, and the maximum probability value in the probability distribution of the second section contour category. The calculation of the target loss based on the structural detection loss and the quality assessment loss includes: Based on the difference between the probability distribution of the second section contour category and the corresponding section contour classification label, the section contour classification loss is calculated. The target loss is calculated based on the structure detection loss, the cross-sectional contour classification loss, and the quality assessment loss.

7. The method according to claim 5, characterized in that, The method further includes: For each sample sub-limb structure, the intersection-union ratio (IUU) between the second detection box and the corresponding sub-limb structure annotation box is calculated. If the intersection-union ratio is greater than a preset threshold, and the predicted probability distribution of the second structure category matches the corresponding structure classification label, then the difference between the second image quality score based on the sample sub-limb structure indicated by the second detection box and the image quality score annotated for the corresponding sub-limb structure annotation box is calculated to determine the quality assessment loss. If the crossover ratio is less than or equal to a preset threshold or the predicted probability distribution of the second structure category does not match the corresponding structure classification label, then the quality assessment loss is determined based on the difference between the second image quality score of the sample sub-limb structure marked by the second detection box and the zero value.

8. The method according to any one of claims 1 to 7, characterized in that, For each limb sectional ultrasound image, based on a first image quality score for each of the sub-limb structures in the limb sectional ultrasound image, the image quality score of the limb sectional ultrasound image is determined, including: The first image quality scores of each sub-limb structure in the ultrasound images of the limb cross sections are weighted and summed to obtain the image detail score; Identify the core sub-limb structure in each of the aforementioned sub-limb structures; The basic quality score is obtained based on the matching between the determined core sub-limb structures and the core structure categories under the sub-section types corresponding to the ultrasound images of the limb sections; Based on the baseline quality score and the image detail score, the image quality score of the limb section ultrasound image is determined.

9. The method according to claim 8, characterized in that, The method further includes: If the sub-section type corresponding to the ultrasound image of the limb section is the first sub-section type, then perform the steps of determining the core sub-limb structure in each of the sub-limb structures and subsequent steps. When the sub-section type corresponding to the limb section ultrasound image is the second sub-section type, the image quality score of the limb section ultrasound image is determined based on the image detail score; wherein, the core structure categories under the second sub-section type are more than the core structure categories under the first sub-section type.

10. A device for acquiring standard cross-sectional images of the limbs via prenatal ultrasound during early pregnancy, characterized in that, The device includes: The model prediction module is used to input the limb cross-sectional ultrasound images from the real-time acquired ultrasound video stream into the target prediction model during prenatal testing in early pregnancy, so as to detect the structure of each sub-limb in the limb cross-sectional ultrasound images and predict the first image quality score of each sub-limb structure. The sub-section type identification module is used to determine the sub-section type corresponding to the ultrasound image of the limb section under the limb section type based on the detected sub-limb structures; An image quality assessment module is used to determine the image quality score of each limb section ultrasound image based on a first image quality score of each sub-limb structure in the limb section ultrasound image. The standard section selection module is used to select a standard section image of the same sub-section type from multiple limb section ultrasound images corresponding to the same sub-section type in the ultrasound video stream based on the image quality score; wherein the image quality score of the standard section image is higher than that of the non-standard section image.

Images