A hip joint ultrasound standard surface screening and measuring method and device based on key point detection and timing consistency

Abstract

The application relates to a hip joint ultrasound standard surface screening and measuring method and device based on key point detection and time sequence consistency, and the method comprises the following steps: acquiring a hip joint ultrasound image sequence; detecting a plurality of anatomical key points in the ultrasound image, outputting a multi-channel response heat map, and obtaining anatomical key point coordinates; constructing a comprehensive consistency scoring function based on the anatomical key points and the time sequence change of adjacent frames of the hip joint ultrasound image, and calculating a consistency score; when the consistency score exceeds a preset threshold, the corresponding frame is determined as a candidate standard section frame; the development state of the candidate standard section frame is reviewed and distinguished to obtain a development state classification result; the alpha angle and the beta angle are calculated, the angle line and the key points are superimposed and displayed on an original drawing, a visual result is generated, and a structured report containing the visual result, the anatomical key point coordinates, the angle parameters and the development state classification result is output. Compared with the prior art, the application significantly improves the automation degree and the result interpretability while maintaining high-precision measurement.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical image analysis technology, and in particular to a method and device for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency. Background Technology

[0002] Developmental dysplasia of the hip (DDH) is a common developmental disorder of the hip joint in infants. Early diagnosis and treatment are crucial to preventing subsequent severe hip dysfunction. Clinically, early diagnosis of infantile DDH usually relies on ultrasound imaging, and the acetabular angles (α angle and β angle) are measured using the Graf method to assess acetabular development. However, the current procedure mainly relies on physicians manually selecting standard cross-sectional frames from the ultrasound video and measuring the angles, which has the following problems: (1) The operation is highly subjective and has poor repeatability. Differences in physician experience lead to low consistency in measurement results; (2) Low efficiency. Manually selecting standard sections frame by frame in ultrasound video is time-consuming and laborious; (3) Insufficient algorithms. Most existing deep learning methods are single-frame classification or segmentation methods, lacking geometric structure constraints and temporal consistency analysis, making it difficult to stably identify standard cross-sections; (4) Insufficient interpretability. Although the end-to-end discriminant network can output classification results, it lacks key structural annotations and angular information, making it impossible for doctors to intuitively verify the reliability of the results.

[0003] Chinese patent application CN118644450A discloses an automatic measurement method and system for standard ultrasound images of pediatric hip joints. Addressing the challenges of complex ultrasound image detection models, limited training data, difficulty in achieving high robustness on large-scale images, and limited practical application in clinical practice, this method utilizes an automatic measurement and classification model based on standard ultrasound images. By simultaneously obtaining the segmentation and annotation of N anatomical structures in standard ultrasound image samples and the detection results of corresponding M key points, it provides doctors with more accurate diagnostic information, avoiding misdiagnosis or missed diagnosis. While this method can simulate clinical practice and reduce measurement discrepancies, possesses an end-to-end deep learning network with good consistency and robustness, and improves training performance while reducing training sample data, it is inefficient in multi-scale information fusion and lightweight inference, and requires high configuration costs.

[0004] Some studies have attempted to automate the screening and measurement of ultrasound images using deep neural networks, but most methods still fall short in terms of accuracy, timeliness, and interpretability. On the one hand, while the traditional U-Net structure has strong feature extraction capabilities, it is inefficient in multi-scale information fusion and lightweight inference. On the other hand, the model often ignores the geometric continuity between frames in the ultrasound sequence, resulting in jitter and inconsistencies in the output.

[0005] In summary, current technologies lack an automatic screening and measurement method for hip joint ultrasound images that combines anatomical key point detection, geometric and temporal consistency analysis, and neural network verification mechanisms, which can ensure the stability of standard section recognition and the interpretability of results while guaranteeing detection accuracy. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing infant hip ultrasound diagnosis, such as low efficiency of manual screening, unstable inter-frame recognition, and insufficient interpretability of test results. This invention provides a method and device for screening and measuring standard planes of hip ultrasound that combines anatomical key point detection, geometric-temporal consistency analysis, and convolutional neural network verification, thereby achieving fully automated processing from DICOM raw files to structured measurement reports.

[0007] The objective of this invention can be achieved through the following technical solutions: A method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency, the method comprising: Acquire hip joint ultrasound image sequences, parse metadata, and process the images; The first convolutional neural network is used to detect multiple anatomical key points in the hip joint ultrasound image, and a multi-channel response heatmap with one-to-one correspondence between the channels and the anatomical key points is output. The coordinates of the anatomical key points are obtained by peak search. Based on the geometric relationship of the key anatomical points and the temporal changes of adjacent frames of hip joint ultrasound images, a comprehensive consistency scoring function is constructed to calculate the consistency score. When the consistency score exceeds a preset threshold, the corresponding frame is determined as a candidate standard section frame. The developmental state of the candidate standard section frame is verified and judged by combining the second convolutional neural network with the anatomical key points, and the developmental state classification result and confidence are obtained. Based on the coordinates of the anatomical key points, the bony acetabular angle and cartilaginous acetabular angle are calculated using the Graf method. The angle lines of each angle are superimposed on the anatomical key points and displayed on the original image to generate visualization results. A structured report containing visualization results, anatomical key point coordinates, angle parameters, and developmental status classification results is output.

[0008] Furthermore, the process of processing the hip joint ultrasound image sequence includes: Read DICOM format hip joint ultrasound image sequences and extract metadata, including imaging depth, pixel spacing, and field of view location parameters; Based on the aforementioned metadata, the hip joint ultrasound image is scaled and cropped to obtain a standardized analysis area. The cropped ultrasound images of the hip joint are subjected to brightness, contrast, and size normalization to adapt to the input requirements of the subsequent neural network.

[0009] Furthermore, after outputting the multi-channel response heatmap, the heatmap is subjected to softmax or maximum value normalization processing. The maximum response intensity of the multi-channel response heatmap is used as the location reliability index. The reliability weight of each anatomical key point is adjusted according to the location reliability index, and the consistency score is calculated using the comprehensive consistency scoring function.

[0010] Furthermore, the first convolutional neural network is an ITUNet network that integrates Involution operation and Transformer self-attention mechanism. The loss function of the ITUNet network includes heatmap regression loss and a geometric consistency constraint term that applies soft constraints to the geometric relationship between the distance between anatomical key points and the baseline angle.

[0011] Furthermore, during the pre-training stage of the first convolutional neural network, a detachable upsampling module is set on the decoding path of the ITUNet network. The detachable upsampling module includes a lightweight convolutional or small window attention structure, which enhances gradient propagation and boundary detail learning, performs local super-resolution detail restoration on the feature map, strengthens the alignment of heatmap peaks and edge sensitivity, and reduces the localization deviation of dissecting key points. During the actual inference stage of the first convolutional neural network, the detachable upsampling module is removed or replaced with an identity mapping, and only the standard upsampling of the backbone path is retained for lightweight inference.

[0012] Furthermore, the process of constructing the comprehensive consistency scoring function includes: Connect the key anatomical points to obtain reference baselines, and calculate the angles between each reference baseline and the horizontal line; Calculate the Euclidean distance between each key anatomical point. If the Euclidean distance exceeds the preset reasonable range of anatomical structure, the current hip joint ultrasound image frame is determined to be an abnormal positioning frame and the current frame is deleted. Based on the coordinates of the key anatomical points, the bony acetabular angle and cartilaginous acetabular angle are calculated using the Graf method. Based on the bony acetabular angle and cartilaginous acetabular angle, the bidirectional change in the angle sequence between consecutive hip joint ultrasound images is calculated. An angle penalty function is constructed based on the angle between each reference baseline and the horizontal line. A structural consistency function is constructed based on the Euclidean distance between each anatomical key point. A bidirectional temporal smoothing function is constructed based on the bidirectional change of the angle sequence. A comprehensive consistency scoring function is constructed by combining the angle penalty function, the structural consistency function, and the bidirectional temporal smoothing function.

[0013] Furthermore, the consistency score is obtained by solving the comprehensive consistency score function through a sliding window, combined with the change information of the forward and backward frames.

[0014] Furthermore, the process of using a second convolutional neural network combined with anatomical key points to perform developmental state verification and discrimination on the candidate standard section frames includes: Receive the candidate standard cross-section frames and extract image features using a second convolutional neural network; Based on the image features, the probability distribution of the two categories of judgment results, namely, normal development and developmental delay, is output. The judgment result corresponding to the maximum probability distribution is used as the verification classification result, and the probability distribution value is used as the classification confidence output.

[0015] Furthermore, the second convolutional neural network introduces channel and spatial attention mechanisms in the feature extraction layer to enhance the network's responsiveness to bony structural boundaries, joint cavities, and bone nucleus regions.

[0016] A device for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency, the device comprising: Image preprocessing module: acquires hip joint ultrasound image sequences, parses metadata, and processes the images; Anatomical Key Point Detection Module: Utilizes a first convolutional neural network to detect multiple anatomical key points in the hip joint ultrasound image, outputs a multi-channel response heatmap with each channel corresponding to one of the anatomical key points, and obtains the coordinates of the anatomical key points through peak search; The temporal screening module constructs a comprehensive consistency scoring function based on the geometric relationship of the anatomical key points and the temporal changes of adjacent frames of hip joint ultrasound images, calculates the consistency score, and determines the corresponding frame as a candidate standard section frame when the consistency score exceeds a preset threshold. Verification module: The second convolutional neural network is used in conjunction with the anatomical key points to verify and judge the developmental status of the candidate standard section frames, and the developmental status classification results and confidence scores are obtained. Output module: Based on the coordinates of the anatomical key points, calculate the bony acetabular angle and cartilaginous acetabular angle using the Graf method, overlay the angle lines of each angle with the anatomical key points on the original image, generate visualization results, and output a structured report containing visualization results, anatomical key point coordinates, angle parameters, and developmental status classification results.

[0017] Compared with the prior art, the beneficial effects of the present invention include: 1. This invention discloses an automatic screening and measurement method for hip joint ultrasound images that combines key point detection, geometric and temporal consistency analysis, and a neural network verification mechanism. It achieves automated screening of standard sections and measurement of developmental status in hip joint ultrasound image sequences, significantly improving automation and interpretability while maintaining high-precision measurement, and has promising clinical application prospects. This invention constructs a comprehensive consistency scoring function based on the geometric relationship of anatomical key points and the temporal changes of adjacent frames in hip joint ultrasound images, calculates a consistency score, and determines the corresponding frame as a candidate standard section frame when the consistency score exceeds a preset threshold. This achieves stable screening of standard section frames, significantly reduces detection jitter, and ensures inter-frame continuity and geometric rationality. Finally, this invention generates visualized results and outputs a structured report containing key point coordinates, angle parameters, and classification conclusions, making the detection results interpretable and easy for doctors to directly verify.

[0018] 2. This invention uses the ITUNet network, which integrates Involution operations and Transformer self-attention mechanisms, for anatomical keypoint detection. Involution operations can achieve local adaptive feature modeling with a lower number of parameters, significantly reducing multiply-accumulate operations and memory usage. The Transformer structure can enhance long-range dependency modeling capabilities and improve feature expression efficiency while maintaining resolution. By integrating Involution operations and attention mechanisms into the U-Net framework and introducing a detachable upsampling module during the training phase, both accuracy and lightweight design are achieved, thereby improving the speed and stability of keypoint detection. Using the ITUNet network for anatomical keypoint detection can more accurately locate anatomical keypoints of the hip joint, providing stable coordinate data for subsequent angle calculation and consistency screening, while also ensuring lightweight inference.

[0019] 3. In this invention, a second convolutional neural network is used in conjunction with anatomical key points to perform developmental state verification and discrimination on candidate standard cross-sectional frames, obtain developmental state classification results and confidence levels, and the verification classification integrates image features and angle data, including an attention mechanism, to form a dual verification of structural localization and semantic classification, thus avoiding misjudgment based on a single indicator.

[0020] 4. This invention also uses the maximum response intensity of the heatmap as a location reliability index for the weighted calculation of consistency scores, so that the consistency scores can more accurately reflect the triple constraints of geometric rationality, temporal stability and location reliability. Key points with reliable location dominate the scoring results, while key points with unreliable location are downweighted, thereby reducing the interference of false location on standard cross-section screening and improving the screening accuracy of candidate frames. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the detection of key anatomical points in this invention. Detailed Implementation

[0022] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] Example 1 This embodiment discloses a method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency. The method is as follows: Figure 1 As shown, steps S1-S5 are included, and each step is described in detail below: Step S1: Obtain hip joint ultrasound image sequences, parse metadata, and process the images.

[0024] The process of processing hip joint ultrasound image sequences includes: Read DICOM format hip joint ultrasound image sequences and extract metadata, including imaging depth, pixel spacing, and field of view location parameters; Based on metadata, the hip joint ultrasound images are scaled and cropped to obtain a standardized analysis area and eliminate irrelevant background interference. The cropped hip ultrasound images were normalized in terms of brightness, contrast, and size to adapt to the input requirements of the subsequent neural network.

[0025] This step achieves a unified scale and contrast standard for images across devices, reducing feature shift caused by differences in input distribution.

[0026] Step S2: Detect multiple anatomical key points in the hip joint ultrasound image using the first convolutional neural network, output a multi-channel response heatmap with one-to-one correspondence between the channels and the anatomical key points, and obtain the coordinates of the anatomical key points through peak search.

[0027] Specifically, such as Figure 2 As shown, this method uses an improved ITUNet (Involution-Transformer-U-Net) structure to output a multi-channel key point response heatmap in hip joint ultrasound images. Each channel corresponds to a key anatomical point of the hip joint (such as the upper edge, lower edge, and nucleus of the acetabulum), and its peak position is used as the coordinate of the key point.

[0028] The first convolutional neural network is the ITUNet network, which integrates Involution operation and Transformer self-attention mechanism. The loss function of the ITUNet network includes heatmap regression loss and geometric consistency constraint term that applies soft constraints to the geometric relationship between the distance between anatomical key points and the baseline angle.

[0029] The ITUNet network, based on the traditional U-Net encoder-decoder architecture, simultaneously introduces Involution (a parameter-adaptive, lightweight local feature transformation) and Transformer self-attention mechanism to balance local texture boundaries and long-range dependencies. This structure effectively enhances feature representation capabilities in infant hip joint ultrasound landmark detection tasks, providing stable key structure localization for Graf-based angle measurements.

[0030] To further improve training stability and deployment efficiency, this invention incorporates a detachable upsampling enhancement module on the ITUNet decoding path, characterized by the following design features: Enabled during training: After regular upsampling (deconvolution or interpolation convolution) at each decoding layer, a branch-based fine-grained upsampling enhancement unit is cascaded, which may contain lightweight convolutions or small window attention structures. This branch enhances gradient propagation and boundary detail learning during training, performing local super-resolution detail restoration on the feature map, thereby strengthening the alignment of heatmap peaks and edge sensitivity, and reducing keypoint localization errors.

[0031] Inference phase bypass: Considering the memory and inference latency constraints of clinical deployment, the aforementioned enhancement branches can be removed or replaced with identity mappings during the inference phase, retaining only the standard upsampling of the main path, thereby achieving lightweight inference without changing the output size and channel consistency.

[0032] In the output stage, the network generates multi-channel heatmaps; preferably, each channel heatmap is subjected to softmax or maximum value normalization processing, and its peak response intensity is used as the key point location reliability index for weighting the subsequent geometric-temporal consistency score.

[0033] To enhance geometric constraints, in addition to the conventional heatmap regression loss, the loss function also includes a geometric consistency constraint term, which applies soft constraints to geometric relationships such as the distance between key points and the baseline angle. This allows the network to model and dissect structural relationships while learning the coordinates of key points, reducing structural errors caused by spurious response peaks.

[0034] Through the aforementioned detachable upsampling and geometric consistency joint training mechanism, this method achieves higher localization accuracy during the training phase and lower GPU memory usage during the inference phase without increasing deployment costs. Combined with ITUNet's lightweight design and global modeling capabilities through its attention mechanism, the network outputs sharper and more stable keypoint heatmaps, providing high-quality input for subsequent angle calculations and verification.

[0035] Step S3: Construct a comprehensive consistency scoring function based on the geometric relationship of anatomical key points and the temporal changes of adjacent frames of hip joint ultrasound images, calculate the consistency score, and when the consistency score exceeds a preset threshold, determine the corresponding frame as a candidate standard section frame.

[0036] The consistency score is obtained by solving the comprehensive consistency score function through a sliding window, combined with the change information of the forward and backward frames.

[0037] The process of constructing the comprehensive consistency scoring function includes: Connect the key anatomical points to obtain reference baselines, and calculate the angles between each reference baseline and the horizontal line. ; Calculate the Euclidean distance between each anatomical key point. If the Euclidean distance exceeds the preset reasonable range of anatomical structure, the current hip joint ultrasound image frame is determined to be an abnormal positioning frame and the current frame is deleted. Based on the coordinates of key anatomical points, the bony and cartilaginous acetabular angles, i.e., the α and β angles, were calculated using the Graf method. Based on these angles, the bidirectional changes in the angle sequence between consecutive hip ultrasound images, i.e., the changes in the α and β angles, were then calculated. and ; Angle penalty function is constructed based on the angle between each reference baseline and the horizontal line; structural consistency function is constructed based on the Euclidean distance between each anatomical key point; bidirectional temporal smoothing function is constructed based on the bidirectional change of the angle sequence; and a comprehensive consistency scoring function is constructed by combining the angle penalty function, the structural consistency function, and the bidirectional temporal smoothing function.

[0038] In Graf ultrasound examination, the α angle primarily reflects the development of the acetabulum and the degree of femoral head coverage. Generally, an α angle greater than 60° is considered to indicate good hip joint stability. Measuring this angle helps physicians determine whether there is a risk of hip dislocation or subluxation, and whether the acetabulum is fully developed.

[0039] The beta angle is primarily used to assess the relationship between the soft acetabular roof and the vertical axis, and it is also important for determining hip joint stability. While the specific normal range of the beta angle may vary depending on different studies and standards, it plays a crucial role in assessing hip dysplasia. By measuring the beta angle, physicians can further understand the shape and position of the acetabular roof, thus more accurately assessing the developmental status of the hip joint.

[0040] Specifically, the expression for calculating the bidirectional change in the inter-frame angle sequence of continuous hip joint ultrasound images is as follows: in, For the first α angle in a frame of hip ultrasound image For the first β angle in a frame of ultrasound image of the hip joint.

[0041] The expression for the overall consistency scoring function is: in, , , These are the angle, structure, and time sequence weighting coefficients, respectively. This is the included angle penalty function, used to suppress non-horizontal frames. This is a structural consistency function used to limit the range of variation in distance between key points. It is a bidirectional temporal smoothing function used to suppress sudden angle changes between frames.

[0042] When the score S exceeds the set threshold, the frame is determined to be a candidate standard section. In actual calculation, the system adopts a sliding window strategy (window length 3–5 frames) and calculates a weighted average score by combining information from previous and subsequent frames to suppress occasional jitter and improve the stability of the screening process.

[0043] In another embodiment, after outputting the multi-channel response heatmap in step S2, the heatmap is further subjected to softmax or maximum value normalization. The maximum response intensity of the multi-channel response heatmap is used as the location reliability index. The reliability weight of each anatomical keypoint is adjusted according to the location reliability index, and the consistency score is calculated using a comprehensive consistency scoring function. Specifically, the location reliability index quantifies the location reliability of each anatomical keypoint and adjusts its contribution weight to the geometric-temporal consistency score. The more reliable the location of a keypoint, the greater its impact on the final score; conversely, the corresponding weight is reduced to minimize the interference of location errors.

[0044] Step S4: Use the second convolutional neural network in conjunction with the anatomical key points to perform developmental state verification and discrimination on the candidate standard cross-section frames, and obtain the developmental state classification results and confidence level.

[0045] Specifically, the second convolutional neural network receives candidate frame images and their corresponding key point measurement results. After extracting image features through multi-scale convolution, it fuses them with angle features to form a comprehensive feature vector. After being activated by a fully connected layer and softmax, it outputs the probability distributions of two categories: "normal development" and "poor development".

[0046] The second convolutional neural network is a residual convolutional structure. It introduces channel and spatial attention mechanisms in the feature extraction layer to enhance the network's response to the boundaries of bony structures, joint cavities, and bone nuclei.

[0047] In another embodiment, a Top-K strategy can be used to select several frames with the highest confidence for comprehensive judgment, so as to reduce the risk of misjudgment of a single frame.

[0048] The process of using a second convolutional neural network combined with anatomical key points to perform developmental state verification and discrimination on candidate standard section frames includes: Receive candidate standard cross-section frames and extract image features using a second convolutional neural network; Based on image features, the probability distribution of the two categories of judgment results, namely, normal development and developmental delay, is output. The judgment result corresponding to the maximum probability distribution is used as the verification classification result, and the probability distribution value is used as the classification confidence output.

[0049] This verification mechanism performs semantic-level verification on geometric measurement results, forming a dual constraint of structure and semantics. The verification classification results output by this module, together with the angle measurement values, constitute the final diagnostic output, providing doctors with structured and verifiable automatic analysis conclusions and improving diagnostic accuracy.

[0050] Step S5: Based on the coordinates of the anatomical key points, calculate the bony acetabular angle and cartilaginous acetabular angle using the Graf method. Overlay the angle lines of each angle with the anatomical key points onto the original image to generate a visualization result. Output a structured report containing the visualization result, anatomical key point coordinates, angle parameters, and developmental status classification results.

[0051] In summary, this invention achieves automatic screening of standard sections and measurement of developmental status in hip joint ultrasound images by combining key point detection, geometric-temporal consistency screening, and convolutional neural network verification. This invention significantly improves automation and interpretability of results while maintaining high measurement accuracy, and has promising clinical application prospects.

[0052] Example 2 This embodiment, based on Embodiment 1 above, discloses a hip joint ultrasound standard plane screening and measurement device based on key point detection and temporal consistency. The device includes: Image preprocessing module: acquires hip joint ultrasound image sequences, parses metadata, and processes the images; Anatomical Key Point Detection Module: Utilizes the first convolutional neural network to detect multiple anatomical key points in hip joint ultrasound images, outputs a multi-channel response heatmap with one-to-one correspondence between channels and anatomical key points, and obtains the coordinates of anatomical key points through peak search; Temporal screening module: Based on the geometric relationship of anatomical key points and the temporal changes of adjacent frames of hip joint ultrasound images, a comprehensive consistency scoring function is constructed to calculate the consistency score. When the consistency score exceeds the preset threshold, the corresponding frame is determined as a candidate standard section frame. Verification module: The second convolutional neural network is used in conjunction with anatomical key points to verify and judge the developmental status of candidate standard cross-section frames, and to obtain the developmental status classification results and confidence scores. Output module: Based on the coordinates of anatomical key points, the bony acetabular angle and cartilaginous acetabular angle are calculated using the Graf method. The angle lines of each angle are superimposed on the anatomical key points and displayed on the original image to generate visualization results. The output is a structured report containing visualization results, anatomical key point coordinates, angle parameters and developmental status classification results.

[0053] For details regarding the above modules, please refer to the relevant descriptions and effects in Example 1 for further understanding.

[0054] The device outputs a structured record containing frame indexes, keypoint coordinates, angle values, screening scores, and classification results, which can automatically generate diagnostic reports or be stored in a database. The device supports automatic batch analysis and can display the detection progress and keyframe previews in real time on the user interface. This output format facilitates rapid review and traceability by physicians.

[0055] To adapt to different ultrasound equipment and imaging conditions, the device provides a parameter control interface, which allows for dynamic adjustment of scoring weights via configuration files or a graphical interface. , , The system includes a selection threshold to enable adaptive filtering in different scenarios. Image and keypoint data are exchanged between modules via standardized interfaces. Different network structures (such as lightweight CNN, Transformer, or hybrid architectures) can be replaced as needed to expand system functionality and adapt to performance requirements.

[0056] This design enhances the system's scalability and cross-platform adaptability, enabling deployment on medical workstations, cloud analytics servers, and portable terminals, and supporting rapid integration of algorithm version iterations with clinical deployment.

[0057] Example 3 Based on Embodiment 1, this embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for executing the hip joint ultrasound standard surface screening and measurement method based on key point detection and timing consistency as described above.

[0058] At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the above-mentioned method for screening and measuring standard planes of hip joint ultrasound based on key point detection and timing consistency. Of course, in addition to the software implementation, this invention does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0059] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0060] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for screening and measuring standard ultrasound planes of the hip joint based on key point detection and temporal consistency, characterized in that, The method includes: Acquire hip joint ultrasound image sequences, parse metadata, and process the images; The first convolutional neural network is used to detect multiple anatomical key points in the hip joint ultrasound image, and a multi-channel response heatmap with one-to-one correspondence between the channels and the anatomical key points is output. The coordinates of the anatomical key points are obtained by peak search. Based on the geometric relationship of the key anatomical points and the temporal changes of adjacent frames of hip joint ultrasound images, a comprehensive consistency scoring function is constructed to calculate the consistency score. When the consistency score exceeds a preset threshold, the corresponding frame is determined as a candidate standard section frame. The developmental state of the candidate standard section frame is verified and judged by combining the second convolutional neural network with the anatomical key points, and the developmental state classification result and confidence are obtained. Based on the coordinates of the anatomical key points, the bony acetabular angle and cartilaginous acetabular angle are calculated using the Graf method. The angle lines of each angle are superimposed on the anatomical key points and displayed on the original image to generate visualization results. A structured report containing visualization results, anatomical key point coordinates, angle parameters, and developmental status classification results is output.

2. The method for screening and measuring standard ultrasound planes of the hip joint based on key point detection and temporal consistency according to claim 1, characterized in that, The process of processing the hip joint ultrasound image sequence includes: Read DICOM format hip joint ultrasound image sequences and extract metadata, including imaging depth, pixel spacing, and field of view location parameters; Based on the aforementioned metadata, the hip joint ultrasound image is scaled and cropped to obtain a standardized analysis area. The cropped ultrasound images of the hip joint are subjected to brightness, contrast, and size normalization to adapt to the input requirements of the subsequent neural network.

3. The method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency according to claim 1, characterized in that, After outputting the multi-channel response heatmap, the heatmap is further subjected to softmax or maximum value normalization. The maximum response intensity of the multi-channel response heatmap is used as the location reliability index. The reliability weight of each anatomical key point is adjusted according to the location reliability index, and the consistency score is calculated using the comprehensive consistency scoring function.

4. The method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency according to claim 1, characterized in that, The first convolutional neural network is an ITUNet network that integrates Involution operation and Transformer self-attention mechanism. The loss function of the ITUNet network includes heatmap regression loss and a geometric consistency constraint term that applies soft constraints to the geometric relationship between the distance between anatomical key points and the baseline angle.

5. The method for screening and measuring standard ultrasound planes of the hip joint based on key point detection and temporal consistency according to claim 4, characterized in that, During the pre-training phase of the first convolutional neural network, a detachable upsampling module is set on the decoding path of the ITUNet network. The detachable upsampling module includes a lightweight convolutional or small window attention structure to enhance gradient propagation and boundary detail learning, perform local super-resolution detail restoration on the feature map, strengthen the alignment of heatmap peaks and edge sensitivity, and reduce the localization deviation of dissecting key points. During the actual inference phase of the first convolutional neural network, the detachable upsampling module is removed or replaced with an identity mapping, and only the standard upsampling of the backbone path is retained for lightweight inference.

6. The method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency according to claim 1, characterized in that, The process of constructing the comprehensive consistency scoring function includes: Connect the key anatomical points to obtain reference baselines, and calculate the angles between each reference baseline and the horizontal line; Calculate the Euclidean distance between each key anatomical point. If the Euclidean distance exceeds the preset reasonable range of anatomical structure, the current hip joint ultrasound image frame is determined to be an abnormal positioning frame and the current frame is deleted. Based on the coordinates of the key anatomical points, the bony acetabular angle and cartilaginous acetabular angle are calculated using the Graf method. Based on the bony acetabular angle and cartilaginous acetabular angle, the bidirectional change in the angle sequence between consecutive hip joint ultrasound images is calculated. An angle penalty function is constructed based on the angle between each reference baseline and the horizontal line. A structural consistency function is constructed based on the Euclidean distance between each anatomical key point. A bidirectional temporal smoothing function is constructed based on the bidirectional change of the angle sequence. A comprehensive consistency scoring function is constructed by combining the angle penalty function, the structural consistency function, and the bidirectional temporal smoothing function.

7. The method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency according to claim 1, characterized in that, The consistency score is obtained by solving the comprehensive consistency score function through a sliding window, combined with the change information of the forward and backward frames.

8. The method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency according to claim 1, characterized in that, The process of using a second convolutional neural network combined with anatomical key points to perform developmental state verification and discrimination on the candidate standard section frames includes: Receive the candidate standard cross-section frames and extract image features using a second convolutional neural network; Based on the image features, the probability distribution of the two categories of judgment results, namely, normal development and developmental delay, is output. The judgment result corresponding to the maximum probability distribution is used as the verification classification result, and the probability distribution value is used as the classification confidence output.

9. The method for screening and measuring standard planes of hip joint ultrasound based on key point detection and temporal consistency according to claim 1, characterized in that, The second convolutional neural network introduces channel and spatial attention mechanisms in the feature extraction layer to enhance the network's responsiveness to bony structural boundaries, joint cavities, and bone nucleus regions.

10. A device for screening and measuring standard ultrasonic surfaces of the hip joint based on key point detection and temporal consistency, characterized in that, The device includes: Image preprocessing module: acquires hip joint ultrasound image sequences, parses metadata, and processes the images; Anatomical Key Point Detection Module: Utilizes a first convolutional neural network to detect multiple anatomical key points in the hip joint ultrasound image, outputs a multi-channel response heatmap with each channel corresponding to one of the anatomical key points, and obtains the coordinates of the anatomical key points through peak search; The temporal screening module constructs a comprehensive consistency scoring function based on the geometric relationship of the anatomical key points and the temporal changes of adjacent frames of hip joint ultrasound images, calculates the consistency score, and determines the corresponding frame as a candidate standard section frame when the consistency score exceeds a preset threshold. Verification module: The second convolutional neural network is used in conjunction with the anatomical key points to verify and judge the developmental status of the candidate standard section frames, and the developmental status classification results and confidence scores are obtained. Output module: Based on the coordinates of the anatomical key points, calculate the bony acetabular angle and cartilaginous acetabular angle using the Graf method, overlay the angle lines of each angle with the anatomical key points on the original image, generate visualization results, and output a structured report containing visualization results, anatomical key point coordinates, angle parameters, and developmental status classification results.

Images