Dynamic segmentation and monitoring method and system for achilles tendon broken end low echo area

Abstract

The application relates to the technical field of medical image processing, in particular to a method for dynamic segmentation and monitoring of a low echo area of an Achilles tendon broken end. The method comprises the following steps: S1, in the first frame of ultrasound longitudinal section image of the Achilles tendon part of a user, the entire Achilles tendon broken end ROI is framed, and the left and right side boundaries of the low echo area in the ROI are searched; S2, key points are selected on the left and right side boundaries; S3, ultrasound image sequences are continuously collected, for each collected frame of ultrasound image, the ROI thereof is aligned with the ROI of the last frame of ultrasound image, and the position of the key points in the current frame of ultrasound image is located in the ROI prediction neighborhood of the aligned current frame image; S4, by analyzing the change amplitude and direction of the multi-dimensional distance vectors between the continuous frames, the compression or stretching state of the low echo area is judged; and S5, according to the key point set, the closed boundary of the low echo area of the current frame is fitted in real time, and the maximum horizontal width, area and morphological complexity parameters of the low echo area are calculated.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method and system for dynamic segmentation and monitoring of the hypoechoic area at the Achilles tendon rupture site. Background Technology

[0002] Achilles tendon rupture is one of the most common injuries in sports medicine. The quality of postoperative rehabilitation training directly affects the patient's functional recovery and the risk of re-rupture. In recent years, wearable ultrasound imaging systems have been increasingly used for postoperative rehabilitation monitoring of Achilles tendon rupture due to their portability, non-invasiveness, and real-time capabilities. By continuously observing ultrasound images of the Achilles tendon rupture area, especially the morphological changes of hypoechoic areas (corresponding to hematomas, granulation tissue, and other unhealed tissues) before and after mechanical loading (such as standing and heel raises), the degree of healing can be quantitatively assessed, guiding the intensity of rehabilitation.

[0003] However, in wearable ultrasound monitoring scenarios, automatic segmentation of the hypoechoic area at the Achilles tendon rupture end faces the following technical challenges: probe movement causing region of interest (ROI) drift; low contrast and noise interference in ultrasound images; individual anatomical differences; and lack of temporal continuity leading to abrupt changes in segmentation results. Therefore, there is an urgent need for a robust and highly accurate method for monitoring the condition of the Achilles tendon rupture end that can adapt to dynamic wearable ultrasound monitoring scenarios. Summary of the Invention

[0004] This application provides a method and system for dynamic segmentation and monitoring of the hypoechoic area at the Achilles tendon rupture site. It is adaptable to wearable ultrasound dynamic monitoring scenarios and offers advantages such as strong robustness and high segmentation accuracy. This method and system can adapt to challenges such as probe movement, poor image quality, and significant individual differences, and features an automatic segmentation and monitoring method and system with consistent inter-frame temporal sequence.

[0005] Firstly, a method for dynamic segmentation and monitoring of the hypoechoic area at the Achilles tendon rupture site based on a wearable ultrasound device is provided, comprising the following steps: S1, after the user wears the wearable ultrasound device, the first frame of the ultrasound longitudinal section image of the user's Achilles tendon is acquired in the user's resting seated state. In the first frame of the ultrasound longitudinal section image, the entire region of interest (ROI) at the Achilles tendon rupture site is selected, and the left and right boundaries of the hypoechoic area are searched within the ROI along a direction perpendicular to the long axis of the Achilles tendon, and the initial coordinate set of the points on the left and right boundaries of the hypoechoic area is recorded; S2, on the left and right boundaries, several points are selected as key points to be tracked based on the discrete curvature of the points; S3, during the user's Achilles tendon rehabilitation training, a sequence of ultrasound images is continuously acquired, and for each acquired image... For each frame of ultrasound image, its ROI is aligned with the ROI of the previous frame of ultrasound image. Within the predicted neighborhood of the aligned ROI of the current frame image, the key points of the previous frame ultrasound image obtained in step S2 are independently tracked, and the position of the key point in the current frame ultrasound image is located as the key point of the current frame ultrasound image; S4, for each frame of ultrasound image, the horizontal distance between the corresponding key points on the left and right sides is calculated in real time to generate a multidimensional spacing vector. By analyzing the change amplitude and direction of the multidimensional spacing vector between consecutive frames, the compression or stretching state of the hypoechoic area is automatically determined; S5, based on the set of all tracked key points, the closed boundary of the hypoechoic area of ​​the current frame is fitted in real time, and the maximum horizontal width, area and morphological complexity parameters of the hypoechoic area are calculated.

[0006] In some embodiments, on the left and right boundaries, a number of points are selected as key points to be tracked based on the discrete curvature of the points, including: S21, calculating the discrete curvature of each point on each boundary of the left and right boundaries, and selecting a number of points with the largest curvature as candidate points; S22, supplementing the remaining boundary points with candidate points of uniform spacing according to the principle of equal spacing; wherein, the remaining boundary points are the points on all points on the boundary other than the candidate points; S23, calculating the local contrast in the neighborhood of each of the above candidate points, eliminating candidate points with local contrast below a preset threshold, and using the remaining candidate points as key points to be tracked.

[0007] In some embodiments, the method further includes: for each key point, constructing a weighted fused combined feature descriptor vector based on the two-dimensional local binary pattern (LBP) and gray-level gradient orientation histogram (HOG) features of its neighborhood.

[0008] In one example of this embodiment, for each acquired ultrasound image frame, aligning its ROI with the ROI of the previous ultrasound image, and independently tracking key points obtained from the previous ultrasound image in the predicted ROI neighborhood of the aligned current frame image, and locating the position of the key points in the current frame ultrasound image, includes:

[0009] Based on the triaxial acceleration and angular velocity measured by the inertial measurement unit (IMU) in the wearable ultrasound device, a rigid body transformation is performed on the ROI in the previous frame of the ultrasound image to obtain the predicted ROI position of the current frame, thereby coarsely aligning the ROI of the current frame of the ultrasound image with the ROI of the previous frame of the ultrasound image. For each key point in the previous frame, a search window is set within the predicted ROI position of the current frame, centered on its coordinates in the previous frame. A template matching method based on Kalman filtering is used to calculate the similarity between each candidate position and the combined feature descriptor vector of the key point in the previous frame. The position with the highest similarity is taken as the precise position of the key point in the new frame.

[0010] In one example of this example, the method further includes: when the inertial measurement unit in the wearable ultrasound device detects that the transient acceleration exceeds a preset threshold, automatically increasing the key point tracking frame rate and expanding the process noise covariance of the Kalman filter; if the acceleration does not fall back below the threshold within 5 consecutive frames, triggering ROI re-retrieval and re-executing the key point tracking process of step S3.

[0011] In some embodiments, step S4 further includes: personalized baseline calibration: using the initial distribution of key point spacing in the first frame of ultrasound longitudinal section image as a reference template, and calculating the normalized rate of change of spacing relative to the reference template in subsequent dynamic monitoring.

[0012] In one example of this embodiment, step S4 further includes: an anomaly removal mechanism: calculating the median of the change rate of the spacing between each pair of key points in the current frame; if the change rate of the spacing between a pair of key points exceeds a set multiple of the median, it is determined to be a tracking anomaly, and the interpolation result of the spacing between adjacent normal key point pairs is used instead; or, fitting a smooth curve to the key points on the same boundary, and correcting the coordinates of key points that deviate from the curve by more than a set standard deviation to the curve.

[0013] In some embodiments, step S5 further includes: monitoring the dynamic change rate of the maximum horizontal width, area and morphological complexity parameters of the hypoechoic region before and after mechanical loading, generating a healing process curve, and displaying the current tissue mechanical properties in real time on the evaluation interface and indicating the risk of recovery.

[0014] Secondly, a wearable ultrasound-based dynamic segmentation and monitoring system for the hypoechoic area of ​​the Achilles tendon rupture site is provided, comprising: a wearable ultrasound device and a monitoring terminal; wherein the wearable ultrasound device comprises: a wearable ultrasound probe for acquiring ultrasound longitudinal section images of the Achilles tendon; an inertial measurement unit integrated into the base of the wearable ultrasound probe for detecting the acceleration and angular velocity of the probe; a portable ultrasound host with a built-in image acquisition card and an embedded processing unit, the embedded processing unit being configured to perform the steps of the method described in the first aspect; a wireless transmission module for transmitting the processing results to the monitoring terminal in real time; and an evaluation interface for displaying in real time the maximum horizontal width, area, morphological complexity parameters, and risk warning information of the hypoechoic area.

[0015] In some embodiments, the embedded processing unit includes: a first-frame initialization module, used to select a Region of Interest (ROI) in the first frame ultrasound image and search for the left and right boundaries of the hypoechoic area; a key point selection and descriptor vector construction module, used to select key points to be tracked on the boundaries and construct combined feature descriptor vectors; a key point tracking module, used to align the ROI using data from the inertial measurement unit and track the key points frame by frame; a dynamic monitoring and anomaly removal module, used to calculate the key point spacing, identify anomalies, and correct them through consistency and anatomical continuity constraints; and a boundary reconstruction and evaluation module, used to fit the closed boundary of the hypoechoic area and calculate quantitative evaluation parameters.

[0016] In summary, the proposed solution employs a two-tiered strategy of "ROI alignment + fine-grained tracking using combined feature descriptor vectors" to effectively address the minute displacements of wearable ultrasound probes along the skin, ensuring consistent tracking of key points across consecutive frames and exhibiting strong robustness against ROI drift and motion. Furthermore, this application abandons the approach of directly segmenting hypoechoic regions globally, instead tracking sparse key points with highly recognizable textures on both sides of the boundary and then fitting the boundary. This significantly reduces sensitivity to low image contrast, strong speckle noise, and blurred boundaries, resulting in high segmentation accuracy for low-quality ultrasound images. Moreover, the inter-frame tracking mechanism of key points naturally carries temporal information; combined with spacing consistency constraints and anatomical continuity correction, it effectively avoids abrupt changes in the segmentation boundary over time, naturally incorporating temporal continuity. Building upon this foundation, this application also eliminates the influence of individual anatomical differences through personalized static baseline calibration, directly quantifies the rehabilitation biomechanical state through dynamic changes in horizontal spacing, and possesses anti-motion interference capabilities as well as intelligent rehabilitation assessment and risk warning functions, thus comprehensively improving clinical applicability and monitoring reliability. Attached Figure Description

[0017] Figure 1 A flowchart of a method for monitoring the state of Achilles tendon rupture ends provided in this application;

[0018] Figure 2A flowchart of another method for monitoring the condition of Achilles tendon ends provided in this application;

[0019] Figure 3 A schematic diagram of key point extraction and spacing measurement of the hypoechoic area boundary of the Achilles tendon rupture end in the ultrasound image provided in this application;

[0020] Figure 4 A schematic diagram illustrating the key point tracking principle provided in this application;

[0021] Figure 5 The diagram shows the structure of the wearable system and the workflow of the IMU-assisted motion interference suppression module provided in this application. Detailed Implementation

[0022] The solution provided in this application will now be described with reference to the accompanying drawings. In this application, "multiple" refers to two or more objects, and "various kinds" refers to two or more types. Terms such as "first," "second," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or number of objects.

[0023] This application provides a method for dynamic segmentation and monitoring of the hypoechoic area at the Achilles tendon rupture site based on a wearable ultrasound device. The method may include... Figure 1 The steps shown.

[0024] First, in step S1, after the user puts on the wearable ultrasound device, the first frame of ultrasound longitudinal section image of the user's Achilles tendon is acquired in the user's resting seated position. In the first frame of ultrasound longitudinal section image, the entire region of interest (ROI) of the Achilles tendon rupture is selected, and the left and right boundaries of the hypoechoic area are searched within the ROI along the direction perpendicular to the long axis of the Achilles tendon. The initial coordinate set of the left and right boundary points of the hypoechoic area is recorded.

[0025] Wearable ultrasound devices belong to wearable monitoring systems, which also include monitoring terminals located at the rehabilitation physician's office.

[0026] Wearable ultrasound devices may include a wearable ultrasound probe, a portable ultrasound unit, and a wireless transmission module. The wearable ultrasound probe can be secured to the posterior aspect of the lower leg corresponding to the Achilles tendon rupture site using an elastic band. The portable ultrasound unit connects to the probe via a flexible cable. The portable ultrasound unit incorporates an image acquisition card and an embedded processing unit. The embedded processing unit is used for execution... Figure 3 The steps in the method are shown. The wireless transmission module can transmit the processing results to the rehabilitation physician's monitoring terminal in real time.

[0027] In some embodiments, the wearable ultrasound probe is a linear array probe with a center frequency of 7.5 MHz and an imaging depth of 4 cm. In some embodiments, the wearable ultrasound probe base integrates a microelectromechanical inertial measurement unit (IMU). The parameters of the IMU are: triaxial accelerometer ±16g, triaxial gyroscope ±2000° / s, and sampling rate 100Hz.

[0028] In step S1, the embedded processing unit mainly performs first-frame initialization and low-echo zone boundary retrieval. After the patient puts on the wearable ultrasound device, they activate the wearable monitoring system in a resting, seated position (without weight-bearing on the Achilles tendon). The wearable ultrasound device acquires the first frame of a longitudinal ultrasound image (with an image resolution of 512×768 pixels).

[0029] In some embodiments, rehabilitation physicians can manually select a rectangular region of interest (ROI) (approximately 200×400 pixels) containing the Achilles tendon rupture area along the long axis of the Achilles tendon on a touchscreen. In some embodiments, the ROI is automatically selected by a pre-trained object detection network, eliminating the need for manual operation by the physician.

[0030] After determining the ROI, the wearable ultrasound device automatically performs the following sub-steps:

[0031] The image within the ROI is denoised using Gaussian filtering (convolution kernel 5×5, σ=1.2).

[0032] Adaptive binarization is performed on the filtered image to obtain the initial low-echo candidate region;

[0033] Introduce anatomical prior constraints: using the long axis of the Achilles tendon as the reference direction, regions in the binarized results whose long axis projection length is less than 1.5 times the short axis projection length are removed;

[0034] Within the preserved connected components, scan along the direction perpendicular to the long axis of the Achilles tendon. When the gray value gradient exceeds 0.6 times the standard deviation of the image gray level and the gray level of the next 5 consecutive pixels increases, record the position as a boundary point.

[0035] Scanning line by line along the long axis of the Achilles tendon, we obtain the coordinate sets of the left and right boundary points.

[0036] Next, in step S2, on the left and right boundaries, several points are selected as key points to be tracked based on the discrete curvature of the points.

[0037] In step 2, the embedded processing unit mainly performs boundary key point selection and descriptor vector construction.

[0038] On the acquired left and right boundaries, the wearable ultrasound device automatically selects several points with unique local texture features as key points to be tracked.

[0039] In some embodiments, the selection of key points can be achieved through the following steps:

[0040] Step S21, Curvature significance calculation: Calculate the discrete curvature of each point on each boundary in the left and right boundary, and select several points with the largest curvature as candidate points; for example, for the left boundary curve, calculate the discrete curvature point by point (using the three-point method), sort the curvature values ​​from largest to smallest, and select the top 2 points with the largest curvature as candidate points with significant curvature.

[0041] Step S22, Uniform Supplement Selection: Supplement candidate points with uniform spacing from the remaining boundary points according to the principle of equal spacing; wherein, the remaining boundary points are all points on the boundary except for the candidate points; in other words, candidate points with uniform spacing can be supplemented from the remaining boundary points according to the principle of approximately equal spacing.

[0042] Step S23, Local Texture Filtering: Calculate the local contrast within the neighborhood of each candidate point, remove candidate points with local contrast below a preset threshold, and use the remaining candidate points as key points to be tracked. Specifically, for each candidate point, calculate the local contrast within its 16×16 pixel neighborhood (defined as the grayscale standard deviation within the neighborhood). If the local contrast of a candidate point is less than 0.5 times the median contrast of all boundary points' neighborhoods, it is removed and replaced by the next point in the neighborhood that meets the condition.

[0043] For example, using the above method, K=4 key points (L1, L2, L3, L4) are selected on the left boundary, and K=4 key points (R1, R2, R3, R4) are selected on the right boundary, for a total of 8 key points, forming 4 pairs of corresponding relationships.

[0044] For each keypoint, a 32×32 pixel square neighborhood is drawn centered on its coordinates, and a combined feature descriptor vector is constructed. Specifically, gray-level gradient features and local texture features within the neighborhood are extracted and fused in concatenation to obtain a high-dimensional feature vector, which uniquely represents the keypoint. The local texture features can be two-dimensional Local Binary Patterns (LBP), and the gray-level gradient features can be histograms of oriented gradients (HOG). In other words, for each keypoint, a weighted and fused combined feature descriptor vector is constructed based on the two-dimensional Local Binary Patterns (LBP) and histograms of oriented gradients (HOG) features of its neighborhood.

[0045] More specifically, LBP features (dimensions) The uniform LBP operator is used to divide the 32×32 neighborhood into 16 non-overlapping 8×8 pixel sub-windows of 4×4. A 59-dimensional uniform LBP histogram is calculated for each sub-window, and the histograms are concatenated and then reduced to 256 dimensions by PCA.

[0046] HOG features (dimensions) The 32×32 neighborhood is divided into 16 8×8 pixel cell units of 4×4. Each cell is used to calculate the gradient magnitude weighted histogram of 9 gradient directions. 2×2 cells are used to form a block with 50% overlap between blocks, forming a total of 9 blocks. After concatenation, a 324-dimensional HOG feature is obtained.

[0047] Weighted fusion: ,in and These are the normalized LBP and HOG feature vectors, respectively. α is the fusion weight, and we take α=0.5. Finally, each keypoint is uniquely represented by a 580-dimensional combined feature descriptor vector.

[0048] Next, in step S3, during the user's Achilles tendon rehabilitation training, a series of ultrasound images are continuously acquired. For each acquired ultrasound image, its ROI is aligned with the ROI of the previous ultrasound image. Within the ROI prediction neighborhood of the aligned current frame image, the key points of the previous ultrasound image obtained in step S2 are independently tracked, and the position of the key point in the current frame ultrasound image is located as the key point of the current frame ultrasound image.

[0049] In step S3, the embedded processing unit primarily performs keypoint tracking. Specifically, in step S3, the ROI of the ultrasound image is coarsely aligned with the ROI of the previous frame's ultrasound image to obtain the ROI prediction neighborhood. Then, the precise location, or specific location, of the keypoint is obtained within the ROI prediction neighborhood.

[0050] In some embodiments, key point tracking is performed as follows:

[0051] The patient begins heel raise training: from a standing position with feet flat on the ground, slowly raise the heels to approximately 30° plantar flexion, then slowly lower them back down. The system continuously acquires ultrasound image sequences (30fps) during this process.

[0052] For each new image after the first frame, perform the following operations:

[0053] Coarse alignment: The ROI of the current frame is coarsely aligned with the ROI of the previous frame of the ultrasound image. This embodiment uses a template matching method based on image grayscale to estimate the global translation of the ROI, and performs a translation transformation on the ROI of the previous frame to obtain the predicted position of the ROI in the current frame.

[0054] Fine-grained tracking based on Kalman filtering: Within the predicted neighborhood of the ROI after coarse alignment, keypoints obtained in step S2 are tracked independently. Specifically, for each keypoint in the previous frame, a search window is set centered on its coordinates in the previous frame within the predicted ROI location in the current frame. A template matching method based on Kalman filtering is used to calculate the similarity between each candidate position and the keypoint descriptor in the previous frame. The position with the highest similarity is taken as the precise position of the keypoint in the new frame.

[0055] In one example, fine-grained keypoint tracking based on Kalman filtering may include the following specific operations:

[0056] For each keypoint in the previous frame, within the ROI prediction location of the current frame, a 31×31 pixel search window is set, centered on its coordinates from the previous frame. Within the search window, the combined descriptor vector corresponding to the candidate location is calculated pixel-by-pixel, along with the descriptor vector from the previous frame. The cosine similarity between the two is used to select the candidate position with the highest similarity as the observation position z_t;

[0057] Inputting the observed position z_t into the Kalman filter yields the optimal estimated position of the keypoint in the current frame. The state variables of the Kalman filter are: The observed variable is A uniform motion model is adopted.

[0058] In some embodiments, the wearable ultrasound device is also equipped with an IMU, and key point tracking can be performed in step S3 by combining the measurement results of the IMU, as follows:

[0059] Based on the triaxial acceleration and angular velocity measured by the inertial measurement unit (IMU) in the wearable ultrasound device, a rigid body transformation is performed on the ROI in the previous frame of the ultrasound image to obtain the predicted ROI position of the current frame, thereby coarsely aligning the ROI of the current frame of the ultrasound image with the ROI of the previous frame of the ultrasound image. For each key point in the previous frame, a search window is set within the predicted ROI position of the current frame, centered on its coordinates in the previous frame. A template matching method based on Kalman filtering is used to calculate the similarity between each candidate position and the combined feature descriptor vector of the key point in the previous frame. The position with the highest similarity is taken as the precise position of the key point in the new frame.

[0060] In this embodiment, the embedded processing unit reads IMU data at a frequency of 50Hz and calculates the probe's pose change between frames using complementary filtering: the translation vector Δt and the rotation matrix Rt. Using Δt and Rt, a rigid transformation is performed on the ROI bounding box of the previous frame to obtain the predicted ROI position of the current frame, thereby coarsely aligning the current frame ROI with its previous frame ROI.

[0061] An example of this embodiment also provides an anti-motion interference mechanism. Specifically, when the inertial measurement unit in the wearable ultrasound device detects that the transient acceleration exceeds a preset threshold, it automatically increases the key point tracking frame rate and expands the process noise covariance of the Kalman filter. If the acceleration does not fall back below the threshold within 5 consecutive frames, it triggers ROI re-retrieval and re-executes the key point tracking process in step S3.

[0062] For example, when the IMU detects a transient acceleration exceeding 1.5g in any axis, the embedded processing unit automatically performs the following operations:

[0063] Temporarily increase the frame rate for keypoint tracking from 30fps to 60fps;

[0064] Increase the process noise covariance of the Kalman filter (to 3 times the normal value);

[0065] If the acceleration does not fall below the threshold within 5 consecutive frames, the ROI re-search is triggered, and the key point tracking process in step S3 is re-executed.

[0066] Next, in step S4, for each frame of ultrasound image, the horizontal distance between corresponding key points on the left and right sides is calculated in real time to generate a multidimensional spacing vector. By analyzing the change amplitude and direction of the multidimensional spacing vector between consecutive frames, the pressure or stretching state of the hypoechoic area is automatically determined.

[0067] In step S4, the embedded processing unit performs dynamic monitoring and status judgment of the spacing. Specifically, for each frame of ultrasound image, the horizontal distance between corresponding key points on the left and right sides is calculated in real time, generating a multi-dimensional spacing vector (a 4-dimensional vector in this embodiment). By analyzing the amplitude and direction of the changes in the spacing vector between consecutive frames, the compressed or stretched state of the hypoechoic region is automatically determined: when the spacing between each pair of key points generally decreases compared to the previous frame, the hypoechoic region is determined to be in a compressed state (corresponding to the Achilles tendon being stretched); when the spacing between each pair of key points generally increases compared to the previous frame, the hypoechoic region is determined to be in a stretched state.

[0068] In some embodiments, the embedded processing unit can also perform personalized baseline calibration. Specifically, it uses the initial distribution of key point spacing in the first frame of the ultrasound longitudinal section image as a reference template, and calculates the normalized rate of change of the spacing relative to this reference template during subsequent dynamic monitoring. For example, before the patient begins heel raise training, the system continuously acquires 30 frames of images in a weightless, seated, resting state, tracks key points according to the above steps, and calculates the mean spacing between each pair of key points. , as the patient's personalized distance baseline vector In subsequent dynamic monitoring, the normalized rate of change of the spacing relative to the template is calculated: .

[0069] In some embodiments, the embedded processing unit can also perform an anomaly removal mechanism, specifically: calculating the median of the change rate of the spacing between each pair of key points in the current frame; if the change rate of the spacing between a pair of key points exceeds a set multiple of the median, it is determined to be a tracking anomaly, and the interpolation result of the spacing between adjacent normal key point pairs is used instead; or, fitting a smooth curve to the key points on the same boundary, and correcting the coordinates of key points that deviate from the curve by more than a set standard deviation to the curve.

[0070] More specifically, the anomaly removal mechanism can include spacing consistency constraints and anatomical continuity constraints. The spacing consistency constraint calculates the absolute value of the rate of change of the spacing between keypoints in the current frame (4 pairs). Take the median (Median) If a pair of key points If the distance exceeds 2.5 times the median, it is judged as a tracking anomaly, and the linear interpolation result of the spacing between adjacent normal keypoints is used instead.

[0071] Anatomical continuity constraint: For four key points on the same boundary, a cubic smooth spline curve is fitted using the least squares method. If the deviation of a key point from the fitted curve exceeds three times the standard deviation, the coordinates of that key point are corrected to the nearest point on the fitted curve.

[0072] Then, in step S5, based on the set of all tracked key points, the closed boundary of the low echo region of the current frame is fitted in real time, and the maximum horizontal width, area and morphological complexity parameters of the low echo region are calculated.

[0073] In step S5, the embedded processing unit performs boundary reconstruction and parameter calculation. Specifically, based on the set of all tracked keypoints, the closed boundary of the low-echo region in the current frame is fitted in real time: curve interpolation is performed on the keypoints on both sides to obtain smooth left and right boundary curves, which are then connected to form a closed contour. Based on the closed contour, the following parameters are calculated:

[0074] Maximum horizontal width (t): Take the maximum horizontal spacing between 4 pairs of key points. j=1,2,3,4;

[0075] Low echo area A(t): calculated by the number of pixels within the closed contour and converted to mm² based on the image pixel pitch;

[0076] Morphological complexity: Where P is the perimeter of the outline, A is the area, and the closer the C value is to 1, the closer the shape is to a circle. Well-healed tissue should gradually tend to a regular strip shape (C value slightly greater than 1).

[0077] Rehabilitation physicians can monitor the dynamic changes of the above parameters in real time during heel raise training and assess the Achilles tendon healing process through a monitoring terminal.

[0078] In some embodiments, the key points have undergone anomaly removal and personalized baseline calibration. In step S5, the embedded processing unit can reconstruct the low-echo zone boundary in real time based on the key points that have undergone anomaly removal and personalized baseline calibration: cubic spline interpolation is performed on the key points on the left and right sides respectively to obtain smooth left and right boundary curves, which are then connected to form a closed contour. The maximum horizontal width is calculated based on the closed contour. Area A(t) and morphological complexity .

[0079] Dynamic monitoring and risk alerts: The embedded processing unit monitors the dynamic rate of change of the above parameters before and after mechanical loading (heel raise training), generating a healing progress curve. The rehabilitation assessment interface displays the results in real time in the form of a dashboard.

[0080] current Values ​​and rates of change relative to the baseline;

[0081] In this round of training The maximum magnitude of change;

[0082] Load response curve ( vs. plantar flexion angle);

[0083] Risk warning: If If the load increases instead of compressing (i.e., the low-echo region is stretched instead of compressed), or if the rate of change of W_max decreases for three consecutive training rounds, the system will automatically pop up a warning: "Healing progress has stalled, it is recommended to reduce training intensity."

[0084] In some embodiments, to facilitate observation by rehabilitation physicians, the dynamic change rate of the maximum horizontal width, area, and morphological complexity parameters of the hypoechoic region before and after mechanical loading can be monitored to generate a healing process curve. The current tissue mechanical properties are then displayed in real time on the evaluation interface, and rehabilitation risks are indicated. The embedded processing unit automatically saves the peak values ​​from each training session. Parameters such as area A and morphological complexity C are used to generate a periphery healing trend map, which allows rehabilitation physicians to objectively assess the Achilles tendon healing process and quantitatively adjust the rehabilitation prescription for the next stage.

[0085] This application also provides a dynamic segmentation and monitoring system for the hypoechoic area of ​​the Achilles tendon rupture site based on wearable ultrasound. The system includes: a wearable ultrasound device and a monitoring terminal; wherein, the wearable ultrasound device includes:

[0086] A wearable ultrasound probe is used to acquire ultrasound longitudinal section images of the Achilles tendon area; an inertial measurement unit, integrated into the base of the wearable ultrasound probe, is used to detect the probe's acceleration and angular velocity; a portable ultrasound host has a built-in image acquisition card and an embedded processing unit, the embedded processing unit being configured to execute... Figure 1 The steps of the method shown include: a wireless transmission module for transmitting the processing results to the monitoring terminal in real time; and an evaluation interface for displaying the maximum horizontal width, area, morphological complexity parameters, and risk warning information of the low-echo zone in real time.

[0087] In some embodiments, the wearable ultrasound probe is a linear array probe with a center frequency of 7.5 MHz and an imaging depth of 4 cm. It is fixed to the posterior aspect of the lower leg corresponding to the Achilles tendon rupture site by an elastic band. The probe is used to acquire ultrasound longitudinal section images of the Achilles tendon area, with an image resolution of 512×768 pixels and an output frame rate of 30 fps.

[0088] In some embodiments, the inertial measurement unit (IMU) is integrated into the wearable ultrasonic probe base and manufactured using microelectromechanical systems (MEMS) technology. The IMU includes a triaxial accelerometer (range ±16g) and a triaxial gyroscope (range ±2000° / s), with a sampling rate of 100Hz, for real-time detection of the probe's acceleration and angular velocity, providing data support for ROI coarse alignment and motion interference resistance.

[0089] In some embodiments, the portable ultrasound unit is connected to the probe via a flexible ribbon cable and includes a built-in image acquisition card and an embedded processing unit. The embedded processing unit uses an ARM architecture processor (1.5GHz, quad-core), equipped with 4GB of RAM and 32GB of storage, and is configured to execute... Figure 1 The steps of the method are shown. The host also has a built-in Wi-Fi wireless transmission module and a Bluetooth module.

[0090] In some embodiments, the wireless transmission module uses the Wi-Fi protocol to transmit the processed segmentation results, quantitative evaluation parameters, and risk warning information to the monitoring terminal (the rehabilitation physician's workstation or the patient's mobile APP) in real time, with a transmission latency of less than 100ms.

[0091] In some embodiments, the evaluation interface is deployed on the display screen of the monitoring terminal to display in real time the maximum horizontal width, area, morphological complexity parameters of the low-echo zone, and risk warning information. The interface adopts a dashboard design, including three main areas: a real-time value display area, a historical curve area, and a risk warning area.

[0092] In some embodiments, the embedded processing unit is internally divided into the following modules according to their functions: first frame initialization module, key point selection and descriptor vector construction module, key point tracking module, dynamic monitoring and anomaly removal module, and boundary reconstruction and evaluation module.

[0093] The first-frame initialization module is used to select the Region of Interest (ROI) in the first frame of the ultrasound image and search for the left and right boundaries of the hypoechoic area. For example, the first-frame initialization module specifically implements the following functions:

[0094] Receive ROI coordinates from manual selection or output by the object detection network;

[0095] Gaussian filtering and adaptive binarization are applied to the image within the ROI;

[0096] Candidate regions were selected by combining anatomical priors (characteristics of hypoechoic areas extending along the long axis of the Achilles tendon);

[0097] The algorithm searches for boundary points on both sides along a direction perpendicular to the long axis of the Achilles tendon, and outputs a set of boundary point coordinates. The keypoint selection and descriptor vector construction module is used to select keypoints to be tracked on the boundaries and construct combined feature descriptor vectors. Specifically, the keypoint selection and descriptor vector construction module implements the following functions:

[0098] Calculate the discrete curvature of each point on the boundary, and select several points with the largest curvature as significant candidate points of curvature.

[0099] Candidate points with uniform spacing were selected to supplement the selection based on the principle of equal spacing.

[0100] Calculate the local contrast within the neighborhood of candidate points and remove points with low texture discriminability;

[0101] For the selected key points, a weighted fusion combined feature descriptor vector is constructed based on LBP and HOG features.

[0102] The keypoint tracking module is used for ROI alignment using data from the inertial measurement unit and for frame-by-frame tracking of keypoints. Specifically, the keypoint tracking module implements the following functions:

[0103] Read IMU data, calculate the inter-frame pose changes (translation and rotation) of the probe, and perform rigid body transformation on the ROI to achieve coarse alignment;

[0104] Within the predicted neighborhood of the coarsely aligned ROI, template matching is performed based on the cosine similarity of the descriptor vectors;

[0105] The key point locations are optimally estimated by combining a Kalman filter.

[0106] When the IMU detects that the transient acceleration exceeds the threshold, it automatically increases the tracking frame rate and triggers a new ROI search. The dynamic monitoring and anomaly removal module calculates keypoint spacing, identifies anomalies, and corrects them using consistency and dissection continuity constraints. Specifically, the dynamic monitoring and anomaly removal module implements the following functions:

[0107] Perform personalized baseline calibration and record the key point spacing reference template under unloaded static conditions;

[0108] Real-time calculation of the horizontal spacing and normalized rate of change of corresponding key points on the left and right;

[0109] Based on the spacing consistency constraint: calculate the median of the spacing change rate, and replace outliers that exceed the set multiple with interpolation results;

[0110] Based on anatomical continuity constraints: a smooth curve is fitted to key points on the same side, and points deviating from the limits are corrected onto the curve. The boundary reconstruction and evaluation module is used to fit the closed boundary of the hypoechoic region and calculate quantitative evaluation parameters. Specifically, the boundary reconstruction and evaluation module implements the following functions:

[0111] Spline interpolation is performed on the corrected left and right key points to obtain smooth boundary curves;

[0112] Connecting the beginning and end points of the boundary forms a closed profile;

[0113] Calculate the maximum horizontal width, area, and shape complexity parameters;

[0114] Monitor the dynamic rate of change of parameters before and after mechanical loading to generate a healing process curve;

[0115] The assessment interface displays the current tissue biomechanical properties in real time and indicates rehabilitation risks.

[0116] Next, taking a patient undergoing Achilles tendon rupture surgery performing heel raise exercises as an example, we will introduce the workflow of the automatic segmentation and tracking method for the hypoechoic area of ​​the Achilles tendon rupture ends provided by us, as follows:

[0117] 1. The patient puts on the wearable ultrasound probe and starts the system. The first frame initialization module automatically completes the ROI setting and boundary extraction.

[0118] 2. The key point selection module selects 8 key points on the boundary and constructs a descriptor vector.

[0119] 3. The patient begins heel raise training. The key point tracking module tracks the key point position frame by frame at a frame rate of 30fps, and IMU data assists in coarse alignment of ROI in real time.

[0120] 4. The dynamic monitoring module calculates the distance between each key point in real time, and the anomaly removal module detects and corrects tracked anomalies.

[0121] 5. The boundary reconstruction module fits the boundary of the low-echo zone in real time and calculates the maximum width, area, and morphological complexity.

[0122] 6. The assessment interface displays the above parameters and risk warnings in real time, and the wireless transmission module synchronizes the data to the rehabilitation physician's terminal.

[0123] 7. After the training is completed, the system will automatically save all the parameters of this training for long-term healing trend analysis.

[0124] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for dynamic segmentation and monitoring of the hypoechoic region at the Achilles tendon rupture end based on wearable ultrasound equipment, characterized in that, Includes the following steps: S1. After the user puts on the wearable ultrasound device, the first frame of the ultrasound longitudinal section image of the user's Achilles tendon is acquired while the user is in a resting seated position. In the first frame of the ultrasound longitudinal section image, the entire region of interest (ROI) of the Achilles tendon rupture is selected, and the left and right boundaries of the hypoechoic area are searched within the ROI along a direction perpendicular to the long axis of the Achilles tendon. The initial coordinate set of the points on the left and right boundaries of the hypoechoic area is recorded. S2. On the left and right boundaries, several points are selected as key points to be tracked based on the discrete curvature of the points. S3. During the user's Achilles tendon rehabilitation training, a sequence of ultrasound images is continuously acquired. For each frame of ultrasound image acquired, its ROI is compared with the previous frame of ultrasound image. The ROIs are aligned, and within the predicted ROI neighborhood of the aligned current frame image, the key points of the previous frame ultrasound image obtained in step S2 are independently tracked, and the position of the key point in the current frame ultrasound image is located as the key point of the current frame ultrasound image; S4, for each frame ultrasound image, the horizontal distance between the corresponding key points on the left and right sides is calculated in real time, and a multi-dimensional spacing vector is generated. By analyzing the change amplitude and direction of the multi-dimensional spacing vector between consecutive frames, the compression or stretching state of the hypoechoic area is automatically determined; S5, based on the set of all tracked key points, the closed boundary of the hypoechoic area of ​​the current frame is fitted in real time, and the maximum horizontal width, area and morphological complexity parameters of the hypoechoic area are calculated.

2. The method according to claim 1, characterized in that, On the left and right boundaries, several points are selected as key points to be tracked based on the discrete curvature of the points, including: S21, calculate the discrete curvature of each point on each boundary in the left and right side boundaries, and select several points with the largest curvature as candidate points; S22, select candidate points with uniform spacing from the remaining boundary points according to the principle of equal spacing; wherein, the remaining boundary points are the points other than the candidate points among all points on the boundary; S23, calculate the local contrast in the neighborhood of each candidate point, remove candidate points whose local contrast is lower than a preset threshold, and use the remaining candidate points as key points to be tracked.

3. The method according to claim 1, characterized in that, Also includes: For each key point, a weighted and fused combined feature descriptor vector is constructed based on the two-dimensional local binary pattern (LBP) and gray-level gradient orientation histogram (HOG) features of its neighborhood.

4. The method according to claim 3, characterized in that, For each acquired ultrasound image frame, its ROI is aligned with the ROI of the previous ultrasound image frame. Within the predicted ROI neighborhood of the aligned current frame image, key points obtained from the previous ultrasound image frame in step S2 are independently tracked to locate their positions in the current frame ultrasound image, including: Based on the triaxial acceleration and angular velocity measured by the inertial measurement unit (IMU) in the wearable ultrasound device, a rigid body transformation is performed on the ROI in the previous frame of the ultrasound image to obtain the predicted ROI position of the current frame, thereby coarsely aligning the ROI of the current frame of the ultrasound image with the ROI of the previous frame of the ultrasound image. For each key point in the previous frame, a search window is set within the predicted ROI position of the current frame, centered on its coordinates in the previous frame. A template matching method based on Kalman filtering is used to calculate the similarity between each candidate position and the combined feature descriptor vector of the key point in the previous frame. The position with the highest similarity is taken as the precise position of the key point in the new frame.

5. The method according to claim 4, characterized in that, The method further includes: When the inertial measurement unit in the wearable ultrasound device detects that the transient acceleration exceeds the preset threshold, it automatically increases the key point tracking frame rate and expands the process noise covariance of the Kalman filter. If the acceleration does not fall back below the threshold within 5 consecutive frames, it triggers ROI re-retrieval and re-executes the key point tracking process in step S3.

6. The method according to claim 1, characterized in that, Step S4 also includes: Personalized baseline calibration: The initial distribution of key point spacing in the first frame of ultrasound longitudinal section image is used as a reference template, and the normalized rate of change of spacing relative to the reference template is calculated in subsequent dynamic monitoring.

7. The method according to claim 6, characterized in that, Step S4 also includes: an anomaly removal mechanism: calculate the median of the change rate of the spacing between each pair of key points in the current frame. If the change rate of the spacing between a pair of key points exceeds a set multiple of the median, it is determined to be a tracking anomaly and replaced with the interpolation result of the spacing between adjacent normal key point pairs; or, fit a smooth curve to the key points on the same boundary and correct the coordinates of key points that deviate from the curve by more than a set standard deviation to the curve.

8. The method according to claim 1, characterized in that, Step S5 further includes: monitoring the dynamic change rate of the maximum horizontal width, area and morphological complexity parameters of the hypoechoic area before and after mechanical loading, generating a healing process curve, and displaying the current tissue mechanical properties in real time on the evaluation interface and indicating the risk of recovery.

9. A dynamic segmentation and monitoring system for the hypoechoic area of ​​the Achilles tendon rupture end based on wearable ultrasound, characterized in that, include: Wearable ultrasound device and monitoring terminal; wherein, the wearable ultrasound device includes: A wearable ultrasound probe for acquiring ultrasound longitudinal section images of the Achilles tendon; an inertial measurement unit integrated into the wearable ultrasound probe base for detecting the probe's acceleration and angular velocity; a portable ultrasound host with a built-in image acquisition card and an embedded processing unit configured to perform the steps of the method according to any one of claims 1 to 8; a wireless transmission module for transmitting the processing results to the monitoring terminal in real time; and an evaluation interface for displaying in real time the maximum horizontal width, area, morphological complexity parameters, and risk warning information of the hypoechoic region.

10. The system according to claim 9, characterized in that, The embedded processing unit includes: a first-frame initialization module, used to select the Region of Interest (ROI) in the first frame ultrasound image and search for the left and right boundaries of the hypoechoic area; a key point selection and descriptor vector construction module, used to select key points to be tracked on the boundaries and construct combined feature descriptor vectors; a key point tracking module, used to align the ROI using data from the inertial measurement unit and track the key points frame by frame; a dynamic monitoring and anomaly removal module, used to calculate the key point spacing, identify anomalies, and correct them through consistency and anatomical continuity constraints; and a boundary reconstruction and evaluation module, used to fit the closed boundary of the hypoechoic area and calculate quantitative evaluation parameters.

Images