Automatic measuring method for rectus abdominis based on ultrasonic imaging and ultrasonic imaging system
By combining deep learning segmentation and ultrasound elastography, and optimizing the endpoints by utilizing the difference in stiffness between the rectus abdominis muscle and the linea alba, the problem of interference from the linea alba in rectus abdominis muscle measurement was solved, and accurate automatic measurement of the rectus abdominis muscle separation distance was achieved.
Patent Information
- Application Number
- CN202511717688.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-02-10
AI Technical Summary
In existing technologies, when using deep learning segmentation models to measure the distance between the rectus abdominis muscles, the measurement accuracy is easily affected by the features of the linea alba, resulting in poor measurement accuracy. Furthermore, it relies on the doctor's experience, is time-consuming and laborious, and makes it difficult to achieve accurate diagnosis of rectus abdominis diastasis.
A deep learning-based segmentation model was used to segment ultrasound images. Ultrasonic elastography was combined with the hardness measurement results of multiple sampling points on the line of interest. The hardness difference between the rectus abdominis muscle and the linea alba was used to optimize the endpoints and determine the separation distance of the rectus abdominis muscle.
It enables precise positioning of the rectus abdominis muscle separation distance, improves measurement accuracy and robustness, supports fully automated measurement, and reduces reliance on physician experience.
Smart Images
Figure CN121489538A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ultrasound imaging technology, specifically to an automatic measurement method and system for rectus abdominis muscle based on ultrasound imaging. Background Technology
[0002] The rectus abdominis muscle, a key component of the core muscle group, plays a vital role in protecting internal organs and stabilizing the pelvis and lumbar spine. However, during pregnancy, physiological changes significantly increase the risk of diastasis recti in women, leading to decreased abdominal muscle function, lower back pain, prolonged postpartum recovery, and adverse effects on physical appearance. In severe cases, it can even cause hernias and increase the risk of abdominal surgery, greatly reducing the patient's quality of life and negatively impacting their physical and mental health. Therefore, early and accurate diagnosis of diastasis recti is crucial for subsequent treatment and patient rehabilitation.
[0003] Ultrasound examination, due to its non-invasive nature, has become an ideal means of screening and diagnosing diastasis recti. The distance between the separated rectus abdominis muscles can be measured through ultrasound images, serving as a diagnostic basis. However, in clinical practice, relying on experienced ultrasound technicians to measure the distance between the rectus abdominis muscles from multiple ultrasound images is both time-consuming and labor-intensive, and the diagnostic results are highly dependent on the technician's individual experience. Furthermore, the presence of the linea alba can easily interfere with the accurate localization of the rectus abdominis endpoints, and the varying levels of experience among different doctors may lead to inaccurate diagnostic results. Currently, deep learning-based methods are widely used in medical image segmentation due to their accuracy and lack of need for manual parameter tuning. However, current algorithms for predicting diastasis recti lack the ability to learn subtle image features and are easily interfered with by the features of the linea alba, resulting in decreased recognition accuracy and making it impossible to accurately measure the distance between the rectus abdominis muscles.
[0004] Therefore, there is an urgent need to provide an automatic measurement method and system for rectus abdominis muscle based on ultrasound imaging, which can overcome the problem of poor measurement accuracy in existing deep learning measurements of rectus abdominis muscle spacing. Summary of the Invention
[0005] In view of this, it is necessary to provide an automatic measurement method and system for rectus abdominis muscle based on ultrasound imaging, so as to solve the technical problem of poor measurement accuracy when using a single method of deep learning segmentation model for automatic measurement of rectus abdominis muscle in the existing technology.
[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides an automatic measurement method for the rectus abdominis muscle based on ultrasound imaging, comprising: An ultrasound image of the rectus abdominis region is acquired, and the ultrasound image is segmented based on a deep learning segmentation model to obtain the model prediction result. The model prediction result is then binarized to obtain the rectus abdominis segmentation result. Based on the rectus abdominis muscle segmentation results, the initial endpoints of the first and second rectus abdominis muscles, as well as a line of interest traversing the rectus abdominis muscle region, are determined. Based on the spatial location corresponding to the line of interest, the hardness measurement results of multiple sampling points on the line of interest are obtained by ultrasonic elastography. Based on the hardness measurement results and the preset linea alba hardness threshold and the preset rectus abdominis hardness threshold, the first rectus abdominis initial endpoint and the second rectus abdominis initial endpoint are optimized to obtain the first rectus abdominis optimized endpoint and the second rectus abdominis optimized endpoint. The distance between the first optimized endpoint of the rectus abdominis muscle and the second optimized endpoint of the rectus abdominis muscle is taken as the rectus abdominis separation distance.
[0007] In one possible implementation, the rectus abdominis segmentation result includes a left-side connected region and a right-side connected region; then, determining the first initial endpoint of the rectus abdominis muscle, the second initial endpoint of the rectus abdominis muscle, and a line of interest traversing the rectus abdominis muscle region based on the rectus abdominis segmentation result includes: In the image coordinate system of the ultrasound image, the axis-aligned minimum bounding rectangle of the rectus abdominis muscle segmentation result is determined; the image coordinate system has the upper left corner of the image as the origin, the horizontal axis of the human body as the X-axis, and the vertical axis of the human body as the Y-axis. The right endpoint of the left connected region is taken as the initial endpoint of the first rectus abdominis muscle, and the left endpoint of the right connected region is taken as the initial endpoint of the second rectus abdominis muscle. The axis-aligned minimum bounding rectangle is taken as the region of interest. Within the region of interest, a Y-axis ray is drawn for each abscissa in the rectus abdominis muscle segmentation result, and all ordinates of the Y-axis ray intersecting with the rectus abdominis muscle segmentation result are obtained. Calculate the median of all the ordinates, use the median as the center point, and connect the center points to obtain the line of interest.
[0008] In one possible implementation, acquiring the hardness measurement results of multiple sampling points along the line of interest using ultrasonic elastography includes: Echo signals from multiple sampling points along the line of interest were acquired using ultrasonic elastography. The co-directional and quadrature components of the echo signal are determined, and the tissue displacement of each sampling point is determined based on the co-directional and quadrature components. The peak time difference between the two sampling points is determined based on the change in tissue displacement over time. The elastic modulus of each sampling point is determined based on the position difference between two sampling points and the time difference of the peak, and the elastic modulus is used as the hardness measurement result.
[0009] In one possible implementation, the elastic modulus is:
[0010] In the formula, It is the elastic modulus; The difference between the positions of the two sampling points; The peak time difference between the two sampling points; Shear wave velocity; This refers to the tissue density in the rectus abdominis muscle region.
[0011] In one possible implementation, before determining the peak time difference between two spatial points based on the change of tissue displacement over time, the method further includes: The tissue displacement is subjected to low-pass filtering.
[0012] In one possible implementation, optimizing the first and second initial endpoints of the rectus abdominis muscle based on the hardness measurement results and a preset linea alba hardness threshold and a preset rectus abdominis hardness threshold to obtain optimized endpoints of the first and second rectus abdominis muscles includes: Using the horizontal coordinate of the spatial location as the X-axis and the hardness measurement result as the Y-axis, hardness distribution data is obtained; At least one candidate region is determined based on the linea alba hardness threshold and the rectus abdominis hardness threshold; the candidate region is a region whose hardness is less than the linea alba hardness threshold and greater than the rectus abdominis hardness threshold; The leftmost region and the rightmost region in the candidate region are determined, wherein the leftmost region includes a first left boundary line and a first right boundary line, and the rightmost region includes a second left boundary line and a second right boundary line; Calculate the spatial distance between the initial endpoint of the first rectus abdominis muscle and the left and right boundaries of the leftmost region, and take the point on the boundary that is closer to the initial endpoint of the first rectus abdominis muscle as the optimized endpoint of the first rectus abdominis muscle. Calculate the spatial distance between the initial endpoint of the second rectus abdominis muscle and the left and right boundaries of the rightmost region, and take the point on the boundary that is closer to the initial endpoint of the second rectus abdominis muscle as the optimized endpoint of the second rectus abdominis muscle.
[0013] In one possible implementation, the deep learning segmentation model includes an encoder, a first cross-scale aggregation transformer, a second cross-scale aggregation transformer, and a decoder; The encoder is used to extract the first-scale features, second-scale features and third-scale features of the ultrasound image; The first cross-scale aggregation transformer is used to fuse the first scale features and the second scale features based on a cross-attention mechanism to obtain the first fused features; The second cross-scale aggregation transformer is used to perform feature fusion of the first fused feature and the third scale feature based on a cross-attention mechanism to obtain the second fused feature; The decoder is used to decode the first scale feature, the first fusion feature, and the second fusion feature to obtain the model prediction result.
[0014] In one possible implementation, the encoder includes an image block layer, a linear coding layer, two first visual state space blocks, a first image block aggregation layer, two second visual state space blocks, a second image block aggregation layer, and two third visual state space blocks. The image segmentation layer is used to divide the ultrasound image into multiple image blocks; The linear coding layer is used to linearly encode the image blocks to obtain coded features; The first visual state space block is used to extract features from the encoded features to obtain the first scale features; The first image block aggregation layer is used to merge the first scale features into blocks to obtain the first merged features; The second visual state space block is used to extract features from the first merged features to obtain the second scale features; The second image block aggregation layer is used to merge the second scale features into blocks to obtain the second merged features; The third visual state space block is used to extract features from the second merged features to obtain the third scale features.
[0015] In one possible implementation, the deep learning segmentation model uses a loss function that is a weighted sum of a binary cross-entropy loss function and a dice loss function during training.
[0016] In a second aspect, the present invention also provides an ultrasound imaging system, comprising: The rectus abdominis initial endpoint localization unit is used to acquire ultrasound images of the rectus abdominis region, segment the ultrasound images based on a deep learning segmentation model, obtain model prediction results, and binarize the model prediction results to obtain rectus abdominis segmentation results. Based on the rectus abdominis segmentation results, the first rectus abdominis initial endpoint, the second rectus abdominis initial endpoint, and a line of interest that runs through the rectus abdominis region are determined. The hardness measurement unit is used to acquire the hardness measurement results of multiple sampling points on the line of interest based on the spatial position corresponding to the line of interest using ultrasonic elastography. The endpoint optimization unit is used to optimize the first and second initial endpoints of the rectus abdominis muscle based on the hardness measurement results and the preset linea alba hardness threshold and the preset rectus abdominis hardness threshold, so as to obtain the optimized endpoints of the first and second rectus abdominis muscles. The rectus abdominis muscle separation distance determination unit is used to determine the distance between the first optimized endpoint of the rectus abdominis muscle and the second optimized endpoint of the rectus abdominis muscle as the rectus abdominis muscle separation distance.
[0017] The beneficial effects of this invention are as follows: The automatic measurement method for rectus abdominis muscle based on ultrasound imaging provided by this invention, after segmenting the ultrasound image based on a deep learning segmentation model to obtain the first and second initial endpoints of the rectus abdominis muscle, obtains the hardness measurement results of multiple sampling points on the line of interest through ultrasound elastography. Based on the hardness measurement results, the first and second initial endpoints of the rectus abdominis muscle are optimized. It utilizes the elastic difference in hardness between the rectus abdominis muscle and the linea alba, effectively overcoming the problem of easy interference from the linea alba image when traditional segmentation is based solely on a deep learning segmentation model. It combines information from two different dimensions: image morphology and tissue function, thereby achieving precise positioning of the medial boundary of the rectus abdominis muscle and greatly improving the accuracy of rectus abdominis muscle separation distance measurement.
[0018] Furthermore, this invention enables fully automated measurement of the rectus abdominis muscle separation distance, and by incorporating the tissue function parameter of stiffness measurement, it improves the robustness of the assessment of the degree of rectus abdominis muscle separation, providing significant clinical implications for the accurate measurement of the rectus abdominis muscle separation distance. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 A schematic flowchart of an embodiment of the automatic measurement method for rectus abdominis muscle based on ultrasound imaging provided by the present invention; Figure 2 A schematic diagram of an ultrasound image 2 cm above the umbilicus provided by the present invention; Figure 3 This is a schematic diagram of the rectus abdominis muscle segmentation result 2cm above the umbilicus provided by the present invention; Figure 4 For the present invention Figure 1 A schematic diagram of an embodiment of step S102; Figure 5A schematic diagram of an embodiment of the region of interest and line of interest provided by the present invention; Figure 6 For the present invention Figure 1 A schematic flowchart of an embodiment of step S103; Figure 7 A schematic diagram of an embodiment of the shear wave velocity calculation process provided by the present invention; Figure 8 A schematic diagram of an embodiment of the hardness measurement results provided by the present invention; Figure 9 For the present invention Figure 1 A schematic flowchart of an embodiment of step S104; Figure 10 A schematic diagram of an embodiment of the endpoint optimization process provided by the present invention; Figure 11 A schematic diagram of an embodiment of the deep learning segmentation model provided by the present invention; Figure 12 This is a schematic diagram of an embodiment of the ultrasound imaging system provided by the present invention. Detailed Implementation
[0021] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0022] It should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0024] This invention provides an automatic measurement method and ultrasound imaging system for rectus abdominis muscle based on ultrasound imaging, which will be described below.
[0025] Figure 1 This is a schematic flowchart of an embodiment of the automatic measurement method for rectus abdominis muscle based on ultrasound imaging provided by the present invention, as shown below. Figure 1 As shown, the automatic measurement method for the rectus abdominis muscle based on ultrasound imaging includes: S101. Obtain ultrasound images of the rectus abdominis region, segment the ultrasound images based on a deep learning segmentation model, obtain the model prediction results, and binarize the model prediction results to obtain the rectus abdominis segmentation results.
[0026] The ultrasound images can be acquired in real time during step S101, or by calling pre-acquired ultrasound images during step S101.
[0027] Among them, the deep learning segmentation model can be any mature image segmentation model that is currently used in the field of ultrasound imaging technology.
[0028] It should be understood that deep learning segmentation models need to be trained and tested before they can be used in the segmentation process in step S101.
[0029] Specifically, prior to step S101, the process also includes training and testing the deep learning segmentation model: Step 1: Establish the dataset for model training. First, crop the images of patients with diastasis recti and those without diastasis recti, removing redundant information that might interfere with the model's prediction. Then, a professional ultrasound physician annotates the rectus abdominis contour mask in each image and measures the distance. Finally, the cropped original image, the rectus abdominis contour mask, and the distance between the rectus abdominis muscles are retained as a data set.
[0030] Step 2: Set a random seed and divide the sample set into a training set and a test set in a 4:1 ratio. The training set data is augmented by data augmentation (adding noise, rotating, and increasing contrast) to build the dataset for model training.
[0031] Step 3: Construct the initial deep learning segmentation model. Train the initial deep learning segmentation model based on the training set in Step 2. The training process uses the stochastic gradient descent algorithm for iterative training.
[0032] Step 4: Determine if the maximum number of iterations or loss value convergence has been reached. If the maximum number of iterations or loss value convergence has been reached, stop training and test the trained model on the test machine. If the test passes, save the tested model and use it as the deep learning segmentation model. Stop training when the set number of iterations is reached.
[0033] When executing step S101, the saved deep learning segmentation model is invoked.
[0034] like Figure 2 and Figure 3 As shown, Figure 2 The ultrasound image is located 2 cm above the navel. Figure 3 This is the segmentation result of the rectus abdominis muscle 2cm above the umbilicus after segmentation using a deep learning segmentation model. Figure 3 The white area in the diagram represents the rectus abdominis muscle area, while the black area represents the non-rectus abdominis muscle area.
[0035] S102. Based on the rectus abdominis muscle segmentation results, determine the initial endpoints of the first and second rectus abdominis muscles and a line of interest that runs through the rectus abdominis muscle region. S103. Based on the spatial location corresponding to the line of interest, the hardness measurement results of multiple sampling points on the line of interest are obtained by ultrasonic elastography.
[0036] Specifically, ultrasonic elastography involves applying excitation to each sampling point using an ultrasonic probe, receiving the echo signal after the excitation is applied, analyzing the echo signal, and obtaining the hardness measurement result.
[0037] S104. Based on the hardness measurement results and the preset hardness threshold of the linea alba and the preset hardness threshold of the rectus abdominis muscle, optimize the initial endpoints of the first rectus abdominis muscle and the second rectus abdominis muscle to obtain the optimized endpoints of the first rectus abdominis muscle and the second rectus abdominis muscle.
[0038] It should be understood that both the linea alba hardness threshold and the rectus abdominis hardness threshold are specific hardness values, and the linea alba hardness threshold is higher than the rectus abdominis hardness threshold.
[0039] S105. The distance between the optimized endpoints of the first and second rectus abdominis muscles is taken as the rectus abdominis separation distance.
[0040] It should be understood that the automatic rectus abdominis muscle measurement method based on ultrasound imaging in this embodiment of the invention can be implemented in any electronic device based on the automatic rectus abdominis muscle measurement method based on ultrasound imaging, such as an ultrasound imaging system. Specifically, the automatic rectus abdominis muscle measurement method based on ultrasound imaging is stored in the device as a compiled program. When the device is started, the program is invoked, and the automatic rectus abdominis muscle measurement method based on ultrasound imaging is implemented.
[0041] Compared with existing technologies, the automatic measurement method for rectus abdominis muscle based on ultrasound imaging provided in this invention, after segmenting the ultrasound image based on a deep learning segmentation model to obtain the first and second initial endpoints of the rectus abdominis muscle, obtains the hardness measurement results of multiple sampling points on the line of interest through ultrasound elastography. Based on the hardness measurement results, the first and second initial endpoints of the rectus abdominis muscle are optimized. It utilizes the elastic difference in hardness between the rectus abdominis muscle and the linea alba, effectively overcoming the problem of easy interference from the linea alba image when relying solely on deep learning segmentation models for segmentation. It combines information from two different dimensions: image morphology and tissue function, thereby achieving precise positioning of the medial boundary of the rectus abdominis muscle and greatly improving the accuracy of rectus abdominis muscle separation distance measurement.
[0042] Furthermore, the embodiments of the present invention can realize fully automated measurement of the rectus abdominis muscle separation distance, and because it incorporates the tissue function parameter of stiffness measurement results, it improves the robustness of the assessment of the degree of rectus abdominis muscle separation, providing significant clinical significance for the accurate measurement of the rectus abdominis muscle separation distance.
[0043] In some embodiments of the present invention, such as Figure 3 As shown, the rectus abdominis segmentation results include a left-side connected region and a right-side connected region; then, as... Figure 4 As shown, step S102 includes: S401. In the image coordinate system of the ultrasound image, determine the smallest bounding rectangle of the axis alignment of the rectus abdominis muscle segmentation result; the image coordinate system takes the upper left corner of the image as the origin, the horizontal axis of the human body as the X-axis, and the vertical axis of the human body as the Y-axis.
[0044] like Figure 5 As shown, Figure 5 The top left corner is the origin of the coordinate system. The direction from left to right is the positive X-axis, and the direction from top to bottom is the positive Y-axis.
[0045] S402. Take the right endpoint of the left connected region as the initial endpoint of the first rectus abdominis muscle, and the left endpoint of the right connected region as the initial endpoint of the second rectus abdominis muscle. S403. Take the axis-aligned minimum bounding rectangle as the region of interest (ROI). Within the ROI, draw a Y-axis ray for each x-coordinate in the rectus abdominis segmentation result, and obtain all y-coordinates where the Y-axis ray intersects with the rectus abdominis segmentation result.
[0046] Specifically, the axis-aligned minimum bounding rectangle is the region composed of the minimum X-coordinate Xmin, maximum X-coordinate Xmax, minimum Y-coordinate Ymin, and maximum Y-coordinate Ymax of the rectangle. All X-coordinates within this region form the X-coordinate set. The Y-axis rays corresponding to the X-coordinates in this set are traversed to obtain the set of Y-coordinates where the rays intersect the segmentation results. The midpoint of this set is calculated to obtain Ymid. Connecting the corresponding (X, Ymid) points sequentially yields the line of interest.
[0047] Among them, the axis-aligned minimum outer rectangle refers to the smallest rectangle whose sides are completely parallel to the coordinate axes of the coordinate system and can just wrap around the target object.
[0048] In this embodiment of the invention, taking into account irregular noise and other issues present at the boundaries of the segmentation results, the noise can be robustly contained by using the smallest outer rectangle aligned with the axis as the region of interest. This avoids the drastic changes in the region of interest that would occur with a rotating rectangle due to a single noise point.
[0049] S404. Calculate the median of all ordinates, use the median as the center point, and connect the center points to obtain the line of interest (LOI).
[0050] Considering that the interference of the linea alba mainly manifests as blurred boundaries and adhesions, this embodiment of the invention uses the median of all vertical coordinates as the center point and the line connecting the center points as the line of interest. This extracts a pure line of interest within the region of interest that is not affected by the linea alba, which can further improve the accuracy of the subsequent determination of the rectus abdominis muscle separation distance based on the line of interest.
[0051] Specifically, such as Figure 5 As shown, Figure 5 The yellow area in the diagram represents the outline of the rectus abdominis muscle segmentation result, the red rectangle represents the region of interest, and the green line represents the line of interest.
[0052] In some embodiments of the present invention, such as Figure 6 As shown, step S103 includes: S601. Echo signals from multiple sampling points along the line of interest are obtained using ultrasonic elastography.
[0053] Specifically, based on the mapping relationship between the image coordinates of the line of interest and the actual tissue sampling point location, the sampling point location is determined, and an excitation is applied at the sampling point by the ultrasound probe to obtain the echo signal. The excitation method includes acoustic radiation force excitation or low-frequency vibration excitation.
[0054] S602. Determine the co-directional and quadrature components of the echo signal, and determine the tissue displacement of each sampling point based on the co-directional and quadrature components.
[0055] The relationship between tissue displacement and its unidirectional and orthogonal components is as follows:
[0056] In the formula, For organizational displacement; The center frequency of the probe; Speed of sound; For window length; and These are echo signals from two adjacent frames; I These are components in the same direction; Q These are orthogonal components.
[0057] S603. Determine the peak time difference between two sampling points based on the change of tissue displacement over time.
[0058] Specifically, the waveforms of displacement changes over time at multiple sampling points are detected separately, and then cross-correlation calculations are performed on the vibration waveforms at adjacent positions to determine the time offset of the peak, i.e., the peak time difference.
[0059] S604. Determine the elastic modulus of each sampling point based on the position difference and peak time difference between the two sampling points, and use the elastic modulus as the result of hardness measurement.
[0060] In a specific embodiment of the present invention, the elastic modulus is:
[0061] In the formula, It is the elastic modulus; The difference between the positions of the two sampling points; The peak time difference between the two sampling points; Shear wave velocity; This refers to the tissue density in the rectus abdominis muscle region.
[0062] To eliminate the influence of noise on the measurement results, in some embodiments of the present invention, before step S603, the method further includes: performing low-pass filtering on the tissue displacement.
[0063] It should be understood that, in addition to low-pass filtering, other filtering methods can be selected based on the characteristics of tissue displacement data, without making specific limitations here.
[0064] In a specific embodiment of the present invention Figure 7The vertical axis on the left represents the sampling point location, and the horizontal axis represents time. Each curve in the figure represents the waveform of the tissue's vibrational displacement over time at different sampling points. It indicates that when an excitation is applied to a point in the tissue, a shear wave is generated, which propagates outwards like ripples created by a pebble thrown into water. Each curve shows how the tissue at that specific location vibrates over time as the wave propagates to that location. Figure 7 The right side shows the shear wave velocity obtained after cross-correlation calculation. Specifically, two adjacent spatial points are selected, and their vibration waveforms are extracted. Then, after cross-correlation calculation, the time offset corresponding to the maximum value of the cross-correlation coefficient is taken as the time difference required for the shear wave to propagate from the i-th spatial point to the (i+1)-th spatial point. Combined with the positional difference between the two spatial points, the shear wave velocity can be obtained.
[0065] In a specific embodiment of the present invention, the final hardness measurement result is as follows: Figure 8 As shown, the color changes from blue to red, indicating that the elastic modulus increases.
[0066] In some embodiments of the present invention, such as Figure 9 As shown, step S104 includes: S901. Using the horizontal coordinate of the spatial location as the X-axis and the hardness measurement result as the Y-axis, hardness distribution data is obtained. S902. Determine at least one candidate region based on the hardness threshold of the linea alba and the hardness threshold of the rectus abdominis muscle; the candidate region is a region whose hardness is less than the hardness threshold of the linea alba and greater than the hardness threshold of the rectus abdominis muscle. S903. Determine the leftmost region and the rightmost region in the candidate regions. The leftmost region includes the first left boundary line and the first right boundary line, and the rightmost region includes the second left boundary line and the second right boundary line. S904. Calculate the spatial distance between the initial endpoint of the first rectus abdominis muscle and the left and right boundaries of the leftmost region, and take the point on the boundary that is closer to the initial endpoint of the first rectus abdominis muscle as the optimized endpoint of the first rectus abdominis muscle. S905. Calculate the spatial distance between the initial endpoint of the second rectus abdominis muscle and the left and right boundaries of the rightmost region, and take the point on the boundary that is closer to the initial endpoint of the second rectus abdominis muscle as the optimized endpoint of the second rectus abdominis muscle.
[0067] In specific embodiments of the present invention, such as Figure 10 As shown, Figure 10The curves in the diagram represent the stiffness distribution data. The X-coordinates corresponding to points A and B are the x-coordinates of the initial endpoints of the first and second rectus abdominis muscles. The shaded area on the left is the leftmost region, and the shaded area on the right is the rightmost region. After comparison, point C is the optimized endpoint of the first rectus abdominis muscle, and point D is the optimized endpoint of the second rectus abdominis muscle. In other words, the distance between points C and D is the rectus abdominis muscle separation distance.
[0068] In this embodiment of the invention, the candidate region is a reliable region that satisfies the elastic modulus. Within this reliable region, a precise point is determined on its left and right boundaries based on the principle of proximity to the initial endpoint. This ensures the stability of the endpoint detection results and the reasonable inheritance of the initial segmentation results. Consequently, the accuracy of the determined rectus abdominis muscle separation distance can be ensured.
[0069] To further eliminate the influence of the linea alba on the accuracy of measurement results, in some embodiments of the present invention, such as... Figure 11 As shown, the deep learning segmentation model includes an encoder, a first scale-wise aggregation transformer (SAT), a second scale-wise aggregation transformer, and a decoder. The encoder is used to extract the first-scale, second-scale, and third-scale features of ultrasound images; The first cross-scale aggregation transformer is used to fuse the first-scale features and the second-scale features based on a cross-attention mechanism to obtain the first fused features. The second cross-scale aggregation transformer is used to perform feature fusion of the first fused feature and the third scale feature based on the cross attention mechanism to obtain the second fused feature; The decoder is used to decode the first scale feature, the first fusion feature, and the second fusion feature to obtain the model prediction result.
[0070] The first and second cross-scale aggregation transformers simultaneously receive both a shallow Textural Feature Map and a deep Semantic Feature Map. In the attention mechanism, these two inputs are typically assigned different roles: the Semantic Feature Map is used as the Query (Q), and the Textural Feature Map as the Key (K) and Value (V). Similarity is calculated between the Query (Semantic) and the Key (Textural), and the resulting attention weights indicate which texture details are most relevant and important for each high-level semantic feature point. These weights are then used to perform a weighted summation on the Value (Textural), thereby filtering and aggregating the most useful details for the current semantic segmentation task from the original texture features, outputting a texture-enhanced feature map.
[0071] This invention, by adding a first cross-scale aggregation transformer and a second cross-scale aggregation transformer, can perform multi-head cross-attention learning on shallow-scale, detail-rich features and deep-scale, semantically clear features. This suppresses texture noise irrelevant to the segmentation task (e.g., irregular textures within the linea alba, interfering textures from other tissues), while highlighting and enhancing key contours and texture information strongly correlated with the rectus abdominis muscle boundary. This eliminates the technical problem of susceptibility to interference from the linea alba in existing deep learning segmentation models, further improving the accuracy of rectus abdominis muscle segmentation results, and consequently, further improving the accuracy of rectus abdominis muscle separation distance measurement results.
[0072] Furthermore, such as Figure 11 As shown, the encoder includes an image patch partition layer, a linear embedding layer, two first visual state space (VSS) blocks, a first image patch merging layer, two second visual state space blocks, a second image patch merging layer, and two third visual state space blocks. Image segmentation layers are used to divide ultrasound images into multiple image blocks; Linear coding layers are used to linearly encode image patches to obtain coded features; The first visual state space block is used to extract features from the encoded features to obtain the first scale features; The first image block aggregation layer is used to merge the first scale features into blocks to obtain the first merged features; The second visual state space block is used to extract features from the first merged features to obtain the second scale features; The second image block aggregation layer is used to merge the second-scale features into blocks to obtain the second merged features; The third visual state space block is used to extract features from the second merged features to obtain the third scale features.
[0073] The purpose of the first image block aggregation layer and the second image block aggregation layer is to achieve downsampling.
[0074] In this embodiment of the invention, VSS Block is used to replace the original convolutional layer, which is beneficial for the network to model the complex relationship between the rectus abdominis muscle, linea alba, and surrounding tissues, and further improves the accuracy of deep learning segmentation model in segmenting the rectus abdominis muscle and linea alba.
[0075] In specific embodiments of the present invention, such as Figure 11 As shown, the decoder includes three cascaded decoding units, each of which includes a convolutional layer and a transposed convolutional (ConvTranspose) layer. The purpose of the transposed convolutional layer is to achieve upsampling.
[0076] In a specific embodiment of the present invention, the loss function used during training of the deep learning segmentation model is a weighted sum of the binary cross-entropy loss function and the dice loss function.
[0077] The binary cross-entropy loss function is:
[0078] The dice loss function is:
[0079] In the formula, This represents the total number of pixels in the ultrasound image. For each pixel i, the label is 1 for the foreground and 0 for the background. Let be the probability value for pixel i to be predicted as a foreground class, and its value range is [0,1]. TP This represents the number of pixels in the foreground where both the label and the prediction result are foreground pixels. FP This represents the number of pixels predicted as foreground but labeled as background. FN This represents the number of pixels that form the background in the prediction result but are labeled as the foreground.
[0080] In summary, the embodiments of the present invention realize the automatic measurement of the inter-rectus abdominis muscle length, eliminating the need for doctors to manually measure the distance. At the same time, by combining the functional parameter of rigidity, the accuracy of the inter-rectus abdominis muscle length measurement is improved.
[0081] On the other hand, embodiments of the present invention also provide an ultrasound imaging system, such as Figure 12 As shown, the ultrasound imaging system 1200 includes: The rectus abdominis initial endpoint localization unit 1201 is used to acquire ultrasound images of the rectus abdominis region, segment the ultrasound images based on a deep learning segmentation model, obtain model prediction results, and binarize the model prediction results to obtain rectus abdominis segmentation results. Based on the rectus abdominis segmentation results, the first rectus abdominis initial endpoint, the second rectus abdominis initial endpoint, and a line of interest that runs through the rectus abdominis region are determined. The hardness measurement unit 1202 is used to acquire the hardness measurement results of multiple sampling points on the line of interest based on the spatial position corresponding to the line of interest through ultrasonic elastography. The endpoint optimization unit 1203 is used to optimize the initial endpoints of the first rectus abdominis muscle and the second rectus abdominis muscle based on the hardness measurement results and the preset hardness thresholds of the linea alba and the preset hardness thresholds of the rectus abdominis muscle, so as to obtain the optimized endpoints of the first rectus abdominis muscle and the optimized endpoints of the second rectus abdominis muscle. The rectus abdominis muscle separation distance determination unit 1204 is used to determine the distance between the first optimized endpoint of the rectus abdominis muscle and the second optimized endpoint of the rectus abdominis muscle as the rectus abdominis muscle separation distance.
[0082] The ultrasound imaging system 1200 provided in the above embodiments can realize the technical solutions described in the above embodiments of the automatic measurement method for rectus abdominis muscle based on ultrasound imaging. The specific implementation principles of each module or unit can be found in the corresponding content in the above embodiments of the automatic measurement method for rectus abdominis muscle based on ultrasound imaging, and will not be repeated here.
[0083] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0084] The above provides a detailed description of the automatic measurement method and ultrasound imaging system for rectus abdominis muscle based on ultrasound imaging provided by the present invention. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. An automatic measurement method for rectus abdominis muscle based on ultrasound imaging, characterized in that, include: An ultrasound image of the rectus abdominis region is acquired, and the ultrasound image is segmented based on a deep learning segmentation model to obtain the model prediction result. The model prediction result is then binarized to obtain the rectus abdominis segmentation result. Based on the rectus abdominis muscle segmentation results, the initial endpoints of the first and second rectus abdominis muscles, as well as a line of interest traversing the rectus abdominis muscle region, are determined. Based on the spatial location corresponding to the line of interest, the hardness measurement results of multiple sampling points on the line of interest are obtained by ultrasonic elastography. Based on the hardness measurement results and the preset linea alba hardness threshold and the preset rectus abdominis hardness threshold, the first rectus abdominis initial endpoint and the second rectus abdominis initial endpoint are optimized to obtain the first rectus abdominis optimized endpoint and the second rectus abdominis optimized endpoint. The distance between the first optimized endpoint of the rectus abdominis muscle and the second optimized endpoint of the rectus abdominis muscle is taken as the rectus abdominis separation distance.
2. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 1, characterized in that, The rectus abdominis segmentation result includes a left-side connected region and a right-side connected region; therefore, determining the first initial endpoint of the rectus abdominis muscle, the second initial endpoint of the rectus abdominis muscle, and a line of interest traversing the rectus abdominis muscle region based on the rectus abdominis segmentation result includes: In the image coordinate system of the ultrasound image, the axis-aligned minimum bounding rectangle of the rectus abdominis muscle segmentation result is determined; the image coordinate system has the upper left corner of the image as the origin, the horizontal axis of the human body as the X-axis, and the vertical axis of the human body as the Y-axis. The right endpoint of the left connected region is taken as the initial endpoint of the first rectus abdominis muscle, and the left endpoint of the right connected region is taken as the initial endpoint of the second rectus abdominis muscle. The axis-aligned minimum bounding rectangle is taken as the region of interest. Within the region of interest, a Y-axis ray is drawn for each abscissa in the rectus abdominis muscle segmentation result, and all ordinates of the Y-axis ray intersecting with the rectus abdominis muscle segmentation result are obtained. Calculate the median of all the ordinates, use the median as the center point, and connect the center points to obtain the line of interest.
3. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 1, characterized in that, The process of obtaining hardness measurement results at multiple sampling points along the line of interest using ultrasonic elastography includes: Echo signals from multiple sampling points along the line of interest were acquired using ultrasonic elastography. The co-directional and quadrature components of the echo signal are determined, and the tissue displacement of each sampling point is determined based on the co-directional and quadrature components. The peak time difference between the two sampling points is determined based on the change in tissue displacement over time. The elastic modulus of each sampling point is determined based on the position difference between two sampling points and the time difference of the peak, and the elastic modulus is used as the hardness measurement result.
4. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 3, characterized in that, The elastic modulus is: In the formula, It is the elastic modulus; The difference between the positions of the two sampling points; The peak time difference between the two sampling points; Shear wave velocity; This refers to the tissue density in the rectus abdominis muscle region.
5. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 3, characterized in that, Before determining the peak time difference between two spatial points based on the change of tissue displacement over time, the method further includes: The tissue displacement is subjected to low-pass filtering.
6. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 1, characterized in that, The optimization of the first and second initial endpoints of the rectus abdominis muscle based on the hardness measurement results and preset hardness thresholds for the linea alba and rectus abdominis muscles, to obtain optimized endpoints for the first and second rectus abdominis muscles, includes: Using the horizontal coordinate of the spatial location as the X-axis and the hardness measurement result as the Y-axis, hardness distribution data is obtained; At least one candidate region is determined based on the linea alba hardness threshold and the rectus abdominis hardness threshold; the candidate region is a region whose hardness is less than the linea alba hardness threshold and greater than the rectus abdominis hardness threshold; The leftmost region and the rightmost region in the candidate region are determined, wherein the leftmost region includes a first left boundary line and a first right boundary line, and the rightmost region includes a second left boundary line and a second right boundary line; Calculate the spatial distance between the initial endpoint of the first rectus abdominis muscle and the left and right boundaries of the leftmost region, and take the point on the boundary that is closer to the initial endpoint of the first rectus abdominis muscle as the optimized endpoint of the first rectus abdominis muscle. Calculate the spatial distance between the initial endpoint of the second rectus abdominis muscle and the left and right boundaries of the rightmost region, and take the point on the boundary that is closer to the initial endpoint of the second rectus abdominis muscle as the optimized endpoint of the second rectus abdominis muscle.
7. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 1, characterized in that, The deep learning segmentation model includes an encoder, a first cross-scale aggregation transformer, a second cross-scale aggregation transformer, and a decoder; The encoder is used to extract the first-scale features, second-scale features and third-scale features of the ultrasound image; The first cross-scale aggregation transformer is used to fuse the first scale features and the second scale features based on a cross-attention mechanism to obtain the first fused features; The second cross-scale aggregation transformer is used to perform feature fusion of the first fused feature and the third scale feature based on a cross-attention mechanism to obtain the second fused feature; The decoder is used to decode the first scale feature, the first fusion feature, and the second fusion feature to obtain the model prediction result.
8. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 7, characterized in that, The encoder includes an image block layer, a linear coding layer, two first visual state space blocks, a first image block aggregation layer, two second visual state space blocks, a second image block aggregation layer, and two third visual state space blocks. The image segmentation layer is used to divide the ultrasound image into multiple image blocks; The linear coding layer is used to linearly encode the image blocks to obtain coded features; The first visual state space block is used to extract features from the encoded features to obtain the first scale features; The first image block aggregation layer is used to merge the first scale features into blocks to obtain the first merged features; The second visual state space block is used to extract features from the first merged features to obtain the second scale features; The second image block aggregation layer is used to merge the second scale features into blocks to obtain the second merged features; The third visual state space block is used to extract features from the second merged features to obtain the third scale features.
9. The automatic measurement method for rectus abdominis muscle based on ultrasound imaging according to claim 1, characterized in that, The deep learning segmentation model uses a weighted sum of a binary cross-entropy loss function and a dice loss function during training.
10. An ultrasound imaging system, characterized in that, include: The rectus abdominis initial endpoint localization unit is used to acquire ultrasound images of the rectus abdominis region, segment the ultrasound images based on a deep learning segmentation model, obtain model prediction results, and binarize the model prediction results to obtain rectus abdominis segmentation results. Based on the rectus abdominis segmentation results, the first rectus abdominis initial endpoint, the second rectus abdominis initial endpoint, and a line of interest that runs through the rectus abdominis region are determined. The hardness measurement unit is used to acquire the hardness measurement results of multiple sampling points on the line of interest based on the spatial position corresponding to the line of interest using ultrasonic elastography. The endpoint optimization unit is used to optimize the first and second initial endpoints of the rectus abdominis muscle based on the hardness measurement results and the preset linea alba hardness threshold and the preset rectus abdominis hardness threshold, so as to obtain the optimized endpoints of the first and second rectus abdominis muscles. The rectus abdominis muscle separation distance determination unit is used to determine the distance between the first optimized endpoint of the rectus abdominis muscle and the second optimized endpoint of the rectus abdominis muscle as the rectus abdominis muscle separation distance.