Clinical anesthesia ultrasonic image assisted positioning guiding method and system

Abstract

The invention relates to the technical field of clinical anesthesia ultrasonic navigation, and discloses a clinical anesthesia ultrasonic image assisted positioning guiding method and system. The method comprises the following steps: acquiring a real-time ultrasonic image flow of a target area; carrying out inter-frame stability analysis and screening clear frames to form an optimized image set; implementing multi-scale filtering to enhance the tissue edge; detecting an anatomical mark point and recording a three-dimensional position; establishing a dynamic space reference system and calculating a central point of a target area; deriving an initial motion path of the guiding instrument and executing initial positioning; synchronously capturing actual positioning point coordinates, and comparing the actual positioning point coordinates with a central point to generate a position difference index; calibrating a reference system according to the index and generating an optimized motion path; and an instrument is controlled to finish accurate positioning. The anatomical mark point identification accuracy is improved through image frame screening; positioning errors are dynamically compensated through a closed-loop calibration mechanism, and the navigation precision and operation reliability of anesthesia puncture are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clinical anesthesia ultrasound navigation, in particular to a clinical anesthesia ultrasound image assisted positioning and guiding method and system. BACKGROUND

[0002] In clinical anesthesia operation, especially in regional anesthesia techniques such as nerve block, accurately injecting anesthetic drugs around the target nerve is the key to ensure the effect and safety of anesthesia. Ultrasound imaging technology has become a standard means of realizing visual guidance, which allows the operator to observe the anatomical structure and the path of the puncture needle in real time. The existing ultrasound-guided puncture technology usually relies on the operator to hold the ultrasound probe to obtain images, while observing the screen and manually adjusting the direction and depth of the puncture needle. The effectiveness of this method is highly dependent on the experience and hand-eye coordination of the operator, and there is a risk of image blurring and positioning error due to physiological movement or operation jitter.

[0003] More advanced auxiliary systems attempt to provide the operator with a puncture path planning by combining ultrasound images with a spatial positioning system. Such systems usually complete a one-time spatial registration at the beginning of the process, that is, a fixed mapping relationship between the image coordinate system and the actual body position of the patient is established. In the subsequent puncture process, the system navigates according to this static spatial relationship. This static navigation mode is difficult to cope with the inevitable soft tissue deformation, unintentional small displacement of the probe or dynamic changes in the patient's body position in actual operation, resulting in deviation between the pre-planned path and the real-time anatomical position, and the navigation accuracy decreases.

[0004] Ultrasound images themselves have the characteristics of large noise and low contrast, and real-time image streams are mixed with low-quality frames caused by tissue movement or unstable probe contact. The conventional processing method is either to directly process the continuous image stream or to simply extract a single frame for analysis. This will expose the key anatomical landmark point recognition algorithm to unstable image data, causing inaccurate or failed feature point detection, thereby directly affecting the accuracy of the initial spatial registration and the reliability of the entire navigation system. SUMMARY

[0005] The purpose of the present application is to provide a clinical anesthesia ultrasound image assisted positioning and guiding method and system to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides a clinical anesthesia ultrasound image assisted positioning and guiding method, which comprises: starting an ultrasound image acquisition device and aligning the target area of the clinical anesthesia operation to continuously obtain a real-time ultrasound image stream; performing inter-frame stability analysis on the real-time ultrasound image stream, and selecting image frames with satisfactory clarity to form an optimized image set; The multi-scale filtering process is performed on the optimized image set to enhance the edge information of key tissues in the image; The pre-defined anatomical landmark points are detected from the enhanced image, and three-dimensional position data of the landmark points are recorded; A dynamic spatial reference frame is established using the three-dimensional position data of the anatomical landmark points, and a central reference point of the target region is calculated; An initial motion path of the guiding instrument is derived based on the dynamic spatial reference frame and the central reference point; Preliminary driving instructions are formed based on the initial motion path and transmitted to the guiding instrument; The guiding instrument performs a preliminary positioning operation in response to the preliminary driving instructions, while capturing real-time coordinates of the actual positioning point; The real-time coordinates of the actual positioning point are compared with the central reference point to generate a position difference index; The dynamic spatial reference frame is calibrated according to the position difference index to generate an optimized motion path; Final driving instructions are generated based on the optimized motion path to control the guiding instrument to achieve accurate positioning.

[0007] Preferably, the starting of the ultrasound image acquisition device and the alignment of the target region of the clinical anesthesia operation to continuously acquire a real-time ultrasound image stream includes: Adjust the incident angle and depth parameters of the ultrasound probe to ensure that the complete anatomical range of the target region is covered; Collect the original ultrasound signal at a fixed sampling frequency and convert it into a digital image sequence; Real-time monitoring of the signal-to-noise ratio of the image sequence dynamically adjusts the gain setting to maintain image quality stability.

[0008] Preferably, the inter-frame stability analysis of the real-time ultrasound image stream, and the selection of image frames with satisfactory clarity to form an optimized image set includes: Calculate the similarity measure between consecutive image frames, eliminate frames with severe motion artifacts, apply a clarity evaluation algorithm to score the remaining image frames, and retain image frames with scores above a threshold; The retained image frames are arranged in chronological order to form an optimized image set for subsequent processing.

[0009] Preferably, the multi-scale filtering process performed on the optimized image set to enhance the edge information of key tissues in the image includes: Gaussian pyramid decomposition is used to optimize the image set to obtain image layers at different resolutions, and adaptive threshold segmentation is performed on each image layer to enhance the tissue boundary profile; Fusion of multi-scale edge information generates a comprehensive enhanced image to highlight key anatomical structures.

[0010] Preferably, the step of detecting predefined anatomical landmarks from the enhanced image and recording the three-dimensional position data of the landmarks includes: Load the pre-stored anatomical template, perform template matching with the enhanced image, identify the center point of the successfully matched region as the anatomical landmark, and extract its pixel coordinates; By combining the spatial positioning data of the ultrasound probe, the pixel coordinates are converted into three-dimensional spatial coordinates.

[0011] Preferably, the step of establishing a dynamic spatial reference system using the three-dimensional position data of anatomical landmarks and calculating the center reference point of the target region includes: Multiple anatomical landmarks are selected as reference base points to construct a least-squares fitting plane; The direction of the reference frame is determined by the normal vector of the fitted plane, with the origin set at the geometric center of the marker point; Calculate the weighted average position of all pixels within the target area to obtain the coordinates of the center reference point.

[0012] Preferably, deriving the initial motion path of the guiding device based on the dynamic space reference frame and the central reference point includes: Input the current position of the guiding device into the dynamic space reference frame and calculate the relative displacement vector; Based on the constraints of the machine's motion, plan a smooth path from the current position to the central reference point; The smooth path is discretized into time series points to form the initial motion path data.

[0013] Preferably, the guiding device responds to the initial drive command to perform an initial positioning operation, while simultaneously capturing the real-time coordinates of the actual positioning point, including: The drive guide moves along the initial motion path and reads the encoder feedback position in real time; Use auxiliary sensors to measure the coordinates of the contact point between the instrument tip and the actual tissue; By fusing encoder data and sensor data, the real-time coordinates of the actual positioning point are obtained.

[0014] Preferably, the step of calibrating the dynamic spatial reference frame based on the position difference index to generate an optimized motion path includes: The Euclidean distance between the real-time coordinates of the actual positioning point and the central reference point is calculated as a position difference index. Adjust the rotation and translation parameters of the dynamic spatial reference system according to the position difference index; The path of the guiding device from its current position to the calibrated reference point is replanned to generate an optimized motion path.

[0015] Preferably, the present invention also includes a clinical anesthesia ultrasound image-assisted positioning and guidance system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the clinical anesthesia ultrasound image-assisted positioning and guidance method as described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By proactively performing inter-frame stability analysis on the real-time ultrasound image stream, image frames meeting the required clarity are selected to form an optimized image set, providing a high-quality and stable data foundation for subsequent processing. This pre-screening mechanism filters out blurred and noisy frames caused by physiological motion or operational interference, ensuring the uniformity and high signal-to-noise ratio of the input images. Anatomical landmark detection is performed based on the optimized image set, improving the accuracy and robustness of the feature recognition algorithm, avoiding interference from low-quality image data in the extraction of key tissue edge information, providing a more reliable and consistent data source for the subsequent establishment of a spatial reference system, and enhancing the overall system stability.

[0017] After the guided instrument performs initial positioning, the three-dimensional coordinates of its actual arrival point are captured in real time. These actual coordinates are then compared with the theoretical target center point calculated based on the image, generating a quantified positional difference index. Based on this difference index, the previously established dynamic spatial reference system is calibrated and corrected in real time, thereby generating an optimized motion path. This closed-loop feedback mechanism transforms one-time static registration into continuous dynamic calibration, enabling the navigation system to sense and compensate for positional deviations caused by factors such as soft tissue deformation, system mechanical errors, or initial registration bias. The dynamic spatial reference system remains synchronized with the actual anatomical position through iterative calibration, allowing the guided instrument to gradually approach the theoretical target point, ultimately achieving positioning accuracy far exceeding that of static navigation methods. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the clinical anesthesia ultrasound image-assisted positioning and guidance method described in this invention. Figure 2 A flowchart for acquiring a real-time ultrasound image stream; Figure 3 This is a flowchart of multi-scale filtering and edge enhancement. Figure 4 A schematic diagram of the three-dimensional spatial coordinate distribution of anatomical landmarks; Figure 5 A comparison chart showing the deviation between the initial and optimized motion paths of the guiding device. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a clinical anesthesia ultrasound image-assisted positioning and guidance method. The method includes: activating an ultrasound image acquisition device and aligning it with the target area of ​​the clinical anesthesia operation, continuously acquiring a real-time ultrasound image stream; performing inter-frame stability analysis on the real-time ultrasound image stream, selecting image frames with acceptable clarity to form an optimized image set; performing multi-scale filtering on the optimized image set to enhance the edge information of key tissues in the images; detecting predefined anatomical landmarks from the enhanced images and recording the three-dimensional position data of the landmarks; establishing a dynamic spatial reference system using the three-dimensional position data of the anatomical landmarks, and calculating the central reference point of the target area; deriving the initial motion path of the guiding device based on the dynamic spatial reference system and the central reference point; generating preliminary driving commands based on the initial motion path and transmitting them to the guiding device; the guiding device responding to the preliminary driving commands to perform preliminary positioning operations, while simultaneously capturing the real-time coordinates of the actual positioning point; comparing the real-time coordinates of the actual positioning point with the central reference point to generate a position difference index; calibrating the dynamic spatial reference system based on the position difference index to generate an optimized motion path; and generating a final driving command based on the optimized motion path to control the guiding device to achieve precise positioning. This method improves the accuracy and reliability of anesthesia operations by combining image processing, spatial modeling, and closed-loop control.

[0021] Ultrasound image acquisition equipment typically includes an ultrasound probe, a signal processing unit, and an image output interface. After the equipment is started, the operator places the probe near the target area on the patient's body surface, adjusts the probe orientation so that its sound beam covers target structures such as neurovascular bundles or the spinal canal, and the equipment continuously captures raw ultrasound signals at a preset frame rate and converts them into a digital image sequence. The image stream is transmitted to the processing unit in real time through a high-speed data interface. After receiving the image stream, the inter-frame stability analysis module calculates the similarity measure between consecutive frames, such as using normalized cross-correlation or structural similarity index to assess changes in image content, and removes blurred frames caused by patient movement or probe jitter. The remaining frames are scored using sharpness evaluation algorithms such as the Tenengrad gradient method or the Laplacian variance method, and only frames with scores higher than a threshold are retained and arranged in chronological order to form an optimized image set. Multi-scale filtering processing uses Gaussian pyramid decomposition to optimize the image set, generating image layers with multiple resolution levels. An adaptive threshold segmentation algorithm is applied to each layer to enhance tissue boundaries, and then edge information is fused through inverse transformation to generate a comprehensive enhanced image, highlighting the contours of key structures such as nerves and blood vessels. The anatomical landmark detection module loads pre-stored templates, such as typical image patterns of transverse processes, spinous processes, or nerve sheaths, and performs normalized cross-correlation matching with the enhanced image. It identifies the center points of regions with high matching degrees as landmarks and, combined with the six-DOF data provided by the probe spatial locator, converts pixel coordinates into three-dimensional spatial coordinates. In the dynamic spatial reference system establishment process, multiple landmarks are selected as base points. A plane is fitted using principal component analysis or least squares, and the reference system direction is defined by the plane normal vector. The origin is set at the geometric center of the landmarks. The gray-weighted average position of the target region's pixels is then calculated as the central reference point. The initial motion path derivation inputs the current pose of the guiding device into the reference system, calculates the displacement vector to the central reference point, considers device joint limitations and motion smoothness constraints, and uses B-splines or Bézier curves to plan the path and discretize it into time series points. The initial drive command is converted into pulse signals or bus commands and sent to the device controller, driving the motor or hydraulic actuator to move the device. Simultaneously, an optical encoder or electromagnetic sensor provides real-time feedback of the device's end-effector coordinates, which are compared with the central reference point to generate Euclidean distance as a difference index. The calibration module adjusts the rotation matrix and translation vector of the reference system according to the difference index, replans the path to generate an optimized motion path, and finally drives the command to control the device to gradually approach the target to achieve sub-millimeter positioning.

[0022] Example 1: See Figure 2In practical implementation, the clinical anesthesia ultrasound image-assisted positioning and guidance method starts the ultrasound image acquisition device and aligns it with the target area of ​​the clinical anesthesia operation, continuously acquiring real-time ultrasound image streams. This includes adjusting the incident angle and depth parameters of the ultrasound probe. The incident angle of the ultrasound probe is finely adjusted through a robotic arm system or a manual operation interface. The goal of adjusting the incident angle is to make the central axis of the ultrasound beam perpendicularly aligned with the center point of the target anatomical structure, such as a nerve plexus or blood vessel wall, to minimize sound wave refraction and scattering effects. The depth parameter is set according to the estimated anatomical depth of the target area. The depth parameter selection covers the complete range from the skin surface to a certain safe distance behind the target structure, ensuring that the image includes all relevant tissue layers. The ultrasound image acquisition device acquires the raw ultrasound signal at a fixed sampling frequency, which is usually set between 15 frames per second and 30 frames per second to meet real-time requirements. The raw ultrasound signal is processed by a beamformer, which uses a delay summation algorithm to focus the sound wave energy. Then, the signal amplitude is extracted by an envelope detector, and after logarithmic compression and dynamic range adjustment, it is finally converted into a B-mode digital image sequence. The digital image sequence is cached in memory in bitmap format and transmitted to the image processing pipeline in real time. The system monitors the signal-to-noise ratio (SNR) of image sequences in real time. SNR monitoring is achieved by calculating the ratio of the standard deviation to the average value of pixel intensity in a selected uniform region of the image. When the SNR is lower than a preset threshold, the system dynamically adjusts the gain settings. Gain adjustment includes time gain compensation and lateral gain control. Time gain compensation gradually increases the gain value as the depth increases to compensate for sound wave attenuation. Lateral gain control adjusts the gain distribution of the lateral scan lines to balance image brightness. The gain adjustment process is based on a feedback loop, continuously sampling image data and updating gain parameters to maintain image quality stability.

[0023] In some embodiments, the incident angle adjustment of the ultrasound probe can be combined with an optical tracking system. This system provides real-time position and orientation data of the probe in three-dimensional space. The incident angle is automatically calibrated based on a predefined ideal incident vector. Depth parameters can be adaptively set by querying a patient anatomy database, which stores typical depth values ​​corresponding to different body types, ensuring personalized parameters. Optionally, the fixed sampling frequency can be dynamically adjusted based on the image content. For example, the sampling rate can be reduced to 10 frames per second during slow tissue movement to reduce computational load, and increased to 30 frames per second during fast movement to capture more detail. This dynamic adjustment is based on calculations of inter-frame differences. It is understood that the real-time monitoring of the signal-to-noise ratio is integrated into the image acquisition hardware, using a dedicated integrated circuit for rapid pixel statistics to achieve millisecond-level response.

[0024] In practical implementation, inter-frame stability analysis is performed on the real-time ultrasound image stream. Image frames meeting the required sharpness are selected to form an optimized image set. This includes calculating the similarity metric between consecutive image frames using a normalized cross-correlation method. This method calculates the correlation coefficient of pixel values ​​in the overlapping region of two images. Frames with correlation coefficients below a set threshold are identified as frames with severe motion artifacts and discarded. Frames with severe motion artifacts are typically caused by patient breathing, heartbeat, or probe jitter. The remaining image frames are scored using a sharpness assessment algorithm. This algorithm uses the Tenengrad gradient method, which calculates the gradient magnitudes in the horizontal and vertical directions based on the Sobel operator. The sum of squared gradients is then used as the sharpness score. Image frames with sharpness scores above a preset threshold are retained. These retained image frames are arranged in timestamp order to form the optimized image set, which is used as input for subsequent image enhancement steps. The inter-frame stability analysis module employs a sliding window mechanism to continuously process the latest input image frame sequence. The window size can be configured from 5 to 10 frames to balance latency and stability.

[0025] In some embodiments, similarity measurement can be combined with the structural similarity index method. The structural similarity index method evaluates the similarity of image brightness, contrast, and structural information, improving motion detection accuracy. The sharpness evaluation algorithm can employ the Laplacian variance method, which calculates the variance value of the image after Laplacian filtering; a higher variance value indicates a sharper image. Optionally, optimizing the formation of the image set can introduce a priority mechanism, prioritizing the retention of frames with the highest sharpness scores to ensure optimal set quality. The sharpness score threshold is dynamically adjusted based on historical image data to adapt to different operating environments. It can be understood that inter-frame stability analysis is implemented as a parallel computing task, utilizing the graphics processor to simultaneously process the similarity and sharpness calculations of multiple frames, improving processing efficiency.

[0026] In practical implementation, the startup process of the ultrasound image acquisition equipment includes initializing the probe drive circuit and signal processing unit. The probe drive circuit generates a high-voltage pulse to excite the ultrasound transducer to emit sound waves. The signal processing unit receives the echo signal and performs analog-to-digital conversion. The digital signal is then processed by a digital signal processor to perform beamforming. Beamforming uses dynamic focusing technology, adjusting the focusing delay according to different depths to improve image resolution. The incident angle is adjusted by the servo motor of the robotic arm. The servo motor rotates the probe according to the input angle value, which is provided by the operator or the automatic planning system. Depth parameters are set by changing the detection range knob of the ultrasound equipment or through the software interface. The detection range is displayed in millimeters to ensure coverage of the target area. The fixed sampling frequency is synchronized by the system clock. The clock signal triggers the analog-to-digital converter to sample at fixed intervals. The sampled data is stored in a first-in-first-out buffer, which is large enough to hold several seconds of image stream to prevent data loss. Real-time monitoring of the signal-to-noise ratio uses a software algorithm to periodically analyze image frames. The algorithm selects the central region of the image as a uniform region and calculates the coefficient of variation of the pixel values ​​in this region. When the coefficient of variation exceeds a threshold, gain adjustment is triggered. The gain adjustment command is sent to the gain controller via a digital bus, and the gain controller modifies the amplifier parameters.

[0027] In some embodiments, the incident angle adjustment of the ultrasound probe can be integrated with an inertial measurement unit (IMU). The IMU provides probe tilt angle data, and the incident angle is compensated for by combining the tilt angle to improve alignment accuracy. The depth parameter can be automatically identified using anatomical landmarks in the ultrasound image, such as determining tissue boundaries through edge detection, and the depth can be dynamically set. Optionally, the fixed sampling frequency can be locked with an external synchronization signal, such as an electrocardiogram (ECG) signal, to achieve sampling at a specific phase of the cardiac cycle, reducing the influence of cardiac motion. It is understood that the real-time signal-to-noise ratio (SNR) monitoring process employs a multi-region averaging method, dividing the image into multiple sub-regions, calculating the SNR of each sub-region, and then averaging the results to improve monitoring robustness.

[0028] In its implementation, the inter-frame stability analysis module reads continuous frame data from the image stream buffer. When calculating the similarity metric, the normalized cross-correlation method divides the current frame and the previous frame into multiple small blocks, calculates the correlation coefficient of each block, and then takes the average as the overall similarity. The similarity threshold is set according to the operation type; for example, a higher threshold is used in static operations to tolerate minor motion, and frames with severe motion artifacts are marked as invalid and removed from the stream. The sharpness evaluation algorithm uses the Tenengrad gradient method to calculate the image gradient using a 3x3 Sobel convolution kernel. The sum of squared gradients is quickly calculated using integral image technology. The sharpness score is normalized to the range of 0-1, and the threshold is set above 0.5 to ensure that only high-quality frames are retained. The optimized image set is maintained using a linked list data structure. When a new frame is added, its sharpness is checked, and if it meets the conditions, it is inserted at the end of the linked list. The linked list capacity is limited to the latest 20 frames to prevent memory overflow. The inter-frame stability analysis is periodically synchronized with image acquisition. The analysis process is triggered when each new frame arrives, and the analysis results update the optimized image set in real time.

[0029] Example 2: See Figure 3In the specific implementation, the multi-scale filtering process performs Gaussian pyramid decomposition on the optimized image set. The optimized image set consists of image frames with acceptable sharpness output by the inter-frame stability analysis module. The Gaussian pyramid decomposition process first applies a Gaussian kernel to each frame in the optimized image set for low-pass filtering. The size of the Gaussian kernel is set according to the image resolution, usually using a 5x5 or 7x7 convolution window. The variance parameter of the Gaussian kernel is adjusted based on the image noise level. The low-pass filtered image is downsampled with a downsampling factor of 2 to generate multiple resolution layers. For example, starting from the original image size of 640x480, the pyramid structure includes progressively decreasing resolution layers such as 320x240 and 160x120. Each resolution layer is processed independently to capture edge features at different scales. The low-pass filtering and downsampling operations are repeated until the image size reaches the preset minimum threshold to ensure that the pyramid covers multi-scale information from fine-grained to coarse-grained. Adaptive thresholding is performed within each resolution layer. It employs a local window statistical method, where the local window size is proportional to the resolution layer size. For example, a smaller window (11x11 pixels) is used in high-resolution layers, while a larger window (21x21 pixels) is used in low-resolution layers. The local window statistical method calculates the mean and standard deviation of pixels within the window. Then, the Sauvola algorithm is applied to dynamically calculate the threshold based on the mean and standard deviation. The Sauvola algorithm adapts to different contrast regions by adjusting parameters, strengthening the boundaries of low-contrast regions. After segmentation, a binary image is generated. The binary image uses either the Canny or Sobel operator to extract edge contours. The Canny operator uses double thresholding to detect strong and weak edges, while the Sobel operator calculates the gradient magnitude. After edge extraction, a binarized edge map is generated, and the edge map for each resolution layer is stored as independent data. Multi-scale edge information fusion employs a weighted stacking method, assigning weights based on the importance of resolution layers. Higher-resolution edge maps retain more details and are assigned higher weights, such as 0.5. Lower-resolution edge maps provide robustness, with weights decreasing layer by layer, such as 0.25 and 0.125. After weight allocation, the edge maps of all layers are summed at the pixel level to generate a preliminary fused image. The preliminary fused image is then filled with edge gaps using a morphological closing operation. This morphological closing operation uses circular or square structuring elements, with the size of the structuring elements set according to the edge width. After the closing operation, a comprehensive enhanced image is generated. This comprehensive enhanced image highlights the continuous contours of key tissues such as the perimembranes or blood vessel walls. The comprehensive enhanced image is used as output for subsequent anatomical landmark detection.

[0030] In some embodiments, Gaussian pyramid decomposition can be replaced by Laplacian pyramid decomposition. Laplacian pyramid captures edge information by calculating the differences between Gaussian pyramid layers. Adaptive threshold segmentation can use the NiBlack algorithm, which dynamically adjusts the threshold based on local mean and standard deviation. Optionally, multi-scale edge information fusion can introduce multi-resolution pyramid reconstruction technology, synthesizing and enhancing the image through inverse transform. Weight allocation can be adaptively adjusted according to image content, for example, by dynamically calculating weights based on edge density. It can be understood that the multi-scale filtering processing module is implemented as a parallel computing pipeline, utilizing multi-core processors to process different resolution layers simultaneously, thereby improving processing efficiency.

[0031] In practice, anatomical landmark detection uses pre-stored anatomical templates, extracted from a historical image database containing numerous ultrasound images acquired during clinical anesthesia procedures. Each template represents a typical structure of the target region, such as a triangular hyperechoic area of ​​the intervertebral foramen or an elliptical hypoechoic pattern of a nerve trunk. The template images are normalized to ensure consistent size and contrast, and stored in a template database as a numerical matrix. This database supports querying by anatomical structure type. Template matching uses normalized cross-correlation as the matching metric. Normalized cross-correlation calculates the correlation coefficient between the template and a sliding window in the enhanced image. The sliding window size is consistent with the template size, with a step size of 1 or 2 pixels to ensure coverage. The matching process slides the window pixel-by-pixel across the enhanced image, calculating the correlation coefficient at each location and generating a correlation coefficient map. The correlation coefficient map identifies local maxima through peak detection, using these local maxima as candidate landmarks. These candidate landmarks undergo non-maximum suppression (NMS), comparing correlation coefficient values ​​within their neighborhood, retaining the maximum value, and eliminating overlapping candidate points. Finally, the center point of the successfully matched region is determined, and this center point is recorded as the image coordinates of the anatomical landmark. The three-dimensional position data recording is combined with the spatial positioning system of the ultrasound probe. The spatial positioning system uses an electromagnetic tracker or an optical tracker. The electromagnetic tracker detects changes in the magnetic field through a sensor coil, while the optical tracker uses an infrared camera to capture reflective markers. The spatial positioning system provides the ultrasound probe's six-degree-of-freedom position and attitude data in the global coordinate system. The position and attitude data are represented in the form of a transformation matrix. The image pixel coordinates of the anatomical landmarks are mapped to the probe coordinate system through inverse perspective transformation, and then transformed to global three-dimensional spatial coordinates through the probe transformation matrix. The coordinate data is stored in floating-point format and is accompanied by a timestamp and landmark identifier for subsequent establishment of a dynamic spatial reference system.

[0032] In some embodiments, anatomical templates can be generated based on machine learning models, such as convolutional neural networks, which learn feature representations from training data. Template matching can employ mutual information or feature matching methods instead of normalized cross-correlation. Optionally, 3D position data recording can integrate inertial measurement unit (IMU) data, which provides probe orientation compensation to improve coordinate accuracy. Non-maximum suppression can be combined with geometric constraints, such as retaining only points that conform to anatomical relationships. It is understood that the anatomical landmark detection module is implemented as a real-time processing loop, triggering detection immediately upon the arrival of a new enhanced image to ensure data timeliness.

[0033] In practical implementation, the Gaussian pyramid decomposition for multi-scale filtering is implemented by inputting the optimized image set in list form. Each frame in the list is processed sequentially. Gaussian kernel convolution uses a separable filter for optimized calculation, performing horizontal convolution first and then vertical convolution to reduce computational complexity. Downsampling operations use nearest neighbor or bilinear interpolation methods to maintain image smoothness. Images at each resolution level are stored in multi-level caches, with cache size dynamically allocated according to pyramid depth. The local window statistical method for adaptive thresholding is executed in parallel on the GPU, with local statistics calculated independently for each pixel. The parameters in the Sauvola algorithm formula are preset based on the global contrast of the image. After thresholding, the binary image is analyzed using connected component analysis to remove noise points. Edge extraction operators such as the Canny operator automatically set high and low thresholds based on the image gradient histogram to ensure edge continuity. The weighted superposition of multi-scale edge information fusion is performed in the frequency domain, converted to the frequency domain by Fast Fourier Transform, and then subjected to inverse weighted transform to improve fusion speed. Morphological closing operations use a multi-scale structuring element sequence to gradually fill gaps of different sizes. Contrast stretching is performed before the overall enhanced image output to enhance visual effects.

[0034] The template loading process for anatomical landmark detection involves a template database index, which is quickly retrieved based on anatomical structure names and image feature hashes. Normalized cross-correlation calculation for template matching is accelerated using integral images. After pre-calculation of the integral images, the sum within the window is quickly calculated. After the correlation coefficient map is generated, Gaussian smoothing filtering is applied to remove noise peaks. The neighborhood size for non-maximum suppression is adjusted according to the template size, typically set to 1.5 times the template width and height. Coordinate transformation for 3D position data recording uses homogeneous coordinate matrix multiplication. Matrix operations are optimized using a hardware acceleration library. Timestamps are synchronized with the image acquisition clock. Landmark identifiers include type encoding and confidence scores, with the confidence score calculated based on the matching correlation coefficient.

[0035] Optionally, multi-scale filtering can introduce wavelet transform to replace Gaussian pyramids. Wavelet transform provides more flexible frequency band division, and adaptive threshold segmentation can be combined with deep learning segmentation networks to directly output edge maps. Anatomical landmark detection can use keypoint detection algorithms such as SIFT or ORB, extracting feature points and then performing descriptor matching. 3D coordinate transformation can incorporate distortion correction to compensate for probe lens errors. In essence, the entire processing chain is designed with a modular architecture, with modules communicating through standard interfaces and supporting plug-in algorithm replacement.

[0036] In practice, the parameter configuration for multi-scale filtering is managed through a configuration file. This file specifies parameters such as the number of pyramid layers, Gaussian kernel size, and threshold algorithm type. These parameters are preset based on the operating environment; for example, the Gaussian kernel variance is increased in noisy images. The window size for adaptive thresholding is proportional to the image resolution to ensure local adaptability. The binarized edge map after edge extraction undergoes thinning processing, with the thinning algorithm reducing edge width while preserving single-pixel edges. The weight allocation curve for multi-scale fusion can be customized, for example, using exponential or linear decay. Post-processing of the fused image includes histogram equalization to enhance overall contrast.

[0037] The template database for anatomical landmark detection supports online updates. New templates are added through expert annotation. Multiple templates are matched in parallel during template matching, and the template with the highest correlation coefficient is selected. Post-processing of candidate landmarks includes cluster analysis, using algorithms such as DBSCAN to group neighboring points, and selecting cluster centers as the final landmarks. Uncertainty estimation is added during 3D coordinate recording, and the uncertainty is calculated based on a positioning error model. Data from the spatial positioning system is smoothed using Kalman filtering to reduce jitter effects. The coordinate data format adopts the standard ROS or DICOM protocol for easy integration.

[0038] See Figure 4 This is a scatter plot, with the horizontal axis representing the X-coordinate of the anatomical landmarks, the vertical axis representing the Y-coordinate, and color intensity corresponding to the Z-coordinate, visually presenting the three-dimensional spatial distribution of anatomical landmarks. Points at different locations and in different colors correspond to landmarks of predefined anatomical structures such as the intervertebral foramen and nerve trunks, demonstrating both the distribution pattern of landmarks on the ultrasound image plane and distinguishing their depth differences through color. These three-dimensional coordinates are the core basis for subsequently establishing a dynamic spatial reference system; their accuracy directly determines the precision of the reference system construction and is crucial data support for achieving precise instrument positioning in anesthesia ultrasound navigation.

[0039] Example 3: In specific implementation, a dynamic spatial reference system is established using the three-dimensional position data of anatomical landmarks. The three-dimensional position data of the anatomical landmarks comes from the coordinates of predefined anatomical points detected and recorded in the previous stage. For example, in lumbar anesthesia, the anatomical landmarks can be the tips of the transverse processes and the apex of the spinous processes of two adjacent lumbar vertebrae. These three points form a triangular plane. Multiple anatomical landmarks are selected as reference base points, with at least three reference base points that are not collinear. A least-squares fitting plane is constructed. The least-squares fitting plane is achieved by solving the problem of minimizing the sum of the squares of the distances from the point set to the plane. The general form of the plane equation is Ax + By + Cz+D=0. The eigenvectors of the point set covariance matrix are calculated; the eigenvector corresponding to the smallest eigenvalue is the normal vector of the plane. The normal vector of the fitted plane determines the direction of the reference frame, which is defined as the Z-axis direction of the dynamic space reference frame. The X-axis direction is determined by the direction of the first principal component of the point set's principal component analysis, or by defining it through the direction of the line connecting two specific anatomical landmarks. The Y-axis direction is derived from the cross product of the Z-axis and X-axis, following the right-hand coordinate system rule. The origin of the dynamic space reference frame is set at the geometric center of all anatomical landmarks involved in the fitting, obtained by calculating the arithmetic mean of all point coordinates. The center reference point of the target region is calculated. The target region in ultrasound images is represented as a specific connected region in the enhanced image, such as a cross-sectional area of ​​a nerve plexus or blood vessel. The weighted average position of all pixels within the target region is calculated, with weights determined based on the pixel's grayscale value. Higher grayscale values ​​indicate stronger tissue echoes and a greater contribution to the weighted average. The coordinates of the center reference point are obtained using the following formula: in: A three-dimensional coordinate vector representing the central reference point. This represents the total number of pixels within the target area. This represents the three-dimensional spatial coordinates corresponding to the i-th pixel. The weight represents the weight of the i-th pixel. The weighted average is obtained by normalizing the gray value of the pixel, for example, by directly using the gray value or the square of the gray value as the weight. The weighted average is calculated in the dynamic space reference system, and the resulting coordinates are represented as the offset vector relative to the origin of the dynamic space reference system.

[0040] In some embodiments, the selection of reference points may include four or more anatomical landmarks. A random sampling consensus algorithm is used to robustly fit the plane, reducing the influence of outliers. The axial definition of the dynamic spatial reference system can adopt all three principal component directions of principal component analysis, respectively serving as the X, Y, and Z axes. Optionally, a distance attenuation factor can be introduced in the weight calculation of the central reference point, assigning higher weights to pixels closer to the geometric center. The weighted average calculation can be performed in the image coordinate system and then transformed to three-dimensional space. It can be understood that the process of establishing the dynamic spatial reference system is repeated periodically, and the reference system parameters are updated in real time as new anatomical landmark data arrives.

[0041] In practice, the initial motion path of the guiding device is derived based on the dynamic spatial reference frame and the central reference point. The current position of the guiding device is input into the dynamic spatial reference frame. The current position of the guiding device is fed back by encoders installed on the device base or joints, or by the coordinates of the marker point at the device's end effector captured in real time by an optical tracking system. The current position data is transformed from the global coordinate system to the dynamic spatial reference frame through coordinate transformation. The relative displacement vector from the current position to the central reference point is calculated. The relative displacement vector is a three-dimensional vector representing the direction and distance from the current position of the device to the central reference point. Combined with the device's motion constraints, including joint angle limits of the robotic arm, link length limits, maximum motion speed, and acceleration limits, a smooth path from the current position to the central reference point is planned. The smooth path planning uses polynomial interpolation algorithms or spline curve fitting algorithms, such as using cubic spline curves to connect the start and end points, ensuring the continuity of the first and second derivatives of the path, thereby achieving smooth motion. The smooth path needs to avoid known obstacle areas, and obstacle information is identified from preoperative images or real-time ultrasound images. The smooth path is discretized into time series points. The time series points sample the continuous path at fixed time intervals. Each time point corresponds to the coordinates of a spatial point on the path and the posture of the instrument end effector. The posture of the instrument end effector is represented by quaternions or Euler angles to form the initial motion path data. The initial motion path data is stored as an array or linked list structure and used to drive the instrument controller.

[0042] In some embodiments, the calculation of the relative displacement vector can take into account the orientation of the instrument's end effector, incorporating directional deviations into path planning. Smooth path planning can utilize Bézier curves or B-spline curves, providing more control points to adjust the path shape. Optionally, the interval between time series points can be dynamically adjusted based on the instrument's maximum speed; the interval increases during high-speed motion and decreases during fine positioning. Inverse kinematics calculations can be incorporated into the discretization process to directly output the joint angle sequence. It is understood that the initial motion path derivation module is tightly coupled with the real-time positioning system, and path data is transmitted to the motion control unit via shared memory or a message queue.

[0043] In practical implementation, the establishment of the dynamic spatial reference system involves storing the 3D position data of anatomical landmarks in a list format. Each point contains x, y, and z coordinates and a confidence score. The selection of reference points is based on confidence score sorting, prioritizing points with higher scores. The least squares fitting plane uses singular value decomposition (SVD) to solve for the plane equation coefficients. SVD processes the centered coordinate matrix of the point set to obtain a stable numerical solution. The coordinate axis directions of the dynamic spatial reference system are ensured to be orthogonal through Gram-Schmidt orthogonalization. The origin coordinates are precisely calculated using floating-point arithmetic. When calculating the central reference point, pixels in the target region are extracted from the enhanced image using region growing or thresholding. The 3D coordinates of each pixel are obtained by backtracking its image coordinates and the probe transformation matrix. Weight calculation uses the difference between the pixel grayscale value and the image background grayscale value to avoid the influence of background noise. Weighted summation uses double-precision floating-point numbers to ensure accuracy.

[0044] In the initial motion path derivation of the guided device, a homogeneous transformation matrix is ​​used to transform the device's current position from the world coordinate system to the dynamic space reference system, and the relative displacement vector is decomposed into components along each axis of the reference system. The device's motion constraints are read from the device configuration file, including the range of motion limits and maximum velocity values ​​for each joint. Smooth path planning uses fifth-order polynomial interpolation to satisfy the position, velocity, and acceleration boundary conditions at the start and end points. During path discretization, the time interval is set according to the system control cycle, typically between 10 and 50 milliseconds. Each time series point contains position and attitude data, as well as a timestamp for that point. The initial motion path data is sent to the motion controller after verification and integrity checks.

[0045] Optionally, a timestamp can be added to the dynamic spatial reference frame for synchronization with the image stream, and the calculation of the center reference point can fuse multi-frame image data, smoothing the coordinate sequence through Kalman filtering. A virtual force field algorithm can be incorporated into the initial motion path planning to automatically avoid obstacle areas. It is understandable that the entire implementation relies on high-precision timing and coordinate synchronization to ensure the consistency of spatial positioning and motion control.

[0046] Example 4: In specific implementation, the guiding device responds to the initial drive command to perform the initial positioning operation, while simultaneously capturing the real-time coordinates of the actual positioning point. The initial drive command is generated based on the initial motion path, which is a time series of points derived from the dynamic space reference system and the central reference point. The initial drive command is transmitted to the control unit of the guiding device through a digital communication interface. The digital communication interface can use CAN bus, EtherCAT, or serial communication protocol. The command format includes target position, target attitude, motion speed, and acceleration parameters. After receiving the command, the control unit of the guiding device parses it into low-level control signals. The control unit drives the servo motor or stepper motor to move. The servo motor drives the device arm to move through the reducer and transmission mechanism. The device arm consists of multiple joints, each equipped with a high-precision encoder. The encoder provides real-time feedback of joint angle data. The control unit calculates the target angle of each joint according to the inverse kinematics algorithm and uses a PID control algorithm to adjust the motor output, so that the device end effector moves smoothly along the initial motion path. During movement, the system synchronously captures the real-time coordinates of the actual positioning point, which refers to the spatial position of the contact point between the instrument tip and the patient's tissue. Real-time coordinate capture is achieved through multi-sensor fusion, primarily relying on encoders and auxiliary sensors mounted at the joints. The encoder provides feedback on the joint angle, and the theoretical coordinates of the instrument tip are calculated using forward kinematics. Auxiliary sensors, such as optical trackers or electromagnetic sensors, directly measure the coordinates of the instrument tip in the global coordinate system. The optical tracker uses an infrared camera to capture reflective markers fixed to the instrument tip, while the electromagnetic sensor detects the position and orientation of the sensor coil through magnetic field induction. Auxiliary sensor data is sampled at high frequencies, typically 100Hz to 1000Hz, to ensure real-time performance. Encoder and sensor data are combined using a data fusion algorithm. Data fusion employs Kalman filtering or complementary filtering methods to reduce measurement noise and latency, yielding high-precision real-time coordinates of the actual positioning point. The real-time coordinates are timestamped and synchronized with the image stream.

[0047] In some embodiments, the initial drive command may include force feedback parameters. When the instrument tip contacts the tissue, the force sensor detects the contact force and adjusts the motion parameters to avoid excessive pressure. The auxiliary sensor may be an ultrasonic ranging module or a laser rangefinder to directly measure the distance between the tip and the tissue. Optionally, data fusion can be extended to distributed filtering, where each sensor independently filters and then fuses the data to improve robustness. Real-time coordinate data can be compressed during transmission to reduce bandwidth consumption. It is understood that the entire capture process is part of closed-loop control, and the real-time coordinates are used for subsequent comparison and calibration.

[0048] In practical implementation, during the initial positioning operation of the guided device, the control unit reads the first time-series point from the initial motion path data. This time-series point contains position coordinates (x, y, z) and attitude quaternions (qx, qy, qz, qw). The control unit solves for joint angles using inverse kinematics, employing numerical iterative methods such as the Jacobian transpose or analytical methods. Joint angle commands are sent to the motor driver, which outputs a PWM signal to control motor rotation. The encoder provides real-time feedback of the actual joint angles, comparing them with the target angles to generate an error signal. The PID controller adjusts the output based on the error's proportional, integral, and derivative terms, causing the device's end effector to move point by point along the path. During this movement, encoder data is read at a fixed frequency, for example, sampling every 1 millisecond. The encoder resolution is typically 1000 to 10000 lines per revolution, providing high-precision angle feedback. Forward kinematics calculations use the DH parameter model to convert joint angles into theoretical end-effector coordinates, which are represented in the device's base coordinate system.

[0049] Taking an optical tracking system as an example, the measurement process of auxiliary sensors involves an infrared camera capturing the scene at high frequency. Image processing algorithms identify reflective markers at the end effector's end effector. These markers typically consist of three or more non-contiguous reflective spheres. The three-dimensional coordinates of each marker are calculated using triangulation principles. Then, the six-degree-of-freedom pose of the end effector is calculated based on the geometric relationships of the markers, and the pose data is converted to a global coordinate system. Electromagnetic sensors generate a low-frequency magnetic field using a magnetic field generator. The sensor coil at the end effector senses changes in the magnetic field, and the output voltage signal is calculated to determine the position and orientation. Sensor data is smoothed using a Kalman filter. The filtered state equation includes position, velocity, and acceleration, while the observation equation is based on the sensor model, and the filtered output is the actual coordinates. Data fusion employs an extended Kalman filter. The state vector includes the actual position and velocity of the positioning point, and the observation vector consists of the encoder's theoretical coordinates and the sensor's measured coordinates. The fusion formula is as follows: in: The state estimation vector at time k contains the three-dimensional position and three-dimensional velocity of the actual location point; The prior state estimate at time k is predicted from the state at time k-1 using the state transition equation; This represents the Kalman gain matrix, calculated based on the prediction error covariance. The observation vector at time k is composed of encoder and sensor data; The observation matrix represents the state mapping to the observation space. The real-time coordinates of the actual positioning points after fusion are stored in the form of a three-dimensional vector and published to the data bus.

[0050] In practical implementation, the transmission protocol for initial drive commands needs to define a message format, including a start symbol, message length, command type, data payload, and checksum. The control unit receives and parses the message, checking its validity, such as whether the location is within the instrument's workspace. The instrument's acceleration and velocity curves employ an S-shaped acceleration / deceleration design to reduce impact. Encoder feedback data is acquired via a high-speed counter, and the counter value is converted into an angle value. This angle value undergoes temperature compensation and nonlinear correction to improve accuracy. Auxiliary sensors require calibration. The calibration process involves moving the instrument to a known location, comparing sensor readings with actual values, calculating the transformation matrix, and synchronizing sensor data timestamps with the system clock using a network time protocol or hardware trigger signal.

[0051] The Kalman filter implementation for data fusion requires initializing the state vector and covariance matrix. The state vector is initialized with the first observation, and the covariance matrix is ​​set based on the sensor accuracy. The filtering period is consistent with the sensor sampling period, for example, a prediction and update step is performed every 10 milliseconds. The uncertainty of the fused coordinates is evaluated using the covariance matrix, and recalibration is triggered when the uncertainty exceeds a threshold. The real-time coordinate data format adopts a standard structure, including coordinate values, timestamps, confidence levels, and sensor source identifiers, as shown in Table 1. Table 1: Real-time coordinate data format table Field name Data type Description Coordinate X Floating point number Actual positioning point in global coordinate system X axis coordinate Coordinate Y Floating point number Actual positioning point in global coordinate system Y axis coordinate Coordinate Z Floating point number Actual positioning point in global coordinate system Z axis coordinate Timestamp Integer Microsecond-level timestamp of data acquisition Confidence Floating point number Confidence level of coordinate estimation, between 0 and 1 Sensor source Enumeration Identify data source, such as encoder or optical sensor In some embodiments, the guiding device can be hydraulically driven instead of motor-driven. The hydraulic cylinder controls the flow rate via a servo valve, the encoder provides feedback on the piston displacement, and an auxiliary sensor can be integrated with a strain gauge at the device's end to directly measure the contact force and calculate the deformation coordinates. Optionally, data fusion can use particle filtering to process nonlinear motion, adapting to more complex scenarios, and real-time coordinate data can be recorded in a circular buffer for historical backtracking. It is understood that implementation details depend on hardware selection; for example, optical tracking is used in high-precision applications, while ultrasonic sensors are chosen in electromagnetically sensitive environments.

[0052] In practice, after the initial positioning operation of the guided instrument is initiated, the control unit monitors the motion status. If any abnormality is encountered, such as joint overload or excessive path deviation, a safety stop is triggered. Abnormal states are managed by a state machine, which includes idle, moving, paused, and error states. Real-time coordinate capture of the actual positioning point is synchronized with the motion. Coordinate data is transmitted via a real-time publish-subscribe model. Subscribers include a path comparison module and a user interface. The user interface displays the instrument's position superimposed on the ultrasound image, providing visual feedback. Encoder data acquisition uses a dedicated acquisition card to reduce the load on the main processor. Sensor data is input via USB or Ethernet interfaces, and timestamps are precisely marked by a hardware clock. Optionally, the initial drive command can include adaptive parameters to adjust the motion speed online according to tissue stiffness to avoid injury. Auxiliary sensors can be configured in multiple modes, such as simultaneously using optical and electromagnetic sensors, with a voting mechanism selecting the optimal data.

[0053] Example 5: In specific implementation, the real-time coordinates of the actual positioning point are compared with the central reference point to generate a position difference index. The real-time coordinates of the actual positioning point come from the spatial position data captured during the initial positioning operation of the guiding device. The central reference point is obtained by calculating the three-dimensional spatial coordinates of all pixels in the target area using weighted average position calculation. The comparison process calculates the Euclidean distance between the real-time coordinates of the actual positioning point and the central reference point as the position difference index. The Euclidean distance is obtained by calculating the square root of the sum of the squares of the coordinate differences between the two points in the three-axis coordinate system. The position difference index is a scalar value used to quantify the spatial deviation between the end of the guiding device and the center of the target area. The calculation formula for the position difference index is as follows: in: This represents the positional difference index, i.e., the Euclidean distance scalar value; Represents the coordinate components of the actual location point in the X-axis direction of the global coordinate system; Represents the coordinate components of the actual location point in the Y-axis direction of the global coordinate system; Represents the coordinate components of the actual positioning point in the Z-axis direction of the global coordinate system; The coordinate components of the central reference point in the X-axis direction of the global coordinate system; The coordinate components of the central reference point in the Y-axis direction of the global coordinate system; This represents the coordinate components of the central reference point along the Z-axis in the global coordinate system. The position difference index calculation module executes at a fixed frequency, typically synchronized with the sensor data update frequency, for example, calculating the latest difference value every 10 milliseconds. The difference value is then smoothed using a low-pass filter to eliminate high-frequency noise fluctuations, generating a stable and reliable position difference index.

[0054] In some embodiments, the positional difference index can be extended to a vector form including angular deviation, obtained by calculating the angle between the instrument end-effector direction vector and the target direction vector. The Euclidean distance can be calculated using Mahalanobis distance, considering the uncertainty weights of different coordinate axes. Optionally, the positional difference index can be combined with historical difference data to calculate a moving average, reducing the impact of random errors. A tolerance threshold can be introduced in the difference index calculation; when the difference value is less than the threshold, it is considered an acceptable error. It can be understood that the positional difference index is a feedback signal of the closed-loop control system, used to trigger subsequent calibration operations.

[0055] In practice, a dynamic spatial reference system is calibrated based on positional difference indices to generate an optimized motion path. The calibration process adjusts the rotation and translation parameters of the dynamic spatial reference system according to the magnitude and direction of the positional difference indices. When the positional difference indices exceed a preset threshold, the system determines that the dynamic spatial reference system needs calibration. The rotation parameters of the dynamic spatial reference system are represented by rotation matrices or quaternions, and the translation parameters are represented by translation vectors. Calibration employs an iterative nearest-point algorithm or a least-squares registration method. The iterative nearest-point algorithm registers the actual set of positioning points with the theoretical set of reference points. The theoretical set of reference points comes from the expected positions under the dynamic spatial reference system. The optimal rotation matrix and translation vector are calculated through singular value decomposition to minimize the root mean square error between the actual and theoretical point sets. The update formulas for the rotation matrix and translation vector are based on coordinate transformation principles. The new reference system is obtained by left-multiplying the rotation matrix and adding the translation vector. The updated origin position and coordinate axis directions of the calibrated dynamic spatial reference system more accurately reflect the spatial relationships of the actual anatomical structures. The optimized motion path is generated based on the re-planning of the calibrated dynamic spatial reference system. The optimized motion path planning starts from the current position of the guiding device, which is provided by the real-time coordinate capture system. The target point is the calibrated center reference point. The path planning takes into account the device's motion constraints and obstacle information, and uses spline interpolation to generate a smooth trajectory. The optimized motion path is discretized into time series points, each point containing spatial coordinates and device attitude. The optimized motion path data replaces the initial motion path and serves as the input for the final drive command.

[0056] In some embodiments, dynamic spatial reference frame calibration can adjust only the translation parameters while keeping the rotation parameters constant, simplifying computational complexity. Iterative nearest-neighbor algorithms can accelerate the use of kd-tree nearest neighbor search, improving registration speed. Optionally, optimizing motion path planning can employ model predictive control algorithms to optimize the motion trajectory for the next few steps. Path discretization can adaptively adjust the time interval, balancing accuracy and computational load. It can be understood that the calibration process is an incremental adjustment to avoid instrument vibration caused by abrupt changes in the reference frame.

[0057] In practical implementation, the generation of the position difference index includes coordinate system unification. The real-time coordinates of the actual positioning point and the central reference point must be transformed to the same coordinate system, usually the global coordinate system is selected as the reference. The coordinate transformation is achieved through a pre-calibrated transformation matrix. Euclidean distance calculation uses double-precision floating-point arithmetic to ensure numerical accuracy. The calculation results are filtered by a first-order low-pass filter, and the filtering time constant is set according to operational requirements, for example, 0.1 seconds to smooth rapid fluctuations. The position difference index is bound to a timestamp for storage and used for trend analysis. After the dynamic space reference system calibration module is triggered, it reads the actual positioning point coordinate sequence and the corresponding theoretical coordinate sequence within the most recent period. The theoretical coordinate sequence is obtained by transforming the path points under the initial dynamic space reference system. Point set registration uses the unit quaternion method to calculate the rotation matrix. The quaternion method constructs the covariance matrix of the point set and solves for the eigenvector corresponding to the largest eigenvalue as the rotation quaternion. The translation vector is calculated using the centroid difference of the point set. The calibration parameters are subjected to amplitude limiting to prevent over-adjustment. After verifying the rationality of the new reference system parameters, they are updated to the global parameter table.

[0058] In the motion path generation phase, the path planner queries the calibrated dynamic space reference frame parameters, transforms the target point to the new reference frame, and uses a cubic spline curve to connect the current position and the target point. Curve control points are automatically generated based on the device's motion constraints. During path discretization, the time interval is calculated based on the maximum permissible acceleration to ensure smooth motion. The reachability of each discrete point is verified through forward kinematics. The optimized motion path data is encoded into a message in a specific format and sent to the motion controller via a real-time communication bus. The entire calibration and path regeneration process is completed within tens of milliseconds, meeting real-time control requirements. Optionally, the position difference index can be combined with force sensor data; when the contact force is too large, the weight of the difference index is increased. Dynamic space reference frame calibration can introduce weighted registration, assigning higher weights to recent data points. Optimizing the motion path can employ stochastic path planning algorithms, such as the fast expanding random tree algorithm, to handle complex obstacle environments.

[0059] In practical implementation, the threshold setting for the positional difference index is based on clinical accuracy requirements. For example, epidural anesthesia requires millimeter-level accuracy, so the threshold is set to 1 millimeter. When the difference index exceeds the threshold, a calibration event is triggered, and a calibration log is recorded for quality assessment. When implementing the iterative nearest-point algorithm for dynamic spatial reference system calibration, point set preprocessing includes denoising and downsampling. The number of registration iterations is limited to 10 to avoid overfitting, and the registration error is used as a calibration quality assessment index. After optimizing the motion path generation, virtual simulation is performed to verify path safety, detecting joint limitations and collision risks. Only after successful verification is the path deployed for execution. During calibration, the user interface provides visual cues, displaying reference system changes and path update status. In some embodiments, the positional difference index calculation can integrate multi-target point information. When multiple key regions exist, a weighted average difference is calculated. Dynamic spatial reference system calibration can be performed in segments, using different calibration parameters for different anatomical regions. Optionally, the optimized motion path can include a velocity planning curve, dynamically adjusting the instrument's movement speed based on the path curvature. Path data can be compressed during transmission to reduce communication latency.

[0060] See Figure 5 This figure is a line graph comparing the two paths. The horizontal axis represents the time step of motion control, and the vertical axis represents the path deviation between the end effector and the target area. The two curves correspond to the initial path deviation and the optimized path deviation, respectively. This figure intuitively demonstrates the value of the closed-loop calibration mechanism: the initial path has a large deviation due to its difficulty in adapting to dynamic changes such as soft tissue deformation and probe displacement. However, the optimized path, through dynamic adjustment of the reference frame, compresses the deviation to the sub-millimeter level. This not only verifies the effectiveness of iterative calibration using the dynamic spatial reference frame but also provides data support for the precise positioning of instruments in clinical anesthesia operations, making it a key step in improving navigation accuracy.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of clinical anaesthesia ultrasound image assisted positioning guidance, characterized in that, The method sequentially performs the following steps: Start the ultrasound image acquisition device and align the target area of the clinical anesthesia operation, continuously acquire real-time ultrasound image stream; Perform inter-frame stability analysis on the real-time ultrasound image stream, select image frames with satisfactory sharpness to form an optimized image set; Implement multi-scale filtering processing on the optimized image set to enhance the edge information of key tissues in the image; Detect predefined anatomical landmark points from the enhanced image and record the three-dimensional position data of the landmark points; Establish a dynamic spatial reference frame using the three-dimensional position data of the anatomical landmark points and calculate the center reference point of the target area; Deduce the initial motion path of the guiding instrument according to the dynamic spatial reference frame and the center reference point; Form preliminary driving instructions based on the initial motion path and transmit them to the guiding instrument; The guiding instrument responds to the preliminary driving instructions to perform preliminary positioning operations, while capturing the real-time coordinates of the actual positioning points; Compare the real-time coordinates of the actual positioning points with the center reference point to generate a position difference index; Calibrate the dynamic spatial reference frame according to the position difference index to generate an optimized motion path; Generate final driving instructions based on the optimized motion path to control the guiding instrument to achieve precise positioning.

2. The clinical anesthetization ultrasound image assisted positioning guidance method according to claim 1, characterized in that, The start of the ultrasound image acquisition device and the alignment of the target area of the clinical anesthesia operation include: Adjust the incident angle and depth parameters of the ultrasound probe to ensure that the complete anatomical range of the target area is covered; Collect raw ultrasound signals at a fixed sampling frequency and convert them into digital image sequences; Real-time monitor the signal-to-noise ratio of the image sequence and dynamically adjust the gain setting to maintain image quality stability.

3. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 2, wherein, The inter-frame stability analysis of the real-time ultrasound image stream includes: Calculate the similarity measure between consecutive image frames, remove frames with severe motion artifacts, apply a sharpness evaluation algorithm to score the remaining image frames, and retain image frames with scores above a threshold; Arrange the retained image frames in chronological order to form an optimized image set for subsequent processing.

4. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 3, wherein, The multi-scale filtering processing of the optimized image set includes: Apply Gaussian pyramid decomposition to the optimized image set to obtain image layers at different resolutions, perform adaptive threshold segmentation on each image layer, and strengthen the tissue boundary profile; Fuse multi-scale edge information to generate a comprehensive enhanced image to highlight key anatomical structures.

5. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 4, wherein, The detection of predefined anatomical landmark points from the enhanced image and the recording of the three-dimensional position data of the landmark points include: Load the pre-stored anatomical template, perform template matching with the enhanced image, identify the region center point of the matching success as the anatomical landmark point, and extract its pixel coordinates; Convert the pixel coordinates to three-dimensional space coordinates in combination with the spatial positioning data of the ultrasound probe.

6. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 5, wherein, The establishment of a dynamic spatial reference frame using the three-dimensional position data of the anatomical landmark points and the calculation of the center reference point of the target area include: Select multiple anatomical landmark points as reference base points to construct a least squares fitting plane; Determine the reference frame direction with the normal vector of the fitting plane, and set the origin at the geometric center of the landmark points; Calculate the weighted average position of all pixel points in the target area to obtain the center reference point coordinates.

7. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 6, wherein, The initial motion path of the guiding instrument derived from the dynamic spatial reference frame and the center reference point comprises: inputting the current position of the guiding instrument into the dynamic spatial reference frame to calculate a relative displacement vector; planning a smooth path from the current position to the center reference point in combination with the motion constraint of the instrument; discretizing the smooth path into time series points to form initial motion path data.

8. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 7, wherein, The guiding instrument executes a preliminary positioning operation in response to a preliminary driving instruction, while capturing real-time coordinates of the actual positioning point comprises: driving the guiding instrument to move along the initial motion path and reading the encoder feedback position in real time; measuring the coordinates of the contact point between the tip of the instrument and the actual tissue using auxiliary sensors; fusing the encoder data and the sensor data to obtain the real-time coordinates of the actual positioning point.

9. The clinical anesthesial ultrasound image assisted positioning guidance method of claim 8, wherein, The dynamic spatial reference frame is calibrated according to the position difference index to generate an optimized motion path comprises: calculating the Euclidean distance between the real-time coordinates of the actual positioning point and the center reference point as the position difference index; adjusting the rotation and translation parameters of the dynamic spatial reference frame according to the position difference index; replanning the path of the guiding instrument from the current position to the calibrated reference point to generate an optimized motion path.

10. A clinical anesthesiology ultrasound image assisted positioning guidance system comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that, The processor, when executing the computer program, implements the steps of the clinical anesthesia ultrasound image assisted positioning and guiding method according to any one of claims 1 to 9.

Images